Last week Professor Bob Eccles and I published our book “Making Sustainability Financially Relevant: A Man/Machine Collaboration.” We think it’s a timely and much needed book. Over the past five years, “sustainability” and the acronym “ESG” have become politicized at both ends of the spectrum. On the left, critics argue these ideas have been captured by shareholder value and that the real work belongs at the level of whole systems. We respect that view, but companies don’t exist to save the world. They exist to solve problems for their customers and make money without making the world worse, inside the constraints of resources, laws, and regulation. On the right, sustainability has been framed as a progressive liberal agenda that interferes with value creation. Much of that critique is politically motivated, although we’ll concede that companies have too often overpromised on doing well while doing good.

We didn’t write this book to win that argument. We wrote it to change the subject.

Our claim is simple, almost boring. And the boringness is the point. A company can go on offense about sustainability; not with slogans, but with analysis. Understand the business model. Identify the material risks (we start with SASB, but other frameworks work too) disclosed in the 10-K or 20-F. Ask how well the company actually explains the link between those risks and value creation. Map each risk to specific line items on the income statement, balance sheet, and cash flow statement. Then estimate the financial impact. Done properly, the result isn’t a narrative or a score. It’s the kind of analysis companies can use to explain the link between good sustainability performance on material issues and financial performance.

Someone could have had this idea 20 years ago. What’s new is that AI has made it practical. A skilled analyst used to be able to trace a handful of material risks through one company’s disclosures and defend the numbers to a skeptical investment committee. Doing it across 10 companies was heroic. Across 100, impossible. AI breaks that constraint, not by replacing the analyst, but by expanding what one analyst can do. This book shows how, and it shows our work, including the prompts.

We chose ExxonMobil as our test case because it is about as hard a case as we could think of. Sustainability analysis there is actively contested, the numbers are enormous, the scrutiny is intense, and every assumption gets challenged. If the method survives ExxonMobil, it survives anywhere. ExxonMobil isn’t a convenient example. It’s a stress test.

The method survived. But it also broke and that turned out to be the most important findings in the book. When we pushed an expanded framework beyond SASB’s established standards, it produced a large, internally coherent, technically defensible estimate of transition-related demand destruction, running well into the billions. The number was wrong. Not because the mechanism was flawed, but because the policy world it assumed had just moved decisively in the opposite direction. We caught it. We traced the error to three specific design failures and built four targeted fixes. The corrected framework is now stronger than either the original or the broken version.

That episode is the book’s central argument in miniature. AI makes the analysis possible; human judgment makes it trustworthy. Better AI does not reduce the need for human judgment. It raises the bar for what the human has to do. The only thing standing between a credible-looking wrong number and a misled investor is domain expertise applied at exactly the right moment. We came to think of ourselves, throughout, as the human coherence anchors.

Because we were watching the machine the whole time, the book also doubles as an informal field report on AI itself. Where it succeeded, where it failed, and what the humans had to do. We ran the analysis through several systems, and we tell you by name which agent produced what. When two independently trained systems agreed on the central findings but diverged on the tail risks, the disagreement was itself the signal. It pointed straight at the assumptions that most deserved human scrutiny.

Part 1 is the case study—the method, built and tested. Part 2 is the operating manual--what to actually do, on Monday, if you hold one of the roles this work touches. We wrote for the board director, the CEO, the CFO, and the CSO; for the long-term investor, the activist investor, the ESG and impact investor, and the private-equity and private-credit investor; for the buy-side analyst and the sell-side analyst; and for the NGO seeking constructive engagement. The through-line is that the gap this method closes is usually organizational, not analytical. In most companies, the sustainability team owns the sustainability report, and the finance team owns the financial report, and no one owns the relationship integrating the two. Which is true integrated reporting. This book makes that gap hard to leave unaddressed and makes possible the original vision for integrated reporting.

A few words on what the book is not. It is not a claim that AI can replace financial judgment. It is not a verdict on what ExxonMobil is worth or what anyone’s allocation to fossil fuels should be. And it does not address negative externalities that don’t affect value creation which are obviously deeply important and must be addressed. Our honest view is that more of that burden belongs to public policy than to companies and investors, who can only do so much.

We self-published this book, fast, through our own imprint, because AI is moving quickly and we didn’t want to wait out a traditional book production cycle. We don’t think of the methodology as definitive, and we expect to revise it as the tools improve. What we do think is eternal is the underlying idea—the granular mapping of material sustainability issues to the drivers of financial performance and valuation. There is a price on the book, and on the related IP we’ll be releasing, because it reflects two decades of work and because, candidly, people tend to use what they’ve paid for. We’re academics. We’re not going to get rich from a book. This isn’t our goal. Our goal is to help companies make more rigorous the link between sustainability performance and financial performance. This will improve internal capital allocation and decision making. It will also improve communications to shareholders and other stakeholders.

Finally, we want the arguments here challenged. Both of us are at the point in our careers where critical feedback is the most useful thing a reader can give us. We have no career anxiety about hearing it. So, tell us where we’re wrong.

Each Monday from here, we’ll take up one theme from the book. We start next week with the gap at the heart of it: companies disclose their sustainability risks to satisfy securities law, then price them at zero in the financial models that actually drive decisions. We hope you’ll come argue with us.

In both the faculty lunch and the midnight chat window, the idea emerges from the exchange itself — not from either mind alone.

Consider two scenes.

Scene one: Two accounting professors at lunch. One mentions an empirical puzzle she noticed in earnings announcements. The other asks a clarifying question. An idea forms. By dessert, there is the rough shape of a paper. Neither professor can say, afterward, who had the idea. The conversation produced it.

Scene two: The same professor at 11 pm. She opens a chat window and types a question. AI responds with a clarification and a reframing. She pushes back. AI concedes one point and extends another. By midnight, the rough shape of a paper exists. Who had the idea?

The honest answer, in both cases, is that we do not know and it doesn’t really matter. What matters is whether the idea is good. Whether it advances our understanding of how markets work, how firms behave, how capital gets allocated. Whether it is true.

That question — not who wrote the sentence, but whether the knowledge is any good — is the one academic institutions have stopped asking. This piece summarizes our argument in the paper “The Artisanal Paper Is Dead: Knowledge Production in the Age of the Hybrid Intelligent Team.” for why that is a problem and what should be done about it. The full theoretical foundations are developed in a companion paper, “Production of Knowledge in Human-Machine Collaborations: The Hybrid Intelligent Team as Epistemological Form“ which provides the epistemological scaffolding for the first paper.

The 1926 Workflow

The artisanal model: a lone scholar handcrafting a single empirical paper over three to five years.

The way academic knowledge gets produced in accounting, finance, and management hasn’t changed much in a century. A lone scholar, or a small team, spends three to five years handcrafting a single empirical paper. The process is slow, insular, and obsessive about identifying who deserves credit. The resulting output is often unreadable to the practitioners, policymakers, and regulators who most need the insights and arrives years after the question it addresses has moved on.

We call this the artisanal model of knowledge production. And we argue, with increasing urgency, that it is being rendered obsolete, not by a new tool, but by a fundamental shift in the organizational form of research itself.

The replacement is what we call the Hybrid Intelligent Team, or HIT: a structured configuration of human and AI agents whose outputs emerge from iterative loops rather than isolated minds. Not a person with a ChatGPT subscription, but a genuinely differentiated system; one agent generates, another challenges, a third verifies independently without knowing what the others concluded. A human orchestrator holds the whole together, directing the process, evaluating the outputs, and taking responsibility for the claims. The unit of production is not any individual agent. It is the iterative loop connecting them. And if academic institutions do not change their rules — their tenure systems, disclosure policies, journal style requirements — they are simply using 2026 technology to prop up a 1926 workflow.

AI Is Not Your Secretary

The Hybrid Intelligent Team — differentiated agents generate, challenge, and verify, while a human orchestrator holds the loop together and carries responsibility for the claims.

The temptation, when thinking about AI in research, is to treat it as a very powerful assistant: faster literature searches, cleaner code, better grammar. Resist that framing.

The deepest barrier to good scholarship has never been logistical. It has been ideational — not knowing what question is worth asking, not seeing the anomaly in the data that actually matters, not having the conceptual vocabulary to frame what you have found. No AI fixes that. What AI does is collapse everything that comes after the idea — the literature review that used to take three months, the coding task that ate six weeks, the second-best paper that got written because the truly ambitious one would have required expertise the researcher did not have.

These were real frictions. Removing them is genuinely significant. But they were never the source of intellectual value. They were the packaging operation. AI is the world’s most powerful packaging operation, and it now runs at near-zero cost. What this means is not that scholarship gets easier. It means that ideation — the capacity to identify a question that matters, frame it precisely, and recognize when the answer is actually an answer — becomes the only thing that separates good research from competent noise.

We call the relevant concept loop-level ideation. When the loop is functioning, knowledge is not produced by the human or the machine; it is produced by the iterative cycle between them. But this is easy to mistake for ordinary iterative drafting. The test is generative consequence; did the output change what you asked next? Did it produce a distinction you could not have made before the loop ran? When the answer is “Yes” you are not dealing with AI as a packaging operation. When the answer is “No,” when the human had the idea and the machine dressed it up, that is AI-assisted drafting. Both are real. Only one justifies rethinking how we attribute, evaluate, and reward intellectual contribution.

One important asymmetry remains. When two humans collaborate, both bear professional stakes in the output. AI has no stake, no career, no consequences. It cannot be sanctioned, retracted, or held to account. The loop distributes ideation; it does not distribute responsibility. The human in the loop carries all of it.

Six Journals. Four Dissonances. Three Proposals.

We examined the AI policies of the three leading accounting journals (The Accounting Review, Journal of Accounting Research, and Journal of Accounting and Economics) and the three leading finance journals (Journal of Finance, Journal of Financial Economics, and Review of Financial Studies) and found every single one built on flawed premises.

Here’s the most revealing fact. A study covering more than 5.2 million papers published since 2023 found that despite 70 percent of journals adopting disclosure-focused AI policies, only 76 papers, approximately 0.1 percent, explicitly disclosed AI use. The current disclosure regime is not producing transparency. It is producing the appearance of a policy while researchers quietly ignore it.

We identify four specific dissonances. The disclosure direction is backwards. Every journal requires authors to disclose if they used AI. We propose the exact opposite. The default assumption should be that AI was used, and the only meaningful disclosure is when an author has not. The permitted scope has been too narrow in most journals, though Elsevier’s September 2025 updates to JAE and JFE represent genuine progress. The reviewer prohibition at JAE and JFE, which explicitly ban referees from using AI on submitted manuscripts, makes the system structurally asymmetric. Production is AI-assisted at scale; review is constrained to human-only, and the gap will only grow. One of us is an editor receiving 350 submissions per year, many of them AI-enhanced, expected to evaluate each one without recourse to the same capabilities that made those submissions cheap to produce. And the Journal of Finance alone has pointed in the right direction, permitting AI-assisted referee reports and uniquely requiring full prompt documentation for AI-generated datasets — the only replicability provision in any of the six policies.

Three reforms follow.

· First: Unrestricted AI use. The line between AI as companion and AI as replacement, JAR’s current framing, is one no serious researcher recognizes in practice. Its only effect is to ensure that AI use goes undisclosed.

· Second: Flip the disclosure default. Require disclosure when AI was not used. The flipped norm removes the stigma currently attached to AI disclosure and replaces it with a signal cost for non-disclosure.

· Third: Allow AI-assisted peer review. Columbia colleague Oded Netzer suggested including AI as a fourth referee on every paper, with no obligation to respond to it. It would socialize AI as a legitimate participant in the evaluation process and, over time, accumulate data on how AI assessments correlate with eventual outcomes. The confidentiality concern that drives the current prohibition is real but solvable. The right answer is institutional AI infrastructure, not a blanket ban.

Who Would Not Get Tenure Under the New Framework?

If tenure criteria shift from publication count to usefulness, some profiles that have historically thrived would struggle. It is worth being specific.

The Artisanal Technician has mastered a particular methodology (e.g., regression discontinuity, natural experiments, and textual analysis) and applies it, carefully and competently, to question after question. The work is technically rigorous and publishable. It is also, increasingly, AI-replaceable in its core execution. Technical facility without ideational originality is, in the HIT era, execution without intellectual alpha.

The Journal Game Player has learned the stylistic and rhetorical conventions of target journals with sufficient precision to produce papers that clear editorial filters without necessarily advancing understanding. AI has now exposed the game completely. We demonstrate this in an appendix to the artisanal paper. We took a practical, readable piece of research on the Public Company Accounting Oversight Board (PCAOB) and had AI translate it into impenetrable theoretical jargon in approximately 30 seconds. If a 30-second translation produces the same stylistic signal as three years of field immersion, the signal has collapsed.

The Dataset Moat Builder has built tenure cases partly on privileged access to proprietary data. AI substantially erodes the positional component of that advantage. Those whose contribution was primarily access rather than insight face a harder evaluation under a system that asks what the access revealed.

The Political Scholar builds a career by navigating the preferences and tribal loyalties of editors with sufficient precision to produce papers that satisfy them, not by being right about the world but by reflecting back what the gatekeepers already believe. AI dismantles this model. AI-assisted papers search the literature too comprehensively for strategic omission to be sustained, and AI dramatically lowers the cost of testing whether the priors that have governed editorial selection are actually supported by evidence.

None of these profiles is fraudulent. Each represented rational adaptation to the incentive structure the artisanal model created. What we are arguing is that the incentive structure is being made obsolete.

The Recognizer Problem. This Is the One That Matters.

AI-assisted production scales faster than the supply of skilled Recognizers, shifting the binding constraint from producing knowledge to evaluating it.

AI accelerates production. It does not replace judgment. The central constraint of the HIT era is not producing knowledge. It is recognizing it.

A well-run Hybrid Intelligent Team can produce warrantable outputs — claims that have met the process conditions for genuine knowledge, that are traceable, that can survive adversarial scrutiny — at a scale and speed the artisanal model never could. But warrantable is not the same as warranted. A claim becomes warranted when the community of inquiry takes it up, evaluates it, and confirms that it actually advances understanding. Warrantable is what the production process can achieve. Warranted is what the community confers. The loop can produce the first. Only the community can produce the second.

The problem is the Recognizer. The loop can only surface knowledge that the human at its center is capable of recognizing as knowledge. If the orchestrator cannot tell the difference between a genuine reframing — a finding that changes what questions are askable — and a sophisticated recombination that merely rearranges what is already known in fluent prose, the loop’s most distinctive outputs get passed through uncritically or discarded for reasons the orchestrator cannot articulate. The loop is bounded by the person at its center.

This is not abstract. The machines can generate text that is fluent, accurate on retrievable facts, and structurally coherent. What they cannot do is produce text that is right in the way a domain expert’s judgment of rightness differs from a checklist of correctness criteria. The human in the loop is not merely approving or rejecting outputs. She is reading drafts, coordinating across agents, proposing revisions, pushing back on claims that are locally coherent but wrong in ways only domain expertise can detect, and taking responsibility for conclusions she cannot offload to the systems that generated them. The gap between accurate and right is where tacit knowledge lives, and no AI yet bridges it.

Now zoom out to the field level. AI-assisted production is scaling faster than the supply of skilled Recognizers — scholars with the domain knowledge, the tacit understanding, and the accountability to evaluate whether a knowledge claim is genuinely warrantable--rather than merely fluent. We will not run out of papers. We will run out of credible evaluation.

The Ph.D. Problem Is Deeper Than It Looks

The concern is not just that PhD students will outsource their work to AI. It is that they will outsource the cognitive processes through which understanding is built.

When a doctoral student spends two years learning econometrics, she is not merely learning to run regressions. She is learning how data is generated, what the assumptions embedded in an estimator actually require, and what distinguishes a robust finding from a fragile one. That understanding cannot be acquired by prompting AI to run the analysis. It is acquired by doing it wrong and learning why.

The same applies to writing. Writers do not first think and then write; they think by writing. A PhD student who delegates first-draft generation to AI before internalizing this discipline is not saving time. They are outsourcing the cognitive process through which understanding is actually built. The Recognizer function requires precisely the evaluative capacity that process produces.

This is what we call the bootstrapping problem, You cannot be a credible Recognizer without first having been, at some point, the executor. If programs delegate execution to AI before students have built foundational understanding, they risk producing Ph.D. graduates who can operate AI systems they cannot critically evaluate. The goal of methods training must be reframed, not to produce students who can run a clean regression, but to produce students who can evaluate whether an AI-generated analysis has made the right assumptions and interpreted the right coefficient.

The Teaching Load Question Nobody Is Asking

If a Hybrid Intelligent Team can produce a high-quality paper in three months rather than three years, what justifies the reduced teaching loads that academic research positions currently carry? The argument for lighter loads rests entirely on the premise that research takes a long time and the time freed from teaching produces knowledge that exceeds the value of the teaching foregone. If AI collapses the production timeline dramatically, that argument needs to be remade honestly.

There are three alternatives that are not mutually exclusive: (1) more and better teaching, transformed by AI to be more personalized and interactive; (2) more and better mentorship of junior researchers in exactly the skills AI cannot transmit; and (3) more ambitious research — questions too large or too risky to pursue under the old production function but tractable when the cost of iteration is lower. What we are not willing to accept is using 2026 technology to produce papers at the old artisanal pace while pretending the research subsidy is still justified.

What This Is Really About

Every reform we propose is justified by one criterion. Does it produce better knowledge? Not more papers, not faster review cycles, not cleaner disclosure norms. Better knowledge.

The evidence from other fields is instructive. AlphaFold solved the protein folding problem with accuracy equivalent to experimental methods — a step-function advance, not incremental progress. The pattern across drug discovery, astronomy, and medicine is consistent. AI’s contribution is about expanding the tractable question space, allowing researchers to ask questions that were previously unanswerable because the data scale was too large or the interdisciplinary synthesis too costly.

The accounting and finance equivalent might be figuring out which risk disclosures in SEC filings investors actually use and which are compliance theater nobody reads. SEC Chair Atkins has stated publicly that risk factors in the 10-K have become a repository for too much. The standard academic approaches have not been able to answer that question credibly at scale. A well-run HIT, working with causal rigor across millions of filings, might.

The conclusion is both simple and demanding. Flip the disclosure default. Allow AI-assisted review. Evaluate tenure candidates on epistemic judgment and live intellectual performance — on whether they can defend their ideas in a seminar room under adversarial questioning — rather than on paper counts calibrated to a production function that no longer exists. Develop Recognizer capacity in doctoral programs before the gap between production and evaluation becomes unbridgeable.

The scholars, institutions, and journals that thrive will be those that stop defending a workflow designed for a world that no longer exists. Stop obsessing over the byline. Start asking whether the knowledge is true and useful.

A Note on How This Piece Was Made

The production history of this piece is itself an instance of the argument, so it is worth documenting briefly.

This Substack summary draws on two papers. The first is “The End of the Artisanal Paper: Knowledge Production in the Age of the Hybrid Intelligent Team“ by Robert G. Eccles and Shivaram Rajgopal, which is the empirical and institutional core of the argument — the six-journal analysis, the reform proposals, the tenure archetypes, the PhD and teaching load implications. The second is “Production of Knowledge in Human-Machine Collaborations: The Hybrid Intelligent Team as Epistemological Form“ by Robert G. Eccles a companion paper that provides the theoretical scaffolding — the warranted/warrantable distinction, the Recognizer Problem fully developed, the four structural conditions of the HIT as an epistemological form. Few people will read the companion paper. The Artisanal paper will reach a much wider audience. This piece is designed to extend that reach further still, which is exactly the kind of layered dissemination strategy a HIT makes tractable.

Both papers were themselves produced using the organizational form they describe. The Artisanal paper was produced by a HIT consisting of both authors, two Claude instances with no memory continuity between them, and a Gemini Pro adversarial review. The absence of memory continuity between the Claude instances was not a design flaw; it was the condition the theory describes. The human orchestrators were the only continuous thread across the entire production process, which is precisely what the coherence anchoring function requires. The adversarial review identified five logical vulnerabilities in the draft, three of which were judged genuine and consequential. Those three produced four substantive additions to the paper that were not in the original draft — additions that changed the argument rather than polishing it. The loop produced them. No single node did.

This Substack summary was produced through a parallel process. Shiva drafted one version with his Claude; Bob drafted another with his. The two versions were then combined, using Shiva’s architecture as the base and feeding in the stronger theoretical treatments from Bob’s version. Bob is listed as a co-author. That listing is institutionally correct — Bob is accountable for everything here — and epistemically incomplete, which is exactly the point the paper makes. The byline names the person responsible. It does not describe where the knowledge came from. Those are two different things and conflating them is the error the artisanal model has been making for decades.

I am not raising these questions to argue that SpaceX has done anything wrong. Every accounting position described in this piece is permissible under GAAP. I am raising them because the S-1, like all complex filings from complex companies, reflects choices from within the range of what is allowed — and because understanding where those choices sit within that range is the starting point for understanding what SpaceX actually is as an investment.

The company operates across three segments with fundamentally different economics: Connectivity (Starlink), which is genuinely profitable; Space (rockets and launch services), which loses money but is strategically central; and AI (xAI, Grok, and X), which is burning cash at a rate that would alarm most investors if it were a standalone company. The S-1’s job is to present all three together in a way that makes the whole seem coherent. There are places where the presentation raises questions that the filing itself does not fully answer, and those are the questions worth examining.

The S-1 raises questions that the filing itself does not fully answer. Those are the questions investors need to be asking.

Here are the numbers that matter most:

Before we begin: the bull case, taken seriously

There is a version of the SpaceX story that makes everything in this piece look beside the point. I want to state it as well as a believer would before proceeding, because the accounting questions that follow deserve to be evaluated against the strongest possible version of the opposing argument.

Starlink is one of the most remarkable businesses built in the last twenty years. It generated $11.387 billion in revenue in 2025, grew that revenue by 49.8 percent year-over-year, and produced $7.168 billion of adjusted EBITDA at margins above 60 percent. It did all of this by connecting 10.3 million subscribers across 164 countries to a low-latency global broadband network that did not exist six years ago. There is no comparable infrastructure business in the world growing at this rate at this margin.

SpaceX’s launch record is not a financial projection. It is an operational fact. Since 2023, SpaceX has launched more than 80 percent of all mass to orbit globally, with a mission success rate above 99 percent on Falcon rockets. No competitor is within a decade of matching that combination of cost, cadence, and reliability.

The AI bet is enormous and unproven. But the same was said of Amazon’s cloud infrastructure investment in 2008. The pattern — a profitable, cash-generating core business funding a long-duration, capital-intensive investment in an emerging market — is exactly how some of the most valuable companies ever built got there. Starlink is playing the role of the retail business. The AI segment is playing the role of AWS in 2008.

If that analogy holds, investors who decline to participate because the accounting is complicated will have made the same mistake made by those who sold Amazon in 2001. That is a real risk, and any honest analysis of SpaceX has to acknowledge it.

Precision in accounting serves believers more than skeptics. It is the investor who shares the vision — and who is therefore paying the price that embeds it — who most needs the numbers to be transparent.

Here is my response to that argument.

A skeptic who thinks the AI vision is wrong will decline to invest regardless of disclosure quality. Nothing in this piece is aimed at skeptics. It is aimed at the investor who believes in SpaceX — who thinks Starlink will reach 50 million subscribers, that Starship will transform the economics of space, that AI compute in orbit is a real business in 2030 — and who is therefore paying a price that embeds all of that conviction. That investor is the one most exposed if the accounting choices obscure cost.

If the common control treatment of X overstates the combined balance sheet by tens of billions, it is the investor who paid $1.75 trillion who absorbs the correction. If GPU useful lives are too long and depreciation is understated, it is the long-term holder who funds the eventual catch-up. If the $15 billion Anthropic contract is cancelable on 90 days’ notice and the revenue recognition policy is aggressive, it is the believer who is buying a revenue multiple that may not survive the next contract renegotiation. Accounting presentations that reduce transparency on these points do not shift the underlying risk — they shift who understands it.

The Amazon comparison is exactly right — but not in the way it is usually deployed. Amazon’s losses were not obscured. Jeff Bezos wrote annual shareholder letters that explained precisely what was being sacrificed and precisely what was being built. He gave investors the information they needed to decide whether to share his conviction. He treated investors as partners in a long-duration bet, not as passengers on a vehicle they did not fully understand. That transparency — that insistence on explaining the economics honestly, even when they were ugly — is what made Amazon’s losses acceptable to long-term investors. It is also what made the subsequent returns so large: investors who stayed did so with eyes open.

My take: What I am asking for in this piece is not pessimism about SpaceX. I am asking for Bezos-level transparency about the numbers. Show me what acquisition accounting would have produced for X Holdings. Classify the capex by asset class and useful life so I can assess the depreciation trajectory. Explain which Adjusted EBITDA adjustments will recur in five years. Disclose the Anthropic termination provisions so I can model the revenue risk. Tell me what the related-party transactions look like against independent benchmarks. That kind of disclosure has historically been what separates companies that earn durable investor trust from those that have to re-earn it after the first difficult year.

The governance argument deserves the same treatment. Objecting to the CEO-consent removal provision is not an assertion that Elon Musk will misuse his position. It is a statement about the design of the system. Good governance assumes that all human beings, including the most talented and visionary ones, occasionally act in their own interests. The constraint is not personal; it is structural. Investors in Tesla who raised governance concerns in 2018 were not predicting Musk’s failures — they were identifying a system where the costs of any failure would fall asymmetrically on shareholders rather than management. That concern was vindicated, not refuted, by subsequent events. The question is not whether Musk is trustworthy. It is whether the structure is sound if he is ever wrong.

None of the seven sections that follow argue that SpaceX will fail. They argue that investors who believe it will succeed are buying it at a price that embeds that belief — and that the accounting and disclosure choices in the S-1 make it harder, not easier, to determine whether that price is fair. That is a problem for believers, not for skeptics. Skeptics will not buy it. Believers will. They deserve better numbers.

1. The accounting choice that most benefits management

The most consequential single accounting decision in the S-1 is one that most press coverage has not examined in detail.

SpaceX used common control accounting under ASC 805-50 to retroactively recast its entire financial history to include xAI (acquired February 2026) and X Holdings (acquired by xAI in March 2025). The technical justification is that Elon Musk controls all three entities, making these reorganizations among commonly controlled parties rather than arm’s-length acquisitions. Under this framework, X’s assets flow into SpaceX’s balance sheet at their historical carrying values rather than at fair market value. There is no acquisition-method goodwill recognition and no fair value step-up. The reorganization itself does not, by itself, create acquisition-method goodwill or a fair-value remeasurement event; however, carried-over assets remain subject to applicable impairment guidance.

To understand the magnitude of this choice, consider what would have happened under acquisition accounting. Musk originally paid approximately $44 billion for Twitter in 2022. By most credible independent estimates, the platform’s value subsequently declined substantially. Under acquisition accounting, identifiable assets and liabilities would generally have been measured at acquisition-date fair value, with any excess consideration reflected through goodwill or, depending on the specific facts, other acquisition-accounting effects. The restated financials would have looked materially different.

Instead, the common control framework allows SpaceX to present three years of retrospective financials that include X and xAI as if they had always been part of the same company, at historical book values, without requiring the recognition of value change that acquisition accounting would have demanded.

My take: This treatment is technically permissible under GAAP. What matters to an investor is whether the resulting financial statements accurately convey the economic substance of what occurred. A company that absorbed a platform acquired for $44 billion — at a cost basis reflecting peak social media valuations — at carrying value rather than current fair value is presenting a significantly different picture than the underlying economic reality. Investors should seek disclosure of what acquisition accounting would have produced, even if that alternative is not required.

The question no one is asking: If SpaceX had instead acquired X from an unrelated party today and applied acquisition accounting, what would the fair-value measurement have produced? The S-1 does not appear to provide a full acquisition-accounting sensitivity showing what such measurement would have produced.

2. Spending more than you earn

SpaceX spent $20.7 billion in capital expenditures in 2025. It earned $18.674 billion in revenue. That ratio — 111 percent — is striking by any standard. It means the company deployed more capital than it received from customers in the same year.

This is not automatically a problem. Companies investing aggressively in genuinely productive assets should spend more than they currently earn; that is what capital allocation is for. The question is whether the assets being created justify the spending, and whether the accounting treatment of that spending is appropriate.

Of the $20.7 billion, $12.727 billion went to AI infrastructure — GPU clusters, data centers, and associated systems. The AI segment that all this capital is building generated $3.201 billion of revenue in 2025 and a negative $6.355 billion of adjusted EBITDA. In Q1 2026 alone, AI capex was $7.723 billion while AI revenue was $818 million. The ratio of AI capex to AI revenue in that single quarter is roughly nine times.

The Starship program adds further complexity. SpaceX funded $3.004 billion of Starship research and development through the Space segment in 2025 and another $930 million in Q1 2026 alone. The accounting question here is fundamental: at what point does a developmental launch vehicle cross from an R&D expense — which must be recognized immediately — to a capitalizable asset? If any portion of Starship spending is being capitalized when it should be expensed, the reported GAAP loss is understated. The S-1 does not give investors enough information to make this assessment independently.

GPU clusters present their own challenge. The useful lives assigned to AI computing hardware determine how quickly costs flow through the income statement. The AI industry has not yet developed consensus on the appropriate useful life for a cluster of GPUs that may be commercially obsolete within three to four years as faster models arrive. The difference between a three-year and a five-year useful life on $12.7 billion of AI assets is a depreciation swing of roughly $1.7 billion per year.

My take: A 10 percent change in the capitalization assumptions applied to $20.7 billion of capex implies a $2.07 billion swing in reported income. That is not a rounding error. It is larger than most companies’ entire annual revenues. Investors need far more granularity on the asset class breakdown, useful-life assumptions, and the policy governing when development spending becomes capitalizable.

The question no one is asking: If GPU useful lives were shortened by two years across the AI segment’s existing asset base, how much additional depreciation would flow through the income statement this year?

3. The $11.5 billion gap

SpaceX reported a 2025 GAAP net loss of $4.937 billion and 2025 Adjusted EBITDA of $6.584 billion. The difference between those two measures is roughly $11.5 billion — though it is worth noting that GAAP net loss and Adjusted EBITDA are not directly comparable: the former captures all economic costs including interest expense and taxes, while the latter adds back depreciation, amortization, stock-based compensation, financing costs, and other items. The gap in Q1 2026, between a GAAP net loss of $4.276 billion and Adjusted EBITDA of $1.127 billion, was $5.4 billion in a single quarter. Both figures appear in the filing, but they measure fundamentally different things.

Non-GAAP measures are legal. They are also, when deployed at this scale, a form of narrative management. Adjusted EBITDA adds back depreciation and amortization, stock-based compensation, and other items that companies characterize as nonrecurring or non-cash. The implicit argument is that these exclusions make the underlying cash economics of the business more visible.

The argument has some merit in certain contexts. Depreciation is a backward-looking allocation of past capital spending; it does not directly reflect the current cash generation of a business. For a highly capital-intensive company like SpaceX, the D in EBITDA is genuinely large and genuinely distorts year-to-year earnings comparisons.

But precision matters here. Capital expenditure is not itself excluded from EBITDA. What is excluded is the depreciation of the assets purchased with that capital. The distinction is important, but the economic consequence for a company spending $20.7 billion on assets per year is similar: a very large and growing stream of depreciation charges that EBITDA ignores, charged against a GAAP income statement that those charges make look far worse than the non-GAAP presentation suggests.

SpaceX’s adjusted EBITDA in 2025 also reflected Starlink’s $7.168 billion contribution, which masked the AI segment’s $6.355 billion EBITDA loss. Strip out Starlink and the business is deeply loss-making even on a non-GAAP basis.

My take: Many large public companies have increasingly relied on adjusted profitability measures that are more favorable than GAAP earnings. That trend does not make such measures wrong; it makes them worth scrutinizing. The question any investor should ask is whether the adjustments being made are genuinely non-recurring and non-economic, or whether they are recurring costs being systematically excluded to present a more favorable picture.

The question no one is asking: Provide the full Adjusted EBITDA bridge and classify each adjustment: recurring or not, cash or non-cash, discretionary or structural. Which adjustments would a short seller dispute?

4. The AI segment: losses described as investments

The AI segment is the most interesting accounting challenge in the S-1, and the most consequential for valuation.

In 2025, the AI segment generated $3.201 billion in revenue, spent $12.727 billion in capital expenditures, and lost $6.355 billion on an adjusted EBITDA basis. In Q1 2026, it generated $818 million in revenue and lost $2.469 billion in operating income. The segment’s losses are accelerating faster than its revenue.

The S-1 frames this consistently as investment rather than operational shortfall. The language is that the company is “prioritizing growth and investment to capture significant opportunities.” This framing is not wrong — it accurately describes management’s intent — but it eludes the question of whether those investments are on track to generate returns that justify the capital deployed.

The largest single AI revenue contract disclosed is with Anthropic: $1.25 billion per month through May 2029 for compute capacity at the Colossus data center in Memphis, subject to termination rights reportedly exercisable on 90 days’ notice by either party. Annualized, that is $15 billion — equal to roughly 80 percent of SpaceX’s entire 2025 revenue. This is an enormous single-counterparty concentration. The revenue recognition question is whether SpaceX can recognize this as revenue as compute capacity is made available, or whether recognition depends on actual usage, service-level achievement, or other conditions. The termination caveat makes the collectability and durability of this revenue stream one of the most important undisclosed facts in the filing.

The S-1 also discloses that SpaceX holds an option to acquire the coding startup Cursor for $60 billion later this year, alongside a $10 billion joint development commitment. The accounting for an unexercised option at this scale — whether any expense is recognized, how fair value is determined, what conditions govern exercise — should be a central disclosure point. An option exercisable at $60 billion in Class A stock is a potential dilution event of considerable magnitude for public investors.

My take: The AI segment is being valued by markets as a high-growth technology business. The financial statements describe something closer to a pre-revenue infrastructure company that is burning cash at a rate that would require Starlink profits to sustain indefinitely. Both descriptions may be simultaneously true; the resolution depends entirely on assumptions about future demand for AI compute that no one can verify today.

The question no one is asking: If Anthropic were to exercise its reported 90-day termination right on its $1.25B/month compute contract, what would the AI segment’s revenue and liquidity position look like in the immediately following quarters?

5. The related-party ecosystem

SpaceX’s S-1 discloses an intricate web of transactions between entities controlled by Elon Musk. In 2025, the combined SpaceX entity paid approximately $650 million to Tesla for goods and services per the S-1’s related-party disclosures: SpaceX’s own operations purchased roughly $144 million in goods and services from Tesla (including approximately $131 million in Cybertrucks), while xAI purchased approximately $506 million in Megapack battery storage systems to power data centers. Tesla’s own filing confirms it holds nearly 19 million shares of SpaceX Class A stock.

The S-1 separately discloses that xAI paid lease obligations to Valor entities — a Musk-affiliated infrastructure fund — totaling approximately $885 million in 2025 and another $857 million in just the first two months of 2026 alone, under arrangements partly treated as failed sale-leaseback transactions that required recording billions of dollars of associated obligations as debt. This is a related-party relationship far larger than the Tesla vehicle purchases that have attracted most press attention.

This is a network of companies — SpaceX, xAI, X, Tesla, The Boring Company, Musk Industries LLC — where a single individual sits at the center of multiple material transactions, often setting terms on both sides of each deal. The disclosure framework requires investors to understand the nature, amount, and terms of related-party transactions, including whether terms are comparable to those available from unaffiliated parties. That is the standard against which each of these transactions should be assessed.

In most of these cases, the answer is probably yes — buying Megapacks for data centers is commercially reasonable regardless of who owns the supplier. But “probably yes” is not the same as “demonstrated yes,” and the concentration of these transactions across a single founder-controlled ecosystem creates a disclosure environment in which market terms are difficult to verify independently.

My take: The individual transactions may each be commercially reasonable. What deserves scrutiny is the governance framework around approving them. When the same individual sits on both sides of a material transaction — as counterparty and as controlling shareholder — the structural question is not about intent. It is about the process through which the company can demonstrate to outside investors that terms were in fact comparable to what an unaffiliated party would have accepted. That demonstration requires disclosure that goes beyond what the S-1 currently provides.

The question no one is asking: For every transaction above $10 million with a Musk-affiliated entity, what independent evidence exists that the pricing matched what a genuinely unaffiliated party would have charged or paid?

6. A $28.5 trillion story built on $3.2 billion of AI revenue

SpaceX’s S-1 identifies a total addressable market of $28.5 trillion. Of that, $26.5 trillion is attributed to AI — covering AI infrastructure, consumer subscriptions, digital advertising, and enterprise applications. The remaining roughly $2 trillion covers space and connectivity.

This is a striking construction. The AI TAM of $26.5 trillion is 94 percent of the total opportunity being pitched to investors. The AI segment that is supposed to address that market generated $3.201 billion in revenue in 2025 — a market penetration rate, if we take the TAM at face value, of approximately 0.012 percent. The segment lost $6.355 billion on an adjusted EBITDA basis in the same year.

TAM figures in S-1 filings are legally classified as forward-looking statements and subjected to extensive cautionary language. But they are not inconsequential. They anchor investor expectations, support valuation multiples, and shape the narrative that determines how financial results are interpreted. A company that reports a $6.355 billion EBITDA loss in a segment described as addressing a $26.5 trillion opportunity is implicitly asking investors to treat that loss as the cost of claiming a very large future prize.

The question is not whether the prize exists — AI infrastructure is genuinely significant. The question is whether $26.5 trillion is a rigorous estimate of the opportunity, whether SpaceX is positioned to capture a meaningful share of it, and whether the current rate of capital spending is a rational path to that outcome. The S-1 makes vigorous arguments for all three propositions. It does not provide the evidence that would allow an investor to independently assess them.

My take: There is an established body of research showing that companies with larger reported TAMs tend to raise more capital and trade at higher multiples, independent of whether those TAMs are ever captured. The $28.5 trillion figure in SpaceX’s S-1 is doing real work in the valuation of this offering. Investors should be clear-eyed about the gap between that figure and the $3.2 billion of AI revenue that actually exists today.

7. When the audit committee’s escalation ladder leads to the same person

SpaceX is a controlled company. Musk holds 85.1 percent of voting power through dual-class shares. Class B shares carry 10 votes each. As the S-1 structure makes clear, the only person who can remove Musk as CEO or Chairman is Musk himself, since removal requires a vote of Class B holders whose majority he controls.

As a controlled company, SpaceX is exempt from Nasdaq’s requirements for independent compensation and nominating committees. It reincorporated in Texas in 2024, benefiting from more founder-friendly corporate law. Shareholder disputes are subject to mandatory arbitration, eliminating class-action lawsuits as an accountability mechanism. Shareholders must own at least $1 million in stock, or 3 percent of the company, to force a proposal to a vote.

The governance structure has already drawn formal challenge. CalPERS, the New York State Common Retirement Fund, and the New York City Comptroller’s Office — collectively representing more than $1 trillion in retirement assets and three of the country’s most prominent public pension investors — sent a public letter to SpaceX executives calling the proposed governance structure “the most management-favorable governance structure ever brought to the US public markets at this scale.” They demanded, among other things, the elimination of the CEO’s consent requirement for his own removal.

The audit committee — required to be independent under SEC rules regardless of controlled-company status — is the primary institutional check on financial reporting in this structure. But the effectiveness of that check depends on the committee’s ability to escalate. Escalation paths in a controlled company ultimately lead to the same controlling shareholder. The committee can document its position. It can engage the external auditor. It can, in extremis, resign. What it cannot do is override.

This is not a theoretical concern. The specific accounting judgments at issue in this S-1 — common control treatment for the xAI merger, AI capex capitalization, non-GAAP presentation, related-party pricing — all involve genuine ranges of permissible choice. In each case, the S-1 reflects a position toward the more favorable end of that range. That is entirely normal. It is also the reason independent oversight exists.

Accounting standards routinely offer a range of permissible choices, and any rational management team advised by competent auditors and counsel will select positions within that range that reflect well on the company. That is not a criticism; it is an accurate description of how financial reporting works in every industry. The question is whether, in a structure where the audit committee’s escalation options are more constrained than at most public companies, there are sufficient external checks to ensure that choices at the favorable end of the permissible range are also the most defensible ones available. That is a question about institutional design, and it deserves a structural answer.

My take: I have written before that what preparers call immaterial, resourceful investors call raw material. The same is true of accounting choices within a permissible range. The choices that matter most are often the ones that receive the least disclosure: the alternatives considered, the analysis performed, the auditor’s initial reaction, and the basis on which one position was selected over another. SpaceX’s first proxy statement and 10-K will reveal how much of that process is made visible to the investing public.

The choices that matter most are often the ones that receive the least disclosure: the alternatives considered, the analysis performed, the auditor’s initial reaction.

What to look for when the 10-K arrives

SpaceX is a genuine technological achievement and a legitimate business of extraordinary scale. Starlink alone — $11.387 billion in revenue and $7.168 billion of adjusted EBITDA in 2025 — would be a large and profitable standalone company. The concerns raised in this piece are not about whether SpaceX is real. It is. In 2024, before the xAI merger, SpaceX reported net income of approximately $791 million. The merger turned a profitable company into a consolidated loss-maker. Understanding which part of the company is doing what is the analytical task the S-1 makes harder than it needs to be.

When SpaceX’s first annual report arrives, the following disclosures will matter most:

• The accounting memoranda supporting the common control treatment of xAI and X Holdings, and the alternatives that were considered — including what acquisition accounting would have produced.

• The asset-class breakdown of $20.7 billion in capex, with useful-life assumptions for each major category and the policy governing when development spending becomes capitalizable.

• A complete, classified reconciliation of the gap between GAAP net loss and Adjusted EBITDA, identifying which adjustments recur and which genuinely do not.

• The revenue recognition policy for the Anthropic compute contract and its single-counterparty concentration risk.

• A comprehensive related-party matrix with pricing benchmarks for every material transaction involving Musk-controlled entities, including the Valor AI infrastructure lease arrangements.

• The audit committee’s report in the first proxy, which will give investors their first real signal of how seriously the company takes its public-company oversight responsibilities and whether the questions raised in this piece have been engaged substantively.

The investors who will understand SpaceX best will not be the ones who accept the $28.5 trillion TAM at face value. They will be the ones who work through the accounting choices quietly embedded in the financial statements and form their own view of the gap between the company’s narrative and its economics.

That gap is real. It is also, potentially, justified. The resolution depends on assumptions about Starship, AI demand, orbital data centers, and Mars that no one — including SpaceX’s management — can verify today. Investors who understand that are equipped to make a genuine judgment. Investors who do not will be surprised by the first year’s GAAP results.

Earlier today I published a shorter version of this analysis in Forbes. If you found your way here from that piece, welcome to Googly — this is what I had to cut to fit a general-audience publication.

If you found your way here from LinkedIn, thank you for the response to that post. Several people asked for the underlying arithmetic, the S-1 page citations, and the Note 7 walkthrough. This is all of that.

And if you are a new subscriber who came directly to Googly: the short version of what you are about to read is that Cerebras Systems completed one of the largest U.S. technology IPOs in recent years, and its 209-page S-1 contains a set of accounting, governance, and capital-structure disclosures that the market, in its current configuration, appeared entirely unbothered by. My argument is not about Cerebras specifically. It is about a market that has been in something close to a permanent bull configuration since 2008, and what that means for the wave of super-IPOs — SpaceX, OpenAI, Anthropic, and others — now approaching the window.

What follows is the full analysis: every claim sourced to a page in the filing, the EPS dilution arithmetic laid out in full, the Note 7 share-count story, the cash-flow and financing structure, the governance rarity question with data, and the deregulation argument I could not fit in Forbes. It runs long. That is the point of Googly — this is where the analysis lives that cannot be compressed without losing something important.

The full analysis begins below.

All financial figures are drawn from the Cerebras Systems Inc. Registration Statement on Form S-1, filed April 17, 2026 (File No. 333-295145; Accession No. 0001628280-26-025762), available at the SEC EDGAR database. Every factual claim is accompanied by an S-1 page number.

Cerebras Systems (CBRS) priced its initial public offering at $185 per share, raised $5.55 billion, and opened at $350 in its Nasdaq debut on May 14, 2026 — 89% above the offering price — and closed its first trading day at $311.07, a 68% first-day gain over the offer price. By proceeds raised and first-day performance, it ranks among the largest U.S. technology IPOs in recent years. The enthusiasm is understandable. Artificial intelligence is transforming the economy, and Cerebras has built something genuinely remarkable: a chip the size of an entire silicon wafer that delivers inference speeds it claims are fifteen times faster than NVIDIA’s flagship products (S-1, p. 1). Few would dispute that the technology is impressive.

My question is a different one. What does the market’s reaction tell us about how investors price financial reporting quality? And does the financial reporting embedded in those 209 pages give us reason to think that quality mattered to the buyers at all?

If I had to summarize what I found in Cerebras’s registration statement in a single sentence, it would be this: the filing contains enough accounting, governance, and capital-structure complexity that a market genuinely attentive to earnings quality would have priced it differently.

Before examining those concerns, it is worth stating the bull case clearly. A persuasive investor could argue that the 2025 income statement is backward-looking, that UAE customer concentration will be diluted as OpenAI and AWS deployments ramp, and that early-stage internal control weaknesses are a remediable feature of fast-scaling technology companies. The OpenAI Master Relationship Agreement, signed in December 2025, and the AWS binding term sheet, signed in March 2026 (S-1, p. F-41), represent the kind of strategic momentum that can transform a company’s trajectory quickly. That momentum is real. The question is not whether the bull case exists; it is whether the current valuation already prices in its success, with little margin for the things the S-1 candidly discloses.

Cerebras is also only the first of what promises to be an extraordinary wave of technology offerings. SpaceX, OpenAI, Anthropic, and a cohort of AI infrastructure companies are widely expected to approach the public markets over the next one to three years, carrying valuations in the hundreds of billions and business models that are, in some cases, even less mature than Cerebras’s. If investors are setting aside financial reporting complexity at the Cerebras scale, the question is what happens when companies of that magnitude arrive at the window. The accounting quality, governance architecture, and capital-structure questions that make Cerebras worth studying are not peculiar to Cerebras. They are the kind of questions that will appear, in some form, in nearly every major technology offering that follows it — and the market’s response to Cerebras tells us something important about how much analytical weight those questions will carry.

A Profit That Is Not Quite What It Seems

Let us begin with the headline financial result. The S-1 reports net income of $237.8 million for fiscal year 2025, a dramatic swing from a net loss of $481.6 million in 2024 (S-1, p. 17). On its face, this looks like a company that has turned the corner. Much coverage of the IPO cited this profitability as a sign of financial maturity.

Buried a few pages deeper is a different story. The company itself — to its credit — discloses a non-GAAP net loss of $75.7 million for the same year (S-1, p. 18). The gap between GAAP income and non-GAAP loss is $313.5 million. The primary explanation is a one-time gain of $363.3 million from the extinguishment of a forward contract liability tied to a failed preferred stock transaction with the UAE-based technology group G42 (S-1, p. 101). When G42’s planned equity investment did not close by April 15, 2025 following CFIUS review, the forward contract expired, and Cerebras recorded a gain equal to the liability’s carrying value. The extinguishment was non-cash: no settlement payment was made or received. The company’s operating loss for the year was $145.9 million (S-1, p. 98).

There is nothing improper about this accounting. The extinguishment appears to be properly recognized. But the arithmetic is stark: without the $363.3 million non-cash gain, the company’s 2025 income statement would not read like a newly profitable operating business. It is worth asking how many investors buying at or near the $350 opening price understood that the GAAP profit line was shaped almost entirely by a canceled preferred stock transaction involving UAE-based G42 rather than by operating performance.

The Per-Share Arithmetic: From $1.64 to Negative $0.40

For fiscal year 2025, the same classes of securities reappear — but on the other side of the ledger. The $363.3 million forward-contract extinguishment gain converted a net loss year into a net income year, and with net income comes the obligation to include dilutive securities in the EPS denominator. The 2025 diluted calculation accordingly absorbs 91.5 million preferred shares (on an as-converted basis, now reflecting the addition of Series G) and 26.7 million option shares under the treasury stock method — more than tripling the denominator from 53.6 million basic shares to 171.8 million diluted shares (S-1, p. F-21). Basic EPS of $1.64 becomes diluted EPS of $1.38, a 16% reduction. The one-time gain is thus doing double duty: it generated the appearance of profitability, and it activated the dilution mechanism that immediately erodes per-share earnings on that same elevated numerator.

Two further categories remain outside the 2025 diluted count. First: 15.2 million RSUs whose vesting condition is the IPO itself — the liquidity event that has now been completed (S-1, p. F-21). Second: 33.4 million Class N shares underlying the OpenAI warrant, for which no vesting conditions had been met at year-end, though certain tranches were assessed as probable of achievement (S-1, p. F-31). Adding the RSUs brings the post-IPO diluted share count to approximately 187 million, reducing EPS to roughly $1.27. The OpenAI warrant represents a further claim on the earnings base as milestones are met. Viewed through the non-GAAP lens — the one the company itself uses to describe underlying economics — the picture shifts more dramatically still: the non-GAAP net loss of $75.7 million divided by that 187-million post-IPO count yields approximately negative $0.40 per share. Basic GAAP EPS of $1.64, post-IPO diluted EPS of roughly $1.27, and non-GAAP economic EPS of negative $0.40 describe the same year from very different vantage points. None of these figures is hidden; all three are derivable from the filing. Whether the spread between them is being priced is the question this article is asking.

The reason these per-share figures matter to valuation is not arcane. Equity markets price companies on multiples of earnings, and the earnings number changes materially depending on which share count is used. At the IPO price of $185, basic GAAP EPS of $1.64 implies a trailing price-to-earnings ratio of approximately 113 times. Applying the post-IPO diluted count of approximately 187 million shares to the same GAAP numerator reduces EPS to roughly $1.27 and pushes the implied P/E to approximately 146 times — not because the business improved, but because more shares must now share the same earnings pool. At the first-day closing price of $311.07, those same calculations yield trailing P/E ratios of approximately 190 times on the basic figure and approximately 245 times on the post-IPO diluted figure. And on the non-GAAP measure the company itself uses to describe underlying economics — a net loss of $75.7 million — there is no P/E ratio at all. These are not equivalent starting points for a valuation model. Analysts building discounted cash flow projections for Cerebras after its first quarterly report will need to decide which share count and which earnings measure they are using. Those choices have large consequences for any fair-value estimate.

Cash Burn, Capital Intensity, and a Loan That Comes with Strings

The income statement, however, is not where the capital-structure story is most clearly told. That story lives in the cash flow statement and the footnotes. Operating cash flow for fiscal 2025 was negative $10.1 million, while capital expenditures totaled $382.7 million (S-1, p. 105). Free cash flow was approximately negative $393 million. This is not a cash-generative business yet; it is a company deploying capital at scale to build data center infrastructure for future service delivery.

To finance that build-out, Cerebras raised $1.1 billion in Series G preferred stock in September and October 2025, $1.0 billion in Series H preferred stock in January 2026, and received a $1.0 billion working capital loan from OpenAI in January 2026 (S-1, p. 24). The OpenAI loan bears 6% annual interest, matures no later than December 31, 2032, and is subject to a secured promissory note (S-1, p. 91). What the filing makes clear, but what is easy to miss, is the loan’s contingent repayment trigger: if the Master Relationship Agreement is terminated for any reason other than OpenAI’s material uncured breach, OpenAI may direct the bank to cease complying with Cerebras’s instructions regarding the loan funds and may require immediate repayment of the outstanding principal and accrued interest (S-1, p. 25).

In other words, one of the company’s largest pre-IPO financing arrangements and its largest disclosed revenue relationship are tied to the same counterparty by the same agreement. Investors calibrating the downside scenario should start there. What portion of 2026 to 2028 revenue is contractually committed by customers other than OpenAI? What capital expenditures must be incurred before revenue can be recognized under the MRA? What refund obligations apply if capacity milestones are missed? These questions are addressed in the risk factors (S-1, p. 24–29), but it is not clear they have been priced into a basic opening-day market capitalization of roughly $67 billion.

The Customer Concentration Problem

At a basic market capitalization of approximately $67 billion on the day of debut — and a potentially higher fully diluted figure once options, RSUs, and the OpenAI warrant covering 33.4 million shares are included (S-1, p. 14) — the market was paying roughly 130 times trailing revenue of $510 million (S-1, p. 17). That is an extraordinary multiple. It implies a belief that Cerebras can sustain and accelerate its revenue trajectory for many years. Whether that belief is justified depends heavily on a question the S-1 raises with unusual candor: who exactly is buying Cerebras’s products right now?

For the fiscal year just completed, two UAE-affiliated customers accounted for the overwhelming majority of revenue. The Mohamed bin Zayed University of Artificial Intelligence in Abu Dhabi accounted for 62%; G42 Holding Ltd, also Abu Dhabi-based, accounted for another 24% (S-1, p. 24). Together, two UAE-affiliated entities generated 86% of the company’s revenue. MBZUAI represented 77.9% of accounts receivable at year-end 2025 (S-1, p. 25). The geopolitical dimension compounds the concentration: a substantial majority of 2025 revenue came from customers in a region the S-1 acknowledges is subject to export control licensing, geopolitical tensions, and CFIUS scrutiny (S-1, p. 56).

The S-1 discloses that G42 and MBZUAI are considered related parties with respect to each other under ASC 850 (S-1, p. 25). To be precise: this does not automatically mean that Cerebras’s own transactions with either customer constitute related-party transactions in the technical accounting sense — the relevant question is whether the pricing reflects arm’s-length terms, which the S-1 states the company believes it does (S-1, p. 171). But the ASC 850 disclosure raises a distinct economic question: if the two customers accounting for 86% of revenue are themselves interconnected, is this truly diversified demand, or a single Abu Dhabi-centered demand cluster? The basis for the arm’s-length conclusion is difficult to verify independently without publicly available comparable transactions or a described independent assurance process capable of testing the company’s related controls and pricing procedures.

Following the OpenAI arrangement, the S-1 notes that Cerebras is prohibited from selling certain products and services to named competitors of OpenAI (S-1, p. 22). That is an unusual contractual constraint for a company positioning itself as a broad, open AI infrastructure provider. It narrows the customer universe at precisely the moment when the market is valuing the company on the assumption that universe is large.

Material Weaknesses: An Unremediated Two-Year Problem

The S-1 discloses two material weaknesses in internal control over financial reporting, present in both fiscal year 2024 and fiscal year 2025 (S-1, p. 64). The first concerns inadequate accounting expertise: the company lacked sufficient resources to timely review account reconciliations and apply GAAP correctly across revenue recognition, inventory management, data center asset accounting, and equity administration. The second concerns IT general controls: the company failed to maintain adequate information technology controls, including effective segregation of duties.

A material weakness means there is a reasonable possibility that a material misstatement of the financial statements would not be prevented or detected on a timely basis. The company has identified these weaknesses and describes remediation efforts underway (S-1, p. 64). But they are not remediated. They will not be considered fixed until management designs controls that operate effectively for a sufficient period and are tested. As of the IPO, that testing has not been completed.

The S-1 does not describe an internal audit function or an equivalent independent assurance mechanism. Whether one exists in practice is not established by the filing. The relevant concern is not that absence from the document proves absence from the organization; it is that no described internal audit function, combined with unremediated material weaknesses across multiple financial statement areas and complex revenue arrangements, is precisely the combination that heightens audit committee oversight responsibilities most sharply.

The company also changed its independent auditor — from BDO USA to KPMG — in November 2025, mid-way through the fiscal year being audited (S-1, p. 210). The S-1 discloses no disagreements with BDO and no reportable events beyond the pre-existing material weaknesses. The disclosure does not establish a problem with either audit; it establishes a diligence question. The audit committee should be prepared to explain, in executive session, the rationale for the change and what transition procedures KPMG used to gain comfort on opening balances and the prior year’s control findings. Auditor changes mid-year, even when benign, warrant more than a brief footnote in a $5.5 billion offering document.

To be clear: the presence of material weaknesses does not mean the financial statements are wrong. The auditors issued unqualified opinions on both years. But it does mean the underlying control environment carries structural fragility. Academic research, including Doyle, Ge, and McVay’s 2007 study of accruals quality and internal control over financial reporting, has documented that companies with material weaknesses — particularly those affecting multiple accounts as here — tend to exhibit lower accruals quality and weaker internal reporting environments. The first-day price action gives little visible evidence that these risks meaningfully constrained demand.

Governance: A Concentration of a Different Kind

The governance structure also warrants attention. The founder and CEO, Andrew Feldman, serves simultaneously as chairman of the board (S-1, p. 149). Holders of Class B common stock — primarily the founders and early investors — hold twenty votes per share compared to one vote per share for the Class A shares sold in the offering (S-1, p. 13). The board carries a classified structure with staggered three-year terms (S-1, p. 149), and the certificate of incorporation requires a supermajority of 66.7% to amend key provisions (S-1, p. 184). None of these arrangements are unusual for a Silicon Valley founder-led technology company at IPO. What the S-1 admirably discloses in full — and what is somewhat unusual — is that Mr. Feldman pled guilty in 2007 to one count of circumventing accounting controls at a prior employer, Riverstone Networks, and entered into a civil settlement with the SEC arising from the same facts (S-1, p. 148). He served three years of probation and paid a fine.

I raise this not to relitigate an eighteen-year-old matter or to attribute current conduct to a past conviction. The concern is not recidivism — a 2007 guilty plea carries limited predictive power about a company’s current reporting culture. The concern is governance design. When a prior controls-related conviction, dual CEO-and-chair status, twenty-to-one voting rights, supermajority charter protections, and unremediated ICFR weaknesses coexist in the same organization, the board owes investors a clear and affirmative explanation of why its oversight architecture is sufficient. The S-1 discloses the facts. That architectural explanation is not there.

The broader question is how often this combination appears in large U.S. technology IPOs. Material weaknesses at IPO have become relatively common: according to KPMG’s 2024 IPO Material Weakness Study, between 40 and 58 percent of traditional U.S. IPOs in each of the three years from 2021 to 2023 disclosed material weaknesses in their initial registration statements. What is considerably less common is the specific pairing Cerebras presents: unremediated material weaknesses in accounting controls, disclosed in two consecutive annual filings, at a company whose founder and CEO has a prior criminal conviction for circumventing accounting controls at a previous employer. Among large U.S. technology IPOs since 2000, I am not aware of a comparable instance where both conditions were simultaneously disclosed. That is not a categorical claim — no comprehensive published database of such pairings exists — but it is a practitioner observation grounded in reading S-1 filings across many market cycles. The combination is unusual enough to warrant more than the single paragraph the governance risk factor devotes to it.

The Deferred Stock Compensation Overhang

One more financial reporting item deserves attention because it will affect the reported earnings that new investors see first. The company granted restricted stock units with a dual vesting condition: employees must both complete a service period and experience a liquidity event — the IPO — before the awards vest (S-1, p. 98). Because the liquidity condition had not been satisfied as of December 31, 2025, the company recognized zero stock-based compensation expense for these RSUs in either 2024 or 2025. Total unrecognized compensation expense for these awards was $343.5 million at year-end (S-1, p. F-34). The S-1 states that had a liquidity event occurred as of December 31, 2025, $150.5 million of that amount would have been recognized immediately.

The IPO is that liquidity event. Investors should expect a large stock compensation charge in the quarter in which the offering closes. The charge does not affect cash and is a disclosed, non-recurring item that IPO analysts will be aware of. The risk is not that sophisticated institutional investors will be surprised. The risk is that retail investors comparing post-IPO GAAP earnings to S-1 financials will confront a step-down in reported results that the S-1 anticipates but does not present prominently. This deferred compensation overhang is a feature of the accounting structure, not a sign of impropriety — but it is one more item in the S-1 that rewards careful reading.

Is Earnings Quality Being Priced?

In a Forbes column I wrote in December 2022, I asked a room of sixty chief investment officers of state pension funds who among them routinely reads a 10-K. About six hands went up. The question here is related but distinct: not whether investors read the S-1, but whether, having read it or not, the market priced what was in it. A company can fully disclose its accounting complexity and have that disclosure produce no effect on its valuation. That outcome is arguably more troubling than simple inattention.

There is a structural explanation that deserves more than passing acknowledgment. For roughly fifteen years following the Federal Reserve’s quantitative easing programs of 2008 — extended and amplified through the COVID-era liquidity surge — global equity markets operated in conditions that were, in aggregate, highly forgiving of financial reporting complexity. Rates stayed near or below zero, asset prices rose substantially, and the marginal cost of overlooking accounting nuance fell toward zero. In that environment, investors were not necessarily irrational to de-prioritize the footnotes: the discount rate that would penalize earnings quality degradation was itself compressed by policy. We have, in something close to a permanent bull market, trained a generation of investors and analysts to treat financial reporting quality as a secondary variable — something to revisit if things go wrong, but not a primary input to valuation during the upswing. Cerebras’s first-day price action is one data point in a sequence that began in 2008. The question is whether it will take a significant correction before investors rediscover that the sequence has consequences.

There are several structural reasons for declining interest in financial disclosure across markets broadly. Passive indexing now accounts for a very large and growing share of equity assets under management, but passive funds are generally not the marginal buyers setting an IPO’s opening auction — that role falls to institutional allocators, book-running investors, hedge funds, and retail participants in the aftermarket. The point about indexing is less about who sets the IPO price and more about what happens afterward: as passive vehicles accumulate shares following index inclusion, their appraisal of individual company financials is, by construction, irrelevant to their investment decision. Quantitative funds run models on extracted financial data at scale; the nuances that a careful analyst might observe — say, that the primary driver of reported profitability is a non-cash contract cancellation gain — are notoriously difficult to encode systematically. And in an era of AI enthusiasm, a company’s technology narrative can overwhelm the signal from its accounting. The more compelling the story, the less incremental information the numbers seem to provide.

Cerebras is, in many respects, a compelling story. Wafer-scale integration is a genuine technical achievement. The OpenAI relationship is central to a disclosed $24.6 billion backlog of remaining performance obligations (S-1, p. F-19), most of which appears tied to the OpenAI arrangement, though the timing, enforceability, and revenue-recognition mechanics of that backlog depend on capacity delivery milestones that have not yet been met. The AWS term sheet, if it results in definitive agreements, could be genuinely transformational. These are real strategic developments that justify investor attention.

In the late 1990s, Alan Greenspan warned of “irrational exuberance” — a willingness to pay extraordinary prices for assets based on narrative rather than fundamental analysis. The parallels to the current AI investment cycle are imperfect. The underlying technology is real. The economic opportunity is large. But some of the valuation behaviors are familiar: a company that generates most of its revenue from two customers in a geopolitically sensitive region, reports GAAP profitability driven primarily by a one-time non-cash event, carries unremediated internal control weaknesses across multiple financial statement areas, and enters the public market at 130 times trailing revenue. Each of those observations is, on its own, manageable. Together, they raise questions that deserve more than a first-day trading surge as an answer.

Maybe that valuation will prove correct. The history of technology investing is littered with examples of skeptics who were wrong. But the history of financial reporting is also littered with examples of investors who paid for a story and discovered, too late, that the financial statements contained a different one.

A Note on Making IPOs Great Again

There is a recurring argument, made with renewed energy in recent years, that America’s disclosure requirements for newly public companies are too burdensome — that compliance costs, litigation risks, and accumulated governance mandates have driven companies away from public markets, concentrated wealth-creation in private hands, and limited ordinary investors’ access to the early stages of enterprise growth. The argument is not wrong in its diagnosis. The number of listed companies has declined by roughly forty percent since the mid-1990s. Going public is genuinely expensive. And the JOBS Act of 2012, which created the emerging growth company framework Cerebras already uses, was a first-generation response to that problem. These concerns deserve a fair hearing and have support across the political spectrum.

Cerebras is worth considering in this context precisely because it is already an EGC, already benefiting from reduced reporting obligations, and yet its S-1 contains exactly the accounting complexity this article has described. A strong version of the deregulatory argument might observe that investors appeared not to price any of it on day one — which invites the question of why we require the disclosure at all. In a bull market, that argument has surface plausibility. Capital flows toward compelling narratives regardless of what the financial statements say, and financial reporting quality is discounted whether or not it is fully disclosed.

That is precisely where the argument becomes dangerous. Cutting disclosure requirements during a bull market is easy because the information would be largely ignored regardless. The cost shows up at the turn of the cycle, when investors who overpaid begin to examine what they actually bought. The history of financial reporting failures — from the savings and loan crisis through Enron and WorldCom and the restatement wave that followed the dot-com collapse — unfolded after periods of exactly this kind of bull-market inattention. In each case, the information needed to understand the losses had, technically, been disclosed. What changed at the turn was not the accounting but investors’ willingness to engage with it — and in some later instances, their ability to obtain it at all where deregulatory changes had narrowed the disclosure perimeter.

Reducing disclosure requirements during an AI-driven boom may appear sensible today, when the information would be largely set aside regardless. It will look different when the cycle turns and investors discover that the earnings quality metrics, control environment findings, and related-party disclosures they need to understand their portfolios were never required to be provided in the first place. Making IPOs more accessible and more numerous is a legitimate policy goal. The right measure of success, however, is investor outcomes over a full market cycle — not deal volume at the peak.

What Should We Make of This?

Cerebras has disclosed the risks and accounting nuances described above with considerable transparency. The S-1 is, in that sense, doing exactly what it is supposed to do. The question is whether anyone is reading it closely enough.

For audit committee members, institutional investors, and analysts engaged with this filing, the items described here are not grounds for alarm but they are grounds for rigorous inquiry. How will the company demonstrate that its material weaknesses have been remediated, and on what timeline? What is the plan for establishing a described internal audit function? How are the MBZUAI and G42 contracts priced relative to arm’s-length comparables, and who has verified that? What events would trigger the immediate repayment of the $1 billion OpenAI working capital loan, and what is the company’s contingency plan if those events occur (S-1, p. 25)? And what non-UAE revenue, contractually committed, can the company point to for 2026 and 2027?

Jack Ciesielski, the veteran financial reporting watchdog, once told me: “at some point, there will be an accounting tragedy that reinforces the importance of looking behind the numbers.” I am not predicting that Cerebras will be that tragedy. But the conditions under which such tragedies tend to occur — enthusiastic markets, concentrated revenue, control environment weaknesses, and one-time accounting gains presented alongside continuing operating losses — are present here, disclosed in plain sight, yet outweighed — at least on day one — by demand for the AI infrastructure story.

That, perhaps more than any single number in the S-1, is the thing worth examining.

What Mike Green’s Argument Means for Every Corner of Finance and Corporate Life

IMPLICATION I
For Sell-Side and Buy-Side Analysts — The Information You Sell Has Modest Demand, If Any!

Mike Green’s data lands like a gut punch for anyone in the business of fundamental research: approximately 7% of market trading activity involves any fundamental analysis whatsoever, down from 80% in 1995. The fundamental signals you are trying to transmit — this company is worth X based on its cash flows — are being drowned out by mechanical price spirals driven by $1 of passive buying creating $17 of market capitalization. The price does not respond to your earnings revision. It responds to whether net passive flows that day were positive or negative.

This means the entire architecture of sell-side research — modeling earnings, adjusting estimates, updating DCF assumptions, issuing price targets — is operating under a false premise. The price target assumes a world where prices eventually revert to fundamental value. Green’s data suggests that mechanisms are now so weakened that mean reversion may take a generation, not a quarter.

The honest response is to add a passive distortion premium to every large-cap valuation. For any S&P 500 constituent, some meaningful fraction of the current price reflects nothing but mechanical index weight and will persist as long as passive flows persist. Your job, as an analyst, is to figure out how large that fraction is and whether your clients understand they are partly paying for it.

IMPLICATION II
For Investor Relations Professionals — Getting Into the Index Matters More Than Your Investor Day

If Green is right that index inclusion automatically inflates a company’s market capitalization by roughly 40%, then the entire discipline of investor relations needs rethinking. The traditional IR model is built on the premise that informed investors respond to information. But passive doesn’t care about your investor day. Vanguard and BlackRock are not going to underweight you because your gross margin guidance disappointed. They hold you in exact proportion to your market cap regardless.

What matters is not how good your story is. What matters is: are you in the index? And if so, what is your index weight? Index weight determines passive buying pressure, which drives price, which determines your weight, which determines more buying. Getting added to the S&P 500 matters more than any earnings beat. Moving from the 400th to the 100th largest company by market cap matters more than growing revenue 20%. IR professionals should be asking: which analysts and index methodologists have the power to change our classification? How do we maximize our weight in the indices that matter?

IMPLICATION III
For the CFO — Your Stock Price Is Lying to You About Your Cost of Capital

This is the implication I find most alarming as a professor. Green estimates the S&P’s multiple would be approximately 65% lower if passive did not exist. For a typical large-cap constituent, approximately 40% or more of the current stock price reflects passive mechanical buying, not a market assessment of intrinsic value.

The CFO’s cost of equity is derived from that stock price. The WACC, which drives every capital allocation decision, is a function of a market capitalization that is partially fictitious. If your equity is overvalued by 40%, your WACC is artificially low. Artificially low WACCs green-light projects that a proper cost of capital would reject. You are overinvesting. You are chasing growth at scale that the market is mechanically rewarding regardless of returns.

Green is explicit about Apple as an illustration: no meaningful growth in nearly a decade, yet at the highest valuation in its history. Under any honest DCF, you would struggle to justify that price. But Apple’s index weight is enormous. The passive machinery buys more of it every day simply because it is large. The CFO of Apple, looking at their stock price, would conclude the market is rewarding them for something. The market is not rewarding them for anything. And share buybacks become narcotic in this environment: 100% of the buyback flow inflates your market cap, raises your index weight, attracts more passive buying, lifts the stock price, and raises the value of management’s stock options. It is circular self-enrichment enabled by the passive ecosystem.

IMPLICATION IV
For the Board — You Can Sleepwalk and Not Get Caught

The monitoring function of the board rests on one foundational premise: bad decisions eventually show up in a falling stock price. Shareholders vote out directors. Activists arrive. If Green is right, that mechanism is severely impaired for indexed companies. The passive bid provides a mechanical floor divorced from fundamentals. A board that presides over years of mediocre performance may never see the stock price decline that would signal ‘something is wrong,’ because passive flows mask it.

Green is explicit: the S&P has returned roughly 18% per year simply because of passive growth. Every board in the S&P 500 has been surfing a structural tide for a decade. The board that looks at that return and congratulates itself for governance excellence is confusing luck for skill. Passive holders are constitutionally incapable of exit — they cannot sell a stock out of the index because that would generate tracking error. Voice without credible exit is weak. The feedback loop of market discipline is severed.

IMPLICATION V
For the Founder-CEO Who Can Trample Shareholder Rights — Passive Is Your Best Friend

Corporate governance scholarship assumes markets discipline bad actors through the informed short seller, the active institutional voter, and the falling stock price. Green’s framework suggests all three channels are weakening simultaneously. Short sellers are being wiped out. Active institutions are being redeemed. And the stock price of an indexed company has a mechanical support structure that can mask fundamental deterioration for years.

A founder-CEO who runs the company primarily for personal benefit — through related-party transactions, sweetheart compensation, empire-building acquisitions, or dual-class share structures — faces a much softer disciplinary environment than the textbooks assume. The market that is supposed to price in bad governance is increasingly not pricing in anything. It is allocating based on market cap weight. A company that is 1% of the S&P 500 by market cap will receive that proportionate passive bid every single day, regardless of whether the CEO is building long-term value or harvesting the firm for personal gain.

IMPLICATION VI
For Short Sellers — You Are Being Executed by a Force You Cannot Fight

Green’s XIV trade worked because he was betting against a specific passive product, not against the broad market. Shorting the broad market in a world of passive-driven price inflation is a different and much more brutal proposition. The traditional short-selling mechanism requires that overvalued companies eventually get sold down by investors who recognize the problem. In a passive-dominant market, those sellers are being systematically redeemed and replaced by passive buyers who will never sell based on fundamentals.

The short seller’s thesis may be 100% correct — the company is worth 30% less than its trading price — but the passive inflows will keep buying the stock regardless. The short seller carries the cost of the borrow, the daily loss, and the psychological burden of watching the market make them look foolish, all because the mechanism that was supposed to vindicate their thesis has been disabled. Aggregate short interest as a fraction of market cap has collapsed over the past twenty years. The short-selling ecosystem is dying. The market has not gotten more efficient — it has gotten less punishing of overvaluation, which is a completely different thing.

IMPLICATION VII
For Finance and Accounting Curricula — Everything You Learned About the EMH Is Worth Questioning

I want to be direct about this one, because I teach this material. The Efficient Market Hypothesis rests on the assumption that prices reflect information because informed investors trade on it and make money doing so. The Grossman-Stiglitz equilibrium tells us there must be a reward for information gathering; otherwise nobody would gather information, prices would become uninformative, and profits would emerge — attracting informed trading back in. The market self-corrects.

Green’s argument, backed now by the Inelastic Market Hypothesis and a growing body of evidence, is that we have created a system that breaks that equilibrium. We have channeled 95 cents of every retirement dollar into passive vehicles that deliberately do not gather information. We have created a negative loading on the active manager who does. We have inverted the Grossman-Stiglitz equilibrium. The result is that the EMH may be empirically false — not marginally false. The multiplier is 17x, not 1 cent. Every DCF model should now include an explicit discussion of the passive distortion factor. The cost of equity should be presented alongside an honest discussion of whether that cost is real or passive-inflated. The inelastic market hypothesis should be taught alongside — or in some courses instead of — the EMH.

IMPLICATION VIII
For Reverse DCF — What Are You Actually Solving For?

The reverse DCF assumes the current stock price is fully determined by investors’ collective assessment of future cash flows. If 40% of the stock price is a passive distortion premium — a mechanical bid that has nothing to do with fundamentals — then the reverse DCF is solving for the wrong thing. It is treating the passive premium as a forecast and asking what fundamental trajectory would justify it.

The correct formulation would be: Current Price = PV(Future Cash Flows) + Passive Premium. The reverse DCF should isolate the first term, not treat the sum as if it were the first term alone. When you do a reverse DCF on Nvidia and conclude the market is pricing in 30% compound revenue growth for fifteen years, you are overstating what ‘the market’ actually believes about Nvidia’s fundamentals. Part of that implied growth is not a growth forecast at all. It is the mathematical consequence of Nvidia’s enormous index weight multiplied by the passive flow multiplier. No one at Vanguard decided Nvidia deserves a 30% growth premium. The algorithm just keeps buying more of it because it is large, which makes it larger, which means the algorithm buys more.

IMPLICATION IX
For the Board and Governance Community — Exit-Based Discipline Is Structurally Dead

The cornerstone of monitoring board literature — Jensen and Meckling, Fama and Jensen, Shleifer and Vishny — is that concentrated, informed, active shareholders discipline management through both voice and exit. Exit means selling, which moves prices down and signals dissatisfaction. Passive shareholders are structurally incapable of exit. Voice without credible exit is weak. As passive ownership of large-cap stocks rises toward 60% and beyond, the fraction of the shareholder base that retains the ability to exit shrinks toward the point where exit-based governance is no longer viable.

We are building a corporate governance system in which the dominant shareholders are constitutionally incapable of the exit discipline that governance scholarship assumes as foundational. We are replacing informed, active monitoring with mechanical, price-insensitive holding. And we are surprised that boards are not being held accountable?

IMPLICATION X
For Your 401(k) — Should You Sell and Move to the Himalayas?

Mike Green: The single best risk trade: when you get your job and your employer matches your contribution, put it straight into fixed income. You’re doubling your return through the matching and you have no risk. For equities — no, I don’t think forward returns are favorable. We’re approaching an inflection point on contributions as the boomers move into retirement and begin to liquidate. Those net flows create the conditions of the decline I’m worried about. That is certainly what the math suggests.

The deeper point is Green’s challenge to the entire asset accumulation model. A retirement system should be built on income replacement, not asset hoarding. A well-functioning retirement system would convert savings into stable income streams. What we built instead is a system where every individual self-insures for a retirement of undetermined length, hoards assets they are afraid to spend, and in so doing withholds the spending that would become the income of the young. The accounting identity is brutal in its simplicity: the spending of the old becomes the income of the young. We built a machine that disrupts that flow. And then we are surprised by the consequences. The most recent issue of the Financial Analysts Journal contains a paper by Coimbra et al, that specifically focuses on this (https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4406011).

IMPLICATION XI: For Policymakers and Reformers — How to Actually Reverse This

Green does not leave the problem without a diagnosis of the cure. His policy prescriptions, delivered at the end of his talk, are specific, layered, and in several cases draw on legal precedent that already exists but has simply been abandoned. Taken together, they suggest a reform agenda that operates on five distinct levers.

Lever 1: Change the QDIA Designation — The Source Code of the Problem

The most direct fix goes to the origin of the distortion itself. The Pension Protection Act of 2006 designated passive target date funds as the qualified default investment alternative — the vessel into which 95 cents of every retirement dollar automatically flows. That was a regulatory choice, not a law of nature. Changing it does not require dismantling the retirement system. It requires the Department of Labor to revisit and broaden what qualifies as a QDIA. A diversified set of default options — including actively managed vehicles, annuity products, and balanced funds — would stop the single-pipe flow of retirement capital into market-cap-weighted passive funds. This is the highest-leverage intervention available. Everything else is downstream of this. The problem with it, of course, is political: the passive fund industry has enormous institutional power, and any move to alter the default would be framed — dishonestly — as harming ordinary retirement savers.

Lever 2: Enforce Antitrust Against Index Provider Concentration

Green is unambiguous: “We cannot have a scenario in which a financial market is controlled by a few participants like Vanguard, BlackRock, and State Street.” The Bork doctrine that gutted US antitrust enforcement in the 1980s — which reduces the entire antitrust question to short-run consumer price effects — is the wrong framework for financial markets, where the harm is not higher prices but distorted price signals and systemic fragility. The Big Three collectively vote the proxies of, and therefore exercise de facto control over, a majority of the S&P 500. This is a concentration of financial power with no historical precedent. Applying existing antitrust frameworks to index provider concentration — or introducing new legislation that treats dominant passive market share as a systemic risk threshold — would force structural remedies: divestiture of voting rights, mandatory unbundling, or caps on passive market share.

Lever 3: Unbundle Target Date Funds — A 1930s Problem With a 1930s Solution

Green notes this with some exasperation: “We addressed this in the 1930s. It’s called bundling.” Target date funds offered by Vanguard invest only in Vanguard products. Target date funds offered by Fidelity invest only in Fidelity products. This is precisely the kind of tied-product bundling that antitrust and consumer protection law has long prohibited — the same logic that prevented AT&T from requiring you to buy their handset when you subscribed to their phone service. The legal basis is already settled. We simply failed to enforce it. Requiring open-architecture target date funds — where the default vehicle must be able to include products from any qualified manager, not just the platform sponsor — would reintroduce genuine competition for retirement dollars and break the structural moat that the largest passive managers have built around the QDIA designation.

Lever 4: Rescind the SEC’s Special Leverage Exemptions for Index Providers

In 1994, the SEC gave index providers a no-action letter exempting them from the leverage prohibition in the Investment Company Act of 1940 — allowing them to use futures for replication. In 2019, when S&P 500 Growth indices exceeded concentration thresholds that would have disqualified them as diversified funds, the SEC again granted an exemption. These regulatory gifts were handed to index providers on the implicit theory that passive investing is benign and democratizing. We now know that theory is false. Rescinding these exemptions — or at minimum requiring that index providers operating under them demonstrate that their concentration and leverage do not create systemic risk — would re-level the playing field between passive and active strategies. It would also slow the feedback loop between passive buying and index concentration by forcing index providers to operate under the same constraints as everyone else.

Lever 5: Tax Buybacks Differently from Dividends — and Make the Math Honest

Green’s proof is arithmetically simple and politically radioactive. A company that buys back stock instead of paying a dividend captures 100% of the passive reinvestment flow for itself, not the 2% that its market weight would generate through dividend reinvestment. This disproportionately inflates the stock prices — and potentially the compensation — of management at the largest indexed companies. If this bears out in the data, do we need to consider a meaningful differential tax treatment for buybacks and special dividends.

Lever 6: Convert the Retirement System from Asset Accumulation to Income Replacement

This is Green’s deepest and most radical proposal — and the one most consistent with how every functioning pension system outside the US actually works. The problem with the current system is not just passive investing. It is that the system requires every individual to self-insure for a retirement of indeterminate length by accumulating a large enough asset pool. That is an extraordinarily complex and uncertain task, and it is why seniors underspend relative to their assets — they don’t know how long those assets need to last. The solution is to convert accumulated retirement assets into guaranteed income streams: annuities, longevity insurance, defined-benefit-style structures. This would accomplish two things simultaneously. First, it would release the hoarded assets back into consumption — increasing spending by the old, which is the income of the young, which addresses the intergenerational transfer problem Green describes. Second, it would reduce the structural demand for equity-market appreciation as the mechanism of retirement security. You no longer need the S&P 500 to hit 7% per year forever if your retirement income is a fixed stream. The secular demand for passive equity investment would diminish, and with it the passive distortion premium.

What Is Politically Realistic in the Next Five Years?

Lever 6 requires generational political will and is unlikely in the near term. Levers 1 through 5 are all achievable through existing regulatory or legislative authority with moderate political effort. The QDIA revision requires a Department of Labor rulemaking. The antitrust enforcement requires a reorientation of existing law, not new legislation. The bundling prohibition is already legally settled; it needs only enforcement. The SEC exemption rescissions require agency action, not Congressional approval.

What is most likely to happen, if Green’s scenario plays out, is not a deliberate policy intervention but a forced one. The demographic reversal — boomers drawing down — will eventually produce the kind of net passive outflow that runs the multiplier in reverse. That event, when it comes, will do what legislation has not: it will make the distortion undeniable. The question is whether the policy infrastructure to manage it in an orderly way will exist before the event, or whether, as Green suspects, we will be building the lifeboat while taking on water.

Coda: The Voltaire Warning

Mike Green ended his talk with a quote from Voltaire: “Those who can make you believe absurdities can make you commit atrocities.” He left it to his audience to decide who was sharing the bad information — him, or the market.

I think the honest answer is that the absurdity is not in what Green is saying. The absurdity is in what we have collectively accepted: that a market in which 55% of assets are managed by algorithms that have no opinion whatsoever about price can still be called ‘efficient.’ That a multiplier of 17-to-1 is consistent with prices reflecting information. That a system in which every retirement dollar automatically flows into the largest companies regardless of their merits is a fair representation of capitalism. That insider ownership was a signal of management quality in the 1990s when it was actually just a mechanical feature of float-adjusted index rebalancing.

Green is not saying markets will crash tomorrow. He is saying we have built a machine that is slowly consuming the information infrastructure it runs on, that is hollowing out the capital formation system even as it celebrates its own highs, that is exporting this dysfunction to Australia, the UK, Denmark, and everywhere else that looks at American retirement account balances and decides they want what we have.

When the inflection point comes — when the boomers’ net withdrawals exceed the inflows, when volatility becomes self-reinforcing, when the 1% who might try to sell becomes 2% — everyone who built their analysis on the EMH will need to rebuild from scratch. Everyone who priced the cost of capital from a passive-inflated stock will discover their WACC was a fiction. Every board that sleepwalked through a decade of structurally supported stock prices will face the accountability they avoided.

My personal belief that the passive system has become too big to fail. There will be pressure to keep finding new flows to the market. Initially from overseas and later or concurrently from some form of pressure to invest social security obligations into equity markets.

In closing, Mike has, at the very least, made me think hard and question the edifice on which our markets are built. View his statements as testable hypotheses. Gather evidence, and by all means, try to refute his propositions.

Thanks for reading.

Shiva Rajgopal: If you strip out the passive distortion, what would the S&P be trading at?

Mike Green: The net impact would be to lower the multiple by about 65%. Which would make perfect sense, if you actually think about it. The S&P is operating at extreme margins with very low growth prospects. It should be trading at a low multiple. Look at a company like Apple. Hasn’t meaningfully grown in almost a decade. Yet it’s trading at the highest valuations it’s ever traded at. Why? Doesn’t make any sense under a discounted cash flow model. It’s got to be something else.

Mike Green: You’re obviously familiar with the work of Michael Mauboussin and the picking easy games hypothesis — the idea that markets have become more efficient, which makes it harder for active managers to outperform. It’s a hypothesis that maybe active managers are just not very good. The issue is that underlying all of our analysis is the assumption that over time there’s an endogenous return to assets that’s reflective of risk, with cyclical variation around that. But what I’ve actually shown you is that we now have an inflating force — passive flows — driving additional price return out of the market. It’s effectively taken that flat line and turned it into a convex curve with rising valuations over time.

Mike Green: Now the problem is what we think of as alpha — it’s actually just the intercept in the linear equation. Return on my portfolio equals its beta times the market plus some idiosyncratic error term we call alpha. If that remains consistently positive, I’m skilled. The problem is if you use a linear equation to solve a curved surface, another mathematical property emerges: your alphas fall. Your intercepts are pushed down over time. The theoretical relationship is perfectly matched with the empirical relationship. The entire assumption behind Grossman-Stiglitz is that information production is costly and I should be rewarded for doing so. What we’ve actually created is a system that penalizes people for trying to apply information. The harder you work, the more disciplined a value investor you become, the worse your results are going to look relative to the market. Costanza market personified.

“What we’ve actually created is a system that penalizes people for trying to apply information. The harder you work, the more disciplined a value investor you become, the worse your results are going to look relative to the market.” — Mike Green

PART TEN: It Isn’t Just Stocks — Bonds, the Equity-Football Payoff Structure, and the 2022 Rebalancing Crash

Shiva Rajgopal: You’ve said the bond market is equally distorted. Walk us through that.

Mike Green: I manage fixed income, in part because it’s less subject to stupidity. Let me explain by comparing payoff structures. What does the payoff structure of an equity look like? It can go to zero, it can’t go below that in a limited liability framework, and theoretically its value can rise forever — that’s what an equity looks like, with the point of maximum uncertainty off into the future. A high-quality fixed income instrument looks a little bit like an American football in flight. Where’s the point of maximum uncertainty? Typically about two-thirds of the way through its duration. Totally different instruments.

Mike Green: Weighting a fixed income universe on market capitalization is going to basically naturally expose you to the greatest duration uncertainty, and you’re going to put the highest weight on bonds at exactly the point at which they’re going to start trending toward par. It’s a terrible strategy. It’s one of the reasons why fixed income underperforms, but it doesn’t really matter because the game is all about distribution. If target date funds capture 95% of every retirement dollar — which they do today — and target date funds are constructed with market cap-weighted bond indices that charge 3 basis points to help make up for their 50 basis points of underperformance, they’re going to grow. And so actually this is crazy: you’re seeing US fiscal and monetary policy being heavily influenced by totally unthinking investment.

“Weighting a fixed income universe on market capitalization puts the highest weight on bonds at exactly the point at which they’re going to start trending toward par. It’s a terrible strategy. US fiscal and monetary policy is being heavily influenced by totally unthinking investment.” — Mike Green

Mike Green: You guys may be too young to remember negative-yielding bonds, but they were a feature of the market a couple of years ago. The narrative we were told was that it was European insurers engaged in asset-liability matching. That wasn’t true. It was American retirees. Why? Because when you buy things in proportion to their market capitalization, when a long-dated low-coupon bond rises above par, it becomes a larger and larger fraction of the index. This drove an unbelievable expansion of duration in the bond indices. The Austrian century bond was issued in 2020 with a 0% interest rate and a 100-year maturity. It traded to 224. Does anyone know where that’s trading today? Seven. That’s a bond. And who was buying it? Your parents.

Mike Green: Think about it this way: imagine there are only two bonds in the universe, equal issuance — a 2-year bond and a 30-year bond. The Fed cuts interest rates. The 2-year barely moves; duration of about 1.2, rises about 1% for a 1% cut. The 30-year has a duration of around 20 — rises 20%. Now my index is more heavily weighted toward long-dated bonds. Next dollar buys more 30s than 2s. What happens when the Fed unexpectedly hikes? The 30-year falls a lot. Most of the coupons issued in the 2020-to-2022 period are now trading around 65 cents. The net impact is that bond market indices — where that incremental bond money is flowing — are about 35% underweight duration. So when people say nobody wants to buy a 30-year bond, they’re telling the truth. Is it a thoughtful consideration of the fact that that 30-year bond in real terms now offers a 2.6% real yield, which would have caused almost all of your forebears to run screaming from the room to go buy it? Nobody wants to touch it.

Shiva Rajgopal: Why did we go down in 2022? The narratives at the time were all about real discount rates and the return of value investing.

Mike Green: We actually know that there’s very little relationship between interest rates and equities. A brand new paper actually just came out: less than a third of rate changes are actually passed through to equities, with particular emphasis on value stocks. But if I build a portfolio like a target date fund that simply says ‘how much equity should I have? It’s a function of my age. How much bonds? It’s a function of my age’ — well, my age doesn’t change very much. So when the Fed hikes in 2022, it caused bond prices to fall. I’m running a portfolio that’s 65% equities, 35% bonds, and my bond portfolio just fell by 20%. What do I have to do? I have to sell equities and buy bonds. Did I do a discounted cash flow analysis? No. Did I consider the forward earning prospects of the companies in my portfolio? No. I just rebalanced.

Mike Green: Zu Lu, a PhD candidate for Hanno Lustig at Stanford — now a full professor at the University of Washington — wrote on this in 2023. She identified what they call the portfolio rebalance channel and determined that this was the single largest contributor to the puzzlingly large response to monetary policy — and it’s exactly what we saw in 2022. Bonds and equities literally behaved like a double helix. Bonds would sell off, equities would have to be sold to rebalance the portfolio. For those of you following markets at the time, you may remember all sorts of narratives: ‘now we’ve got real interest rates, the worst affected are going to be large cap growth companies with all their growth off into the future, we’re going to see a return to value investing because now we have real discount rates.’ Did any of that happen? No. Didn’t work. The simple reality is interest rates don’t matter in the way that we think. They matter in terms of portfolio construction. Now it’s a flow-based story, not a discounting story.

SECTION ELEVEN: It’s a Global Contagion — The UK Is Sending 55 Cents of Every Retirement Dollar to America

Shiva Rajgopal: Is this a purely American problem?

Mike Green: It depends on the degree of adoption, and foreign markets have their own interesting issues. Australia has actually moved heavily toward index investing for the same reasons as the US. Australia has something called superannuation, very similar to our 401(k) plans — government-sponsored, single fund, you can make allocations. They made those changes in 2014, later than we did but in a more aggressive and centralized manner. And they then encountered an interesting wrinkle: managers approved for participation in the superannuation platforms would occasionally underperform. That’s terrible — you’ve been entrusted with the retirement of individuals. So they introduced legislation in 2020: if you’re an active manager and you underperform your benchmark for two years, you’re banned from participation. So what did everybody do? Indexed. And if you look at valuations in Australia, they show the exact same thing you see in the United States.

Mike Green: The rest of the world has chosen to pursue a US retirement model because it appears so successful. Look at how well our markets have done. Look how rich our seniors are. So, they’ve decided they want to pursue this path. It takes on even more perverse characteristics in places like the UK or Denmark, which is about to make a similar switch. They don’t want to penalize their seniors. They don’t want to restrict them to investing in domestic stocks that would show home country bias. So instead, they choose to allocate on a global basis. That means that in the UK, 15 cents of every retirement dollar is reinvested into the UK. 85 cents is going to the rest of the world. 55 cents of the UK’s retirement funds are flowing to the United States to buy the S&P 500 and our bonds. This is insane when you think about it from a pure generational standpoint. The fact that seniors in the UK are investing less than 15 cents of every dollar into the UK tells you exactly what’s going to happen to long-term productivity in the UK. You cannot possibly replace simple equipment or capital asset depreciation with that type of investment level.

“In the UK, 15 cents of every retirement dollar is reinvested into the UK. 55 cents of UK retirement funds are flowing to the United States to buy the S&P 500 and our bonds. The fact that seniors in the UK are investing less than 15 cents of every dollar into the UK tells you exactly what is going to happen to long-term productivity in the UK.” — Mike Green

Mike Green: As of 2023, about 80% of all retirement accounts had a single product in them: the target date fund. When you guys get jobs, you will be faced with retirement options along the lines of: ‘Congratulations on joining our retirement system. You have 10 investment options: the Vanguard target date fund retiring in 2035, 2045, 2055, 2065, 2075.’ I think they now have 2085 — you’re going to be working a long time. The simple reality is that there are no choices left. When I got my first job, I opened up my 401(k) package and there were 3,000 mutual funds in it. What did I do as a 22-year-old? I kind of randomly picked some. That’s not necessarily a good thing, but it did mean that a lot of money went in very diversified ways. Now it’s all directed the same way.

SECTION TWELVE: What Can Be Done — From the Bork Doctrine to the Bodega

Shiva Rajgopal: Given that you’ve diagnosed the problem, what does the prescription look like?

Mike Green: One — you guys are on the right path. You have a professor who is actually engaging in this stuff. You’re learning it. But nobody else cares. And so things you can be involved in: education. Ultimately many of you will be a fiduciary. You need to understand that all of the data you now look at for S&P 500 returns or bond market returns incorporates feedback loops. We don’t know what the returns to equities look like. Because we corrupted them.

Mike Green: Small is beautiful. The simple reality is that it builds robustness. New managers coming into a market who have to charge slightly higher prices because they don’t have the asset base, but who have a creative and thoughtful idea, should be subsidized. We’ve built a legal framework that penalizes them.

Mike Green: We need to end the preferential treatment of quantitative strategies. I build an index and I can say: this is what my performance is supposed to be. We know that’s a misrepresentation. We also know it’s total garbage. Why do so many people invest in the S&P 500? Because you can show it. What’s Mike Green’s performance going to be? I can’t show that. But I can tell you that over 100 years the S&P has done X, Y, Z, and you should expect something similar. That’s a marketing story. We’ve preferentially enabled quantitative strategies on that basis.

Mike Green: We also need to enforce diversification rules. This should be really straightforward. It doesn’t matter whether your lack of diversification is because the market is not diversified. It’s still not diversified. You should not be receiving an exemption.

Mike Green: One of the most important things — and this is candidly the single most important thing — is that as a society we have to wake back up to what capitalism really is, which is about competition. We have failed to enforce antitrust for decades now. The Bork doctrine that emerged in the 1970s and became the operating regulatory framework for treating antitrust in the 1980s is totally wrong. We need to actually recognize that competition itself is a good. We cannot have a scenario in which a financial market is completely controlled by a few participants like Vanguard, BlackRock, and State Street.

Mike Green: Also: things like target date funds. We addressed this in the 1930s. It’s called bundling. Why is it not illegal to for Vanguard to only offer Vanguard products? I know that sounds like an intrusive state thing, but the simple reality is it has already settled legal basis. We failed to enforce it.

Mike Green: And open market buybacks and accelerated share repurchases. This sounds like heresy. Traditional finance theory will tell you that buybacks and dividends are the same thing. They’re not. Just do the math. If I’m a company in the S&P 500 at 2% of the market capitalization, I pay a dividend on par with the rest of the index. 2% of that could flow back to me in a dividend reinvestment program. That 2% will distort my share price a little bit. But what if instead of paying a dividend, I just buy back my stock? Then 100% of that flow goes into my stock. What does it do? It inflates the value of that stock disproportionately, raising the compensation that senior managers are receiving.

“Traditional finance will tell you that buybacks and dividends are the same thing. They’re not. Just do the math.” — Mike Green

Mike Green: And the last thing: just remember that markets are not designed to deliver retirement services. They’re designed to price the marginal cost of capital in a capitalist society. That’s supposed to improve our allocation of resources. Instead, what we did was decide that markets were about retirement services, and we uniquely blessed publicly traded companies to receive the retirement largess. Historically, somebody would save their own money — I’m going to take 10% of my paycheck and I’m going to set it aside because I really want to open up a bodega. That’s my life dream. But that’s really hard to do if 6% of my paycheck is automatically being withheld. And by the way, those retirement funds have to be invested in my competitor, who I’ll call Walmart. So I’m actually sowing the seeds of my own destruction. Many of the phenomena we see in America today — rising inequality, a huge bias between large corporations and households, the difference between large and small business (and by small business I don’t mean the Russell 2000, I mean real small business) — can be traced back to regulatory choices that we’ve made. We are providing a subsidized cost of capital for major multinationals that then don’t pay taxes. This is absurd.

“We are providing a subsidized cost of capital for major multinationals that then don’t pay taxes. This is absurd.” — Mike Green

And Green ended his formal remarks with Voltaire: “Those who can make you believe absurdities can make you commit atrocities. It’s up to you to decide whether I’m the one sharing the bad information or whether what you’re seeing in markets is the bad information.”

SECTION THIRTEEN: The Q&A Gems

On the Historical Average Valuation — ‘Luck Favors the Prepared Mind’

Student: How do you prepare yourself for the future?

Mike Green: Go back to my survey. The survey results came in and the two curves crossed at almost exactly 16 times — and that’s literally the market’s historical average prior to around 2015. If you run the history of the Shiller PE, it comes back at 16 times. That number could have been 25x and I wouldn’t have been surprised. But because it actually came out at 16x, it spurred me to say: hey, wait a second, that number means something. The expression is ‘luck favors the prepared mind.’ I would actually suggest that being prepared is largely a function of just remembering stuff. When I see 16 times, I think: that’s exactly the market’s historical average. There might be something interesting here. Nobody ever had an explanation for why it was 16 times. Turns out that’s actually just the result of how market participants discounted.

On Whether the Market Is Too Big to Fail

Student: Do you think right now the market is just too big to fail because so many people are putting their retirement money into these indices?

Mike Green: Unfortunately, that is one of the implications. Part of what we’re seeing in the buy-the-dip mentality is a reinforcement of the mechanical properties of money flowing into a higher-multiplier investor. Eventually volatility rises — as we’re starting to see — and that lowers the relative attractiveness of even that higher return and people start to exit. That creates an endogenous feature to likely withdrawals. The single scariest statistic that came out of the pandemic for me was a self-congratulatory note that Vanguard wrote on April 3, 2020, in which they said less than 1% of their customers tried to sell. And my reaction to that is: oh my God, imagine if it had been 2%. We have people extraordinarily well trained not to sell.

Mike Green: A retirement system should be built on income replacement. Most people don’t actually care what their retirement assets are. They just care: “Do I have enough money to service my bills, to eat, and occasionally buy a nice outfit for my grandchildren?” That’s what I care about. But because of the way we built our retirement system, every individual has to self-insure for a retirement of undetermined length. That means they have to accumulate a ton of assets that they are hopeful will last them through their retirement.

Mike Green: You want to know why young people are struggling? It’s a simple accounting identity. The spending of the old becomes the income of the young. They are spending less because they are uncertain as to how long they’re going to need those assets and what the assets are actually going to make. So they are underspending relative to their asset levels, which is exactly why many of you look at your Boomer parents like: ‘you guys are such jerks, you don’t understand what it’s like to be a kid these days.’ But the simple reality is we built a retirement system that didn’t exist for their parents. It’s an asset-hoarding system where the income is uncertain, and as a result there’s less income being spent on young people than we should have.

“You want to know why young people are struggling? It’s a simple accounting identity. The spending of the old becomes the income of the young. They are spending less because they are uncertain as to how long they are going to need those assets.” — Mike Green

Mike Green: Why do we treat investment in the S&P 500 as more favorable than you saving to invest in your own business? The retirees could have sold their business, accumulated other assets, bought an annuity. There are any number of things that exist. But the way we’ve built the system, it rewards the hoarding of assets.

On How It Ends

Student: Is it worth the sacrifice of having lower returns for retirees to stop this?

Mike Green: Unfortunately, that is my conclusion. There’s no way you’re not going to stop this. The question is just: does it end in a crash, in which case we all suddenly have to adjust our behaviors — which is what I think is going to happen — or do we slowly deflate it, recognizing that what we really need to do is convert that asset value into a stable income level so that we can increase consumption.

On Personal Investing

Student: Knowing all of this, how would you act as someone who is 22 years old investing their money?

Mike Green: The single best risk trade you can do: when you get your job and your employer matches your contribution, put it straight into fixed income. You’re doubling your return through their matching and you have no risk associated with it. Then it becomes a question of what do you actually think the forward expected return is going to be for other asset classes. Do I think they’re very favorable for equities? No, I don’t. I think they’re very favorable for fixed income because of the neglect that I showed in the chart before. For equities they actually look quite terrible, in part because we’re approaching an inflection point on contributions as the boomers move into retirement and begin to liquidate these assets. Those net flows actually create the conditions of the decline that I’m worried about. That’s certainly what the math suggests.

Shiva Rajgopal: You developed a framework in 2017 for predicting the effects of passive growth. What does it say?

Mike Green: This is a slide I wrote back in 2017. It hasn’t changed since, which is pretty remarkable. But as we’ve gone through this process, each one of these gets more and more support from the academics, and so unfortunately I feel that we’re getting more right. Things that we should be seeing: an increase in the correlation between securities; an increase in valuation of securities regardless of fundamentals as passive share grows; reduced market elasticity; an increase in market concentration as the momentum bias leads to the largest companies becoming larger; paradoxically, a reduced ability for new companies to become public even as markets hit all-time highs; and portfolio effects dominating cash flow and discounting effects.

Mike Green: We’re now seeing multi-trillion dollar market cap companies rise and fall by $500 billion in a single session. It’s insane. I understand you’re thinking that’s on a percentage basis not very big, but think about the implications for the cash flows that are supposedly embedded in that. It makes no sense whatsoever.

SECTION FIVE: The Correlation Catastrophe — The Day Homestake Mining Saved the World

Shiva Rajgopal: Tell us about the correlation story, because your chart on this is one of the most striking things you present.

Mike Green: Correlation itself is a devilishly tricky exercise — you need a 500-by-500 covariance matrix, all the instantaneous weights of the index, and so on. Or you can do what I did to get through Wharton and cheat! Which is to take advantage of a property called Kendall’s Tau, or binomial correlation, which simplifies the entire process much like the Rockettes down at Rockefeller Center. They’re not identical, but they move their legs at the same time. So binomial correlation simply says: did two stocks rise on the same day? If this happens over and over again, they’re going to be correlated. That’s a much easier calculation, and it tends to track really well. What you can actually see here is the rising co-movement. The fraction of the S&P 500 that’s moving in the same direction every single day has risen dramatically.

Mike Green: I debated the BlackRock chief strategist at one point using this slide. He said: ‘Look, correlations were just as high in the 1930s. There was no passive investing then. You can’t draw an analogy.’ And my reaction was: do you realize you’re pointing to a spot on the chart where the market was down 85%? And everything was incredibly correlated because it was the Great Depression and nobody knew if anything was going to survive. We’re at all-time highs and we’re experiencing levels of correlation that are absolutely absurd.

Mike Green: There are a couple of really important regulatory dates here. The Investment Company Act of 1940 did something really important. It took the closed-end funds and unit trusts from the 1920s that had facilitated a levered coordinated rise of US equities, and forced them to disclose any leverage they use. In the aftermath of the 1930s, that naturally meant mutual funds would advertise themselves as unlevered. Anyone here trade crypto? What happens in crypto when people use leverage? You get coordinated waterfall events. Those are correlation events created by leverage in portfolios. By taking out that leverage, we introduced a time period of basically very little forced selling for an extended period of time. We ended that in 1994, as I already mentioned. Then in 2003 we changed the indices to be float-adjusted market cap weighted, which bought us a little capacity.

Mike Green: And then on February 5, 2018 — the Volmageddon event — something happened that had literally never happened before. Every single stock in the S&P 500 moved in the same direction. That didn’t even happen in 1987. In 1987, Homestake Mining was actually up. This was a complete control of the market by index forces. In 2018 we saw it for the first time. In the fourth quarter of 2018 we then saw it three times. First quarter of 2020 we saw it five times. We’ve been in a relatively modest environment since, but we’re still printing the highest levels in history.

“On February 5, 2018, something happened that had literally never happened before. Every single stock in the S&P 500 moved in the same direction. That didn’t even happen in 1987. In 1987, Homestake Mining was actually up.” — Mike Green

SECTION SIX: The Survey, the 100% Marginal Propensity, and Why Valuation Became a Positive Selection Criterion

Shiva Rajgopal: You did something unusual for an investor — you ran a formal survey of portfolio managers. What did you find?

Mike Green: I was a little disappointed that there hadn’t been a bunch of research on how active managers actually behave. We generally had the idea of downward-sloping demand curves, so I went out and surveyed investors. I asked a fairly straightforward question: you’re a portfolio manager, you have 5% cash in your portfolio — so cash is not a constraint in either direction — you receive a new inflow or a redemption request. What is your marginal propensity to buy or sell, given valuation? The results are somewhat unsurprising. As valuations rise, you become more willing to hold cash — downward-sloping demand curves. As valuations rise, your marginal propensity to sell rises. Cash should be more attractive on a forward-looking basis if valuations are really high. What was surprising was the intersection of the two lines. They crossed at almost exactly 50/50, and remember this is just portfolio managers surveying other portfolio managers, not asking about the Federal Reserve or interest rates. And yet the lines crossed at almost exactly 50/50 at exactly the market’s historical valuation average of 16X earnings. Why? Because it turns out that’s why the market worked the way it worked. People discounted.

Mike Green: I then built an agent-based simulation. I took 10,000 agents, fed them the responses from the 450 different surveys, and somewhat randomly gave them cash and took cash away. And what emerges is a mean-reverting market. Valuations get higher, you become less willing to buy, more willing to sell. Valuations get lower, you become more willing to buy, less willing to sell. That causes mean reversion. The Shiller PE oscillating around its historical mean of 16X was simply a byproduct of market participants who discounted.

Mike Green: When you introduce a passive investor, they have zero price elasticity. They have to own these things. Valuation doesn’t matter. In fact, valuation becomes a positive selection criterion. The higher valued it is, the more of it I am going to buy on a relative basis. So what that does is shift the market away from mean reversion and toward an environment characterized by that marginal propensity of 100%: Did you give me cash? If so, then buy. Did you ask for cash? If so, then sell. Did I bring valuation into it? No. And as passive gains share, valuations rise at an increasingly volatile pace. Unfortunately, this again perfectly matches the empirical data.

Mike Green: A lot of times people say: well, we should just buy small caps, they’re neglected. No. Small caps have risen in valuation roughly 5x. Don’t forget that the largest index product is the Vanguard Total Stock Market Index. We are rapidly moving into a world in which valuations are ridiculously high for almost everything.

“Valuation becomes a positive selection criterion. The higher valued it is, the more of it I am going to buy on a relative basis.” — Mike Green

SECTION SEVEN: The Multiplier Nobody Taught You — $1 Creates $17, and How America Now Invests with Negative Cash

Shiva Rajgopal: Walk us through the mechanics of how passive actually inflates prices.

Mike Green: Most people think the answer is one-for-one. If I go from 95% invested to 100% invested, markets need to rise 5%. The problem is it doesn’t work that way. Because in order to go to that 100% investment, the cash can’t change in value. If I buy and you sell, you now have my cash and you’re a market participant. So the only thing that can change is valuations.

Mike Green: Imagine there’s $1,000 total invested. It’s all held at active managers. They have $950 of equities and 5% in cash. I decide I want to try out this newfangled passive thing. I go to the market in its totality and say I want 1% of the market — $10 — and I’m going to give it to passive investors. Now the active managers have $950 in equity and $40 in cash, and I have $10 in cash. I give that to the passive managers. They now have $10 in cash. Equities are unchanged, cash unchanged in total. But nobody’s at their desired state. The active managers are overinvested. The passive manager is underinvested. Now a transaction has to occur. The active managers want to sell at $9.50 of equities, the passive managers want to buy $9.99 of equities — and that’s not a trade. A trade is where prices meet. So you actually have to solve it iteratively. What you discover is that simply moving 1% requires equities to rise by more than 1% in price. The elimination of active manager cash would actually cause something like a Shiller PE of 900. You simply can’t do it.

Mike Green: So we now have two forces: an inelastic buyer and an unwillingness to hold cash. It is a structural feature of the market. Why do people call cash trash? Because the world’s largest investors treat it as if it is. That’s built into their mandate. It doesn’t actually tell you anything about cash. It doesn’t tell you about inflation or real returns. It just tells you what the investment mandate is.

Mike Green: You guys are fortunate enough to be coming through this program today. Had you come through it seven years ago, you would have never heard that markets are inelastic. They are deeply inelastic. Gabaix and Koijen in their 2021 paper — the Inelastic Market Hypothesis — identified that the EMH, which assumes near-perfect elasticity, is not even remotely correct. The EMH assumes a dollar into the market, because every buyer has a seller, creates about 1 penny of additional market capitalization — basically the bid-ask spread. It turns out using their instrumented variable analysis that number works out to between $5 and $8. A 500-to-800-fold misspecification. Their paper uses an average from 1992 to 2019. My estimate of the current multiplier is somewhere in the neighborhood of 17. One dollar into the market is creating $17 in market capitalization. And for the very largest stocks — like Nvidia — we’re getting multipliers in excess of 100 to 1. One dollar creates $100 of market capitalization. That’s absurd. And it is the definition of a Ponzi, unfortunately.

“One dollar into the market is creating $17 in market capitalization. For stocks like Nvidia, the multiplier exceeds 100 to 1. That’s absurd. And it is the definition of a Ponzi, unfortunately.” — Mike Green

Mike Green: The world’s largest active fund continues to be the Fidelity Contra Fund. It’s about $150 billion in assets. It carries about $4.5 billion in cash — around 3%. The world’s largest fund is the Vanguard Total Market Index, which across its various incarnations is about $1.9 trillion. As of March 31, 2025, its cash balance was negative one billion dollars. Now, when you eventually go to speak with an allocator, you will discover that if you walk in and say ‘I’m running a $1.9 trillion strategy with negative $1 billion in cash,’ they will throw you out of their office for being irresponsible. But this is how America invests. There is no cash in the system. They faced redemptions associated with the tariff tantrum and instead of actually selling, they used a line of credit to meet redemptions.

Mike Green: Volatility on fundamental events like earnings announcements is rising dramatically, perfectly consistent with the theoretical estimates. Goldman Sachs presents this as an average. It’s not. This is a trend. It’s tied directly to the forces of passive.

Mike Green: Valentin Haddad’s work — and I work with Valentin fairly regularly now — is one of the most important papers out there. I would argue that Valentin and his co-authors actually don’t fully appreciate it. His paper went from looking at cross-asset elasticity to looking at the individual securities. This is where the meat of this all lies. The largest stocks would be less liquid, less elastic than the smallest stocks even with constant elasticities. Why? Because their trading volume does not keep pace with their market capitalization. Apple is give or take 100 times the size of Delta Airlines in market capitalization. It trades about 10 times as much. So its liquidity is less than its market cap on an adjusted basis.

Mike Green: But it turns out the empirical data is far more extreme than the theoretical line. And this is a byproduct of two features. First, lack of substitutability. Do I actually care if I’m replicating the S&P 500 whether I buy Delta Airlines or United Airlines? I really don’t because it’s 0.3% of the index. Do I care about Nvidia or Apple? Yes — I need to get those immediately or my tracking error is going to be too high. There is no substitutability for the largest companies. This is presented in log form — 12.5 is give or take $3 trillion. You can see that the elasticity of those stocks is basically zero. This is part of where we’re getting those multipliers of 100x for Nvidia.

Mike Green: On the flip side, the smallest companies have a secondary feature — discretionary managers are being redeemed. And because discretionary managers tend to focus on smaller and more value-oriented stocks, they’re actually providing more liquidity than you would expect given the net redemptions happening in the active manager space.

SECTION EIGHT: The SPAC Story, the IPO Drought, and the Endogenous Momentum Spiral

Shiva Rajgopal: You mentioned that the SEC has been intervening to facilitate passive. Give us an example.

Mike Green: In September of 2019, indices like the S&P 500 Large Cap Growth index exceeded concentration thresholds that would allow them to be marketed as diversified funds. That meant they were no longer eligible to be sold to many retail investors. Well, that would be a tragedy. You can’t allow that to happen. So the SEC stepped in again and said: concentration rules apply, but not to you. You’re just replicating the market.

Mike Green: For decades we’ve seen a phenomenon where even as markets rose to new all-time highs, the quantity of IPOs remained very muted. Venture firms couldn’t get stuff public. And then in 2020, SPACs suddenly showed up. As a hedge fund manager, you use SPACs all the time because they give you a little bit of volatility exposure and they allow you to hide how much cash you have in your portfolio. But SPACs suddenly exploded in 2020. The narratives we heard didn’t tell you the truth, which is this: in the methodology for the Vanguard Total Market Index, special purpose acquisition companies are not eligible for index inclusion. However — there was something called a fast-track IPO. A fast-track IPO occurred if a company listed at greater than the 85th percentile threshold. In 2020, that was about $1.5 billion.

Mike Green: What actually happened with SPACs was people tried doing SPACs of unusual size. Nobody had ever tried a SPAC at that scale. And then suddenly SPACs started showing up in the wild days of 2020 as mechanisms for matured venture properties to come public or for private equity entities to exit. And they triggered fast-track IPO status for the first time. The SPAC is not eligible for index inclusion, but if it qualifies by hitting that $1.5 billion threshold, it becomes eligible for fast-track IPO status, which requires index providers to buy their proportionate share in as few as 5 days. The rules of SPACs are that they can’t offer liquidity from insiders until prices have exceeded typically a 20% threshold for a minimum of 20 days. So the world’s largest buyers have to buy in 5 days. And the only sellers can’t sell for 20 days. What happens to prices? They go up a lot. Anyone remember Nikola? This is a battery-operated truck they pushed down the hill. Vanguard and BlackRock kept buying the entire time.

Mike Green: In 2023, they changed the rules and SPACs became no longer eligible for index inclusion under fast-track IPO status. We haven’t we seen a SPAC since. Private equity payouts have collapsed, venture payouts have collapsed. Why? Because they can’t get liquidity. We have a dysfunctional market from a fundraising standpoint. The only IPOs now actually being done are being led by firms like Coinbase and Robinhood who are pitching them to their own clients: ‘get on the inside track.’ NASDAQ has now actually tried to reintroduce this fast-track IPO status for SpaceX and OpenAI. They effectively are selling access to the index in order to get these companies public and get the listings associated with it.

Mike Green: Every single year I’m confronted with people sharing NASDAQ’s IPO Pulse index saying this is the year the IPOs are coming back. Markets are at all-time highs. It’s not happening. We are simultaneously hollowing out our capital system as we are celebrating its rise.

“We are simultaneously hollowing out our capital system as we are celebrating its rise.” — Mike Green

Mike Green: What if what we’re experiencing in today’s market is not information diffusion, but endogenous momentum? Cliff Asness in the late 1980s and early 1990s identified momentum as effectively information diffusion: I do work on a company, I discover their earnings prospects are better, I buy the security, that causes its price to rise, attracts additional attention, people do additional work and discover I was right, they buy, price continues to rise. That’s momentum as an information diffusion feature. But what if it’s actually something else? When a billion dollars flows into Vanguard, it buys $70 million worth of Apple. That causes Apple’s price to rise slightly. The next day that billion dollars has to buy $70.1 million of Apple, and $70.2 million, and $70.3 million — an endogenous price-weight spiral driven by passive investing.

Mike Green: The evidence of this is extraordinarily strong. I basically took market-neutral portfolios, determined how much flow was impacting each individual security, and ran high and low combinations in a traditional value approach across multiple market-neutral portfolios. The net result is a monthly return of about 1.4%, standard deviation well below the S&P. You’re outperforming the S&P. Your beta is effectively non-existent. Your alpha is basically 100% of your return. And this is just passive. What this means is that the S&P 500 is rising almost 18% a year simply because of the growth of passive. That’s more than the S&P’s return over the last 5 years. In other words, everything we’re seeing at this point appears to be passive.

A Note Before We Begin

I have spent my career studying financial reporting, capital markets, and what prices are supposed to tell us about the world. I was trained — as most finance and accounting scholars were — to believe that markets are reasonably efficient, that prices reflect fundamental value, that discounted cash flow analysis tells you something true, and that the cost of capital is a meaningful concept that disciplines corporate behavior.

After spending an afternoon listening to Mike Green — veteran portfolio manager, and Chief Strategist at Simplify Asset Management — I am no longer sure any of that is true.

Green has spent a decade arguing that the explosive growth of passive investing has not just distorted markets at the margins. It has broken the entire mechanism by which prices are supposed to signal anything. He is not fringe. He is not a crank. The academic literature — slowly, grudgingly — is catching up to what he was writing about ten years ago. And the implications, once you sit with them, are genuinely staggering for everyone: analysts, CFOs, boards, investor relations professionals, short sellers, corporate governance scholars, anyone who has a 401(k) and of course for policy makers.

He says many controversial things. View them in the in the spirit of hypotheses that need to be refuted or supported. Don’t dismiss these ideas outright because they are inconsistent with your worldview or your ideological stance.

What follows is a reconstructed interview drawn from Green’s recent talk in my class for Masters students at Columbia Business School. I have framed his arguments in the form of a conversation, added my own questions to give structure, and then extended the implications into eleven areas. I intend to present his arguments in parts, running into roughly 3000 words in each installment. Nothing here is put in Green’s mouth that he did not say — but the framing, the follow-through, and the 11 implications are mine. Buckle up.

PART ONE: I Love Capitalism. Unfortunately, I’ve Experienced Very Little of It

Shiva Rajgopal: Mike, before you walk us through the mechanics, I’d like to know what drives you to keep making this argument year after year, against considerable resistance.

Mike Green: I actually do this because I care and because I love markets. I actually love capitalism more than almost anyone I know. Unfortunately, I think I’ve experienced very little of it over the course of my career, and this is just the latest component of it. You can slave away at something for 10 years and then all of a sudden wake up one morning and be like, wait a second, this is how to use it. The academic literature is now coming out to support the stuff that I was writing about 10 years ago. We’re also in the process of an incredibly fruitful product development cycle tied to some of the insights that have become apparent to me in the last 6 to 9 months.

Mike Green: But the most important thing is to get educated and understand this, because the simple reality is this is the single most important subject that should be on the minds of every American. It’s tied to our entire retirement system and the reality is nobody understands it or even cares.

Shiva Rajgopal: How big has passive actually gotten?

Mike Green: First, the most important thing to understand: people tend to focus on the managed space — mutual funds or ETFs. ETFs have often just been code for passive investing. But this is kind of the tip of the iceberg. There’s also an entire industry around separately managed accounts and total return swaps and futures, and so on. When we do the full calculations, we come to the conclusion that about 55% of the US equity market by market capitalization is now held in passive strategies. I define passive strictly as market cap-weighted indices. When I started in the industry in the early 1990s — I graduated from Wharton in 1992 — the passive share was around 2%.

PART TWO: The 1990s Rewrite — Insider Ownership, Float, and 30 Years of Regulatory Choices

Shiva Rajgopal: Walk us through how we got from 2% to 55%.

Mike Green: In 1994, Vanguard began to experience what was called tracking error. They were growing and forced to try to buy every security in the index, which was causing their results to deviate from the benchmark. They went to the street and said: what can we do about this? The street said: use futures. They said: we can’t. Futures require margin accounts. Margin accounts are leverage. The 1940 Act prohibits leverage. What did the SEC actually do in 1994? They gave a no-action letter to index providers saying you, as a special group, are allowed to use leverage. That changed the character of the industry in a very material way. It went from portfolio replication being a problem of the portfolio manager to suddenly being a profit arbitrage opportunity for the street.

Mike Green: If you think about what actually happened in the 1990s, it’s really interesting. We used to weight our indices on the basis of market capitalization. Most people think they still are. You guys are in a finance program, so you know they’re actually adjusted market cap weighted. The reason why that was done was because in the 1990s when you had market cap-weighted indices, you also had relatively high market cap companies with relatively low float — like Microsoft, Cisco, Dell. When we switched to using futures to replicate, that meant futures brokers had to go recreate the index by buying up the individual components. That meant they had to buy up the market cap representation of Microsoft when only half the shares were actually outstanding.

Mike Green: Insider ownership became the single best predictor of return as a strategy in the mid-1990s. And as academics are incentivized to do, we constructed all sorts of narratives. I encourage you to go back and read the rich literature on insider ownership and management alignment from the 1990s. It was all bunk. It was a feature of the market. Nobody cares about insider ownership anymore. We truly actually don’t care. A paper just came out that actually highlights that the only source of liquidity for passive investing is by and large the selling of active managers facing redemptions and insiders selling shares. So when Jensen Huang sells shares in Nvidia, it’s celebrated. Why? Because it increases the market capitalization and shares available for Vanguard to buy.

Mike Green: The reason why I tell this long convoluted story is to say this is actually a problem that has been building for 30-plus years. It is a byproduct of misunderstanding of markets as being highly elastic, capable of absorbing almost anything we throw at them. The real story here is that we have abused markets to provide a service we call retirement. It’s not what they were designed to do. Nowhere in your weighted average cost of capital calculations did you say ‘I want to retire and go to Disney World.’ The simple reality is that markets exist to set the marginal cost of capital to drive investment, and we’ve lost sight of that.

Mike Green: John Bogle himself recognized that if everybody indexed, the only words we could use were chaos, catastrophe — the markets would fail. The problem was John never did any research on it. He simply was asserting a truism from the singularity that arises in the Grossman-Stiglitz paradox. But more importantly, he didn’t consider the dynamics of the difference between stock and flow. Most of the analysis done around what share passive could get to, relied on stock analysis — meaning what share was passive. So: let’s run an experiment with agents at 10% passive, 20%, 40%, 50%, 60%, 70. And if we exclude flows, somewhere around 80% we encounter a singularity in which there’s simply not enough capital to maintain continuous markets. I actually just released a paper co-authored with Hari Krishnan and Stephan Sturm on exactly this topic. (https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6438678)

Mike Green: In 2015, something really important happened. Not only did passive continue to gain share, but active flipped to losing assets. This was a byproduct of the Pension Protection Act of 2006 that changed our retirement system from an opt-in system — in which you had to choose to save — to an opt-out system in which you had to choose not to save. They established what’s called a qualified default investment alternative. That QDIA was designated as passive investments, particularly target date funds. 2012 it switched specifically to target date funds. That meant there was no more money flowing into the active manager space because it was not designated under the retirement system that roughly 95 cents of every retirement dollar now flows through. So we actually created conditions under which the flow of money from the active community — these are the people who are supposed to be doing the DCF — goes net negative. In physics terms, the loading factor has now become negative. The value signal gets a negative coefficient on it.

“We actually created conditions under which the flow of money from the active community — these are the people who are supposed to be doing the DCF — goes net negative. In physics terms, the loading factor has now become negative. The value signal gets a negative coefficient on it.” — Mike Green

Mike Green: And it was at this point that people started to notice what they called the Costanza market. You guys are too young to remember the Seinfeld episode where perennial loser George Costanza recognizes that every decision he’s made in his life is wrong. So, if he simply does the opposite, it should work out, and it works out so well he can’t handle the success. We began to refer to the market in 2016 as the Costanza market because it literally was doing things that made no sense whatsoever to seasoned investors. The seasoned, thoughtful investors who are trained on discounted cash flows are being net redeemed. The portfolios that they’re lovingly crafting and putting together that have high expected returns are experiencing outflows. And if markets are inelastic — and we know they are inelastic now — that means you’re getting a negative loading on the value factor and the markets get crazier and crazier and the experts look dumber and dumber and dumber. So nobody wants to listen to them.

Mike Green: JP Morgan did an analysis looking at the percentage of trading that involved any sort of fundamental analysis. You want to know why nobody cares about PhDs from the Columbia Finance or Accounting Program? Because there’s no fundamental analysis being done. Everything is off of price now. Everything is off of arbitrage. It doesn’t matter what the price of the S&P is. It matters: can I replicate the S&P through the individual securities and do so profitably so that I can broker an exchange into an ETF.

Mike Green: Back in 1995, about 80% of trading activity was coming through the discretionary managers. Only about 10% was actually happening on the retail side, partly because of relatively high commissions. Today that is completely flipped. As of 2022, we estimate that only about 10% of trading activity involves a discretionary asset manager, and some subsegment of those are actually then involved in fundamental trading activity. We actually now think this has fallen to around 7%. Speculative trading — the Robin Hoods of the world, who now have free commissions, free option trading — has exploded to around 21% of the market. In other words, they’re about three times the size of active managers. So when you see something like the GameStop saga, it’s actually true. Coordinated effort from retail individuals can cause behaviors in markets that professional managers could never have imagined occurring. The noise traders are dominant.

Mike Green: You want to know what the single most valuable piece of research information on the street is? Last price tag was $4.5 billion for 2025. A single research piece. It’s order flow — payment for order flow. That’s why Robin Hood exists. Citadel writes them a check for about $4 billion every year. You have a Robin Hood account. Let’s just say Citadel knows everything you’re doing.

Part 2 comes on Wednesday.

Instead of the usual CFO elective fare — a crisp accounting case on Enron, maybe a chapter on ASC 606, a few earnest slides about disclosure controls — I asked four cohorts of finance executives to spend an hour with Antigone. The play Sophocles wrote in Athens in 441 BCE. The one about a young woman who buries her brother in defiance of the king’s decree, and pays for it with her life.

I was not confident it would work. CFOs are a pragmatic tribe. They measure things. They certify things. They have 10-Ks to file and audit committees to brief and covenants to watch. Asking them to read a Greek tragedy felt, on paper, like the sort of thing that sounds clever in a faculty office and falls flat the moment you open your mouth on Zoom.

It did not fall flat. It did something else — something I am still thinking about days later — which is why I am writing this.

The Setup

The case I put in front of the group was mundane on purpose. You are the CFO of a publicly traded SaaS company. Your team has found a revenue-recognition question in Q4. If you take an aggressive reading of ASC 606, $18 million of revenue that really belongs in Q1 can be pulled into Q4 — just enough to hit guidance. Outside counsel says the aggressive reading is defensible. The head of sales points out that every competitor books it this way. The CEO is immovable on hitting the number. Everyone in the room has signed off. You have 48 hours before filing. You are the only person who is uneasy.

Then I asked a question that only makes sense if you know the play: In this scenario, is the CFO Antigone, or Creon?

Antigone is the one who defies institutional authority in the name of a higher law — and gets buried alive in a cave for her trouble. Creon is the king who issues the lawful decree, who follows proper procedure, who has the rule book on his side, and who loses his son, his wife, and his moral standing by the end of the play. One does the right thing. The other does the compliant thing. They are both destroyed, but not by the same thing.

The answers poured into the chat. Some said Antigone. Some said Creon. A great many said both, depending on what she decides to do. One participant, unprompted, had written “Antigone” in the chat before I even revealed where we were going. Another wrote something that stuck with me:

“The CFO is Antigone in the 48 hours she feels the unease, and becomes Creon at the moment she signs anyway. Same person, same week, different choice winning.”

That is the sentence I am going to remember from this session. It collapses the whole play into one line of finance prose.

The Character Map

Before opening the floor for stories, I walked the group through the full map, because the real usefulness of the play is not the binary of hero versus tyrant. It is the supporting cast.

Antigone is the CFO who calls the audit-committee chair before signing. Who puts the concern in writing. Who pays the career price — gets labeled difficult, gets dropped from the short list, gets the quiet phone call from a headhunter that says “word is you’re a stickler.”

Creon is the CFO who says the lawyers told me this was fine. Who leans on procedure. Who has the memos and the sign-offs and the precedent and the auditor’s letter. Who is not wrong within his own frame. That is the terrifying thing about Creon — he is never technically wrong.

Ismene, Antigone’s sister, is the character I find most recognizable and most uncomfortable. She agrees with Antigone in private. She refuses to help in public. She rationalizes — we are only women, we cannot fight the king, someone has to survive — and then lives. Ismene is the colleague who pulls you aside after the meeting and says “I wanted to say something too.” Ismene is the majority of every boardroom I have ever sat in. Ismene is, on most days, me.

Haemon is the CFO whose written objection gets documented and then overridden. I raised it. Yes — but was raising it enough?

Tiresias, the blind prophet, is the auditor who shows up too late with the qualified opinion, the advisor who was going to say something eventually, the board member whose post-mortem wisdom arrives in time for the restatement but not in time to prevent it.

And the Chorus — the old men of Thebes who narrate, who fret, who never act — is the disclosure committee. The audit committee that receives the briefing, nods, asks no follow-ups, and moves to the next agenda item.

When I laid this out, the Zoom fell quiet in the way Zoom rarely does. Not the silence of confusion. The silence of recognition.

The Stories

I had not planned for the stories to come the way they did. I asked, half-apologetically, whether anyone was willing to share a moment they had been in one of these roles. I told them not to name companies. I told them Chatham House rules applied. And then I waited, because I have been teaching long enough to know that the first person to speak determines whether you get ten more or none.

The first person to speak was a veteran executive who had spent a career in consumer packaged goods. He talked about channel-stuffing pressure — the quiet, ritualized quarterly exercise of pulling shipments forward to make a number — but then took the conversation somewhere I did not expect. Sometimes, he said, the CFO is the one championing it. Not resisting the pressure. Producing it. And there was a strategic wrinkle I had never heard articulated quite that way: if your product is occupying the biggest share of a retailer’s warehouse, the retailer becomes your partner in clearing it. You want to be their biggest inventory problem. The books close. The commercial relationship deepens. Nobody lies. Everything is technically fine.

He was describing Creon’s Thebes — a polis where everyone is behaving rationally within the rules, and something essential is quietly going wrong.

Next came a technical accounting leader who had run SEC reporting through the transition from ASC 605 to ASC 606. His story was about a company whose revenue-recognition practice — recognize at shipment, not delivery — was going to need to change by hundreds of millions of dollars over a rolling window. The CFO he reported to outright rejected the implication in the first conversation. It took six months of patient education — not confrontation; education — before the organization accepted what the new standard required. He sat in the boardroom explaining it. He watched the division CFOs call him to say this is BS and we don’t like it. He thought about resigning. He stayed, and the company eventually did the right thing. “You feel you’re alone a lot of times,” he said. “But it happened.”

Then he told a second story, from a different company — a very large technology firm moving from one reportable segment to two. And here the gray turned darker. Certain information, under the segment-disclosure rules, would have to be reported externally if it continued to flow to the CEO. So the company simply stopped sending that information to the CEO. Plausible deniability, engineered. The CFO of the business was frustrated because she genuinely wanted to explain the drivers of her business to her CEO. She could not. Not if they wished to avoid the disclosure. He held the line. The career cost was real. “You take that bullet for the compliance,” he said.

He mentioned, almost in passing, that he had been through SEC and DOJ investigations in multiple roles. All I can share, he said, is that it is not the right or wrong answer that matters to regulators. It is whether you can show what processes ran, and what was transparently disclosed internally along the way.

Then came a voice I had not expected to speak — a finance leader based in Zurich who had just typed, minutes earlier, that she had lived this exact case. I asked if she was willing. She was. She had been asked by her CEO to report incorrect revenue. She pushed back. Professionally, she emphasized — I just said, we shouldn’t do that morally, and I don’t feel comfortable signing off. The CEO became hostile. The conversation ran for the better part of a week. “I was put in a position where I had no choice but to quit.”

She paused, and then added something that I think is the single most important sentence anyone said in the 66 minutes of the session:

“From the moment I pushed back, the culture had already changed. The way they looked at me had already changed. I was already a stickler. I already had a bad reputation. Culturally, it was not the best place for me to stay.”

That is the hidden cost the governance textbooks never quite capture. You do not pay the price for resigning. You pay it the instant you hesitate to sign. The room updates its model of you immediately. You may keep the job. You will not be the same person inside it.

The chat lit up. Brave. Thank you for sharing. That is exactly what happened to me. A CFO based in Singapore wrote in to say same thing happened here — after refusing to sign I left the company. A colleague from Mumbai wrote in that he had left a job over something similar and it was traumatic for a while. A participant in Dublin wrote that he had had an identical experience and left later, so it would not be correlated. This is an important detail. The playbook, for those who survive these moments, is not to quit on the spot. It is to quit later, when the graph is less legible.

An FP&A leader from Toronto took us into a different corner of the same problem. Not revenue recognition — an accrual reversal, timed and sized precisely to deliver a 200% bonus payout for everyone. She raised it to the controller, to the CFO, and to the CEO directly. The number went in. The bonuses paid out. Everyone in compliance. Everyone aligned. “Sometimes,” she said quietly, “we still have the responsibility of acting as Antigone, even knowing the consequences.” She is still thinking, she admitted, about what she should have done differently.

A CFO based in the Pacific Northwest — who had, remarkably, begun her career at Arthur Andersen in the months of the indictment — offered the most practical piece of wisdom of the session. She calls it her walk-away fund. Not public-years savings. A distinct pot of money, built deliberately through every role she has ever held, that existed for exactly one purpose: to let her leave any job at any moment if she was asked to compromise her values. She said it plainly: “You always have to do what is right for you to be able to sleep at night.” Several participants later said in the chat that they were going to build one.

Then came the story I had not known to expect. A CFO operating in an emerging market who had recently discovered that over 5% of the costs he was tracking came from invoices with no economic substance — internal fraud, in other words. He raised it to the family office that owned the business. The response was: we are selling the company soon, it is better not to surface this. He spoke quietly. “Honestly, I would rather quit than stay.”

And finally, with five minutes left in the hour, a CFO based in São Paulo who had worked at a major credit-ratings agency through the 2008 financial crisis — her CEO had testified before Congress — closed the session with the clearest statement of principle I heard all day. She has been asked to sign things she did not agree with, at two different firms. The first time, she went to the audit committee, got the filing delayed, got the numbers restated, and quit eighteen months later when the surrounding culture had not changed. The second time, she quit on the spot, without another job lined up.

She earned a nickname at one of those firms. The problem child. She was entirely at peace with it.

“A salary is a salary. Character and reputation last longer than a position.”

What the Poll Said

Midway through the hour I ran a live poll on a wordier version of the original question — given everything we had just discussed, what would you actually do in the CFO’s seat?

About 45% said they would escalate to the audit committee — the Antigone answer. About 54% said they would seek additional outside counsel before deciding — the procedurally safe answer, which is to say the Creon answer in new clothing. A tiny residual chose simply to sign.

I do not love the distribution. I also do not judge it. This was a group of seventy-odd working CFOs and senior finance leaders, most of them with mortgages and tuition bills and ageing parents and headhunters who have long memories. The fact that roughly half of them were, on a Monday morning with no stakes whatsoever, prepared to be Antigone is a higher number than I would have predicted in the privacy of my own head.

What the poll really showed was that there is no consensus answer. Humanity has been arguing about this since 441 BCE. If the question had a clean answer, Sophocles would not have made a play out of it. He would have published a memo.

What I Am Taking Away

Three things keep circling back.

The first is that the technically compliant frame and the materially right frame are not adjacent categories. They sit at different altitudes. A great deal of real-world governance failure happens in the space where every individual actor is technically compliant and the collective result is materially wrong. Andy Fastow, whom I have interviewed, has been consistent about this — his defense has always been that every disclosure was made, every sign-off obtained, every committee briefed. He was not wrong about the procedure. He was wrong about the thing the procedure was meant to prevent. Creon would have understood him completely.

The second is that character precedes decision. The CFOs in the room who had refused to sign were not making a decision in the moment. They had made the decision years earlier, quietly, by accumulating a walk-away fund, or by marrying someone whose career could absorb a gap, or by keeping their children’s private-school tuition at a level they could cover without the bonus, or by some other slow and unglamorous act of architecture that made the eventual no possible. You cannot generate Antigone-level courage on 48 hours’ notice. You can only discover, at the 48-hour mark, whether you have been preparing to be Antigone all along.

The third is that Ismene — the sister who stays silent, who privately agrees, who survives — is the most common role in every institution and almost never the one that gets discussed in leadership development. We train CFOs on how to be Antigone. We do not train them to notice, in themselves, the precise moment they become Ismene. That moment is quieter than the Antigone moment. It is usually rationalized as practicality. The filing has to go out. The auditors signed. The lawyers cleared it. Someone has to hold the function together. Every one of those sentences is defensible. Every one of them is also how institutional failure sounds from the inside.

A Closing Note

I ended the session by pointing out, mostly to myself, that a play written twenty-five centuries ago had done something no accounting case had ever managed to do in any class I had taught. It had gotten a roomful of experienced finance executives to volunteer, on the record, that they had resigned rather than sign. That is not a literary observation. That is an argument for the liberal arts in the training of people who sign 10-Ks.

AI is very good at the how. It will tell you, in seconds, exactly how ASC 606 applies to your bundled contract. It will draft your disclosure language. It will spot inconsistencies in your footnotes.

It is not yet very good at the why. It is not good at the moment in the corridor outside the boardroom when you have to decide who you are. For that, strangely, we still need the Greeks.

Go read Antigone. It takes about two hours. You will see yourself in it somewhere. If you are lucky, you will see yourself in Antigone. If you are honest, you will see yourself in Ismene. Either way, you will not look at your next signature the same way.

Companies buying high-growth, “move fast and break things” targets often discover that internal audit can create immediate friction with the very entrepreneurial culture that justified the deal. Dominique’s approach is to refuse to accept that framing in the first place.

> “This is fundamentally a risk appetite question, and the key is alignment before any audit begins. You should be very explicit with management: for this acquired entity, during this defined period, the risk appetite is intentionally different from the core. Internal audit can and should still look at the entity — but against the risk appetite and the standards that have actually been agreed, not against a group standard that has not yet been adopted.”

Then she said something that I think most boards never hear from their auditors:

> “An M&A has a value creation thesis. Internal audit should be verifying that the things that are supposed to create value are actually happening, not just cataloguing what could break. Risk is not inherently negative. Risk is an event. What matters is how you manage it. An internal auditor who only thinks about downside risk is going to undermine the very rationale for an acquisition.”

Three months into most integrations, there is a list of everything that could go wrong. There is rarely a list of whether the acquirer is actually doing the things that were supposed to make the deal worth the price. That is a huge, expensive, and largely unaddressed audit question.

---

Auditing AI: what it actually looked like on the ground

Uber has been using machine learning from early in its history — it is embedded in how the app works — so Dominique’s team had developed methodologies for auditing ML models before the current generative-AI wave. By the end of 2025, they were doing the equivalent for GenAI. She was blunt about what the team looked like:

> “We had engineers embedded in the audit team who could actually go into the models — examine whether models were appropriately tuned, whether they were being fed the right data, whether they were secure, whether they were hallucinating. We could go to management and report that specific models were not calibrated well enough to create the best value while serving stakeholder interests.”

That is a sentence no team of accountants can say. It is what she meant in Part 1 when she said the team must mirror the business.

But the most important thing she said about AI was not technical. It was about fiduciary responsibility.

> “Governance has always been about decision making. When the technology is making profound decisions — decisions that used to be made by management — what does that mean for fiduciary responsibility? Whether it is a human decision augmented by GenAI, or a fully automated AI decision, the board needs to ask how it exercises oversight over a decision-making process that no longer looks like what it used to.”

Then the diagnosis every board should sit with:

> “The speed of adoption is the biggest risk right now. Organizations are deploying these tools with extraordinary trust and very few mechanisms to stay in control. That gap between adoption speed and governance maturity is where boards and CAEs need to focus urgently.”

---

For the CFO without an internal audit function: four triggers and one path

A good portion of our conversation was for the CFO who does not yet have an internal audit function — who is at a mid-sized private company, who is not obligated by any rule to establish one, and who has to make the case to leadership.

Dominique’s advice starts with the uncomfortable truth that nobody is forcing the conversation:

> “In a private company — and in many public companies depending on where they are listed — nobody is forcing you to have internal audit. You have to want it. So your first task is to create that desire.”

The way to create it, she argues, is not to propose a permanent function. It is to commission a single, externally provided audit on one area where the low-hanging fruit is visible. Done well, it opens the door. The CFO builds from there — perhaps adding one internal lead who manages the mandate before scaling to a team.

When should the CFO commission an engagement at all? She gave me four triggers, and they are worth writing down:

1. Pure reassurance. Something matters so much — business continuity or legal survival — that you need belts and suspenders. You know it is working, but you need to *know* with certainty.

2. New ventures. You are going somewhere the organization has no prior experience, and you want someone helping you think about risk as you enter that territory.

3. Endemic dysfunction. Everyone knows there is a persistent problem — a process that has been broken for years, that people have complained about repeatedly, that nobody seems able to fix. Internal audit can put everything on the table and get to the root cause.

4. Scale. The company has grown to the point where leadership can no longer directly observe everything, and the organization is running on trust. You want an independent mechanism to verify that the trust is well placed.

She closed this part of the conversation with an observation that should unsettle every publicly traded company:

> “The best internal audit I have seen often exists inside private companies — because management chose to have it. When it’s wanted, it works. In publicly traded companies, especially in the US, internal audit too often starts as an obligation, and the CAE is left to prove relevance against a headwind.”

---

Five things I took away, for CFOs and boards

If I had to compress the two conversations into an action list, it would be this:

1. Recalibrate the risk portfolio. If 80 percent of value destruction is strategic and 15 percent is operational, make sure your audit plan reflects those proportions. If your function is 70 percent SOX and 30 percent everything else, you are auditing the tail.

2. Protect functional independence, regardless of the administrative line. Whoever the CAE reports to, the access must be enterprise-wide, and the CFO must signal that clearly to the CTO, the CRO, and the rest of the C-suite.

3. Build a team that mirrors the business. Engineers, operations specialists, product people — proportional to how much those functions drive the company’s actual value. A homogeneous team of accountants cannot audit a technology company.

4. Align on risk appetite before every audit. What you are auditing against — the standards, the thresholds, the tolerances — is as important as the audit itself. Especially in post-acquisition integration and new market entry.

5. Treat AI governance as a board-level imperative. The gap between the speed of AI adoption and the maturity of AI governance is the single most urgent area for boards and CAEs to close. It is, at its core, a question about fiduciary responsibility when decisions no longer look like they used to.

Dominique recommends that CFOs and board members read Domain 1 (Purpose) and Domain 3 (Governing the Internal Audit Function) of the Global Internal Audit Standards, effective January 1, 2025, from the IIA. The standards are at theiia.org. They are not long, and they are what the profession says the function is for.

This concludes the two-part conversation with Dominique Vincenti. Thanks for reading. If you found it useful, the best thing you can do is forward it to a CFO or board member who files internal audit under “SOX compliance.”

I believe internal audit, as a profession, has a strange structural problem. Almost every CFO and board member I meet has a rough mental model of what internal audit does — and for most of them, the rough model is wrong. Not catastrophically wrong. Just wrong in ways that quietly cost their organizations enormous amounts of value.

Dominique Vincenti is one of the world’s foremost authorities on the function. Thirty-five years. Chief Audit Executive (CAE) at Uber through its IPO, before that at Nordstrom. A seat at the table in 2003 as the profession negotiated with the SEC’s Chief Accountant over how Sarbanes–Oxley would actually be executed. She led the drafting of the purpose and governance sections in the 2025 Global Internal Audit Standards — the first comprehensive rewrite in decades.

What follows is the first half of our conversation drawn from her talk to my CFO class, organized around the argument I found most striking: that the boundary of what internal audit is *for* has been badly misdrawn at most companies, and correcting it is the single highest-leverage governance move available to a CFO.

---

The frame: what is a CFO actually doing?

Dominique opened not with internal audit but with the CFO’s role, because she wanted to set the stakes before we got to the technical question.

“You are a member of the executive team. You participate in setting the overall strategic vision, direction, and culture of your organization. Specifically, you translate that vision into financial reality — fiscal health, resource allocation, risk management, data-driven insights. You are the value architect, balancing short-term performance with long-term investment and operational efficiency.”

Internal audit’s purpose, as articulated in the 2025 standards, is to strengthen an organization’s ability to create, protect, and sustain value — by providing the board and management with independent, risk-based, and objective assurance, advice, insight, and foresight. Same north star as the CFO. Value.

“Internal audit should offer management a safe, unfiltered mirror — an objective view of whether the company is really making good choices for all its stakeholders. That’s its function. And its scope is governance and risk management as a whole — which is much larger than most people assume.”

That phrase — “much larger than most people assume” — is the hinge of the whole conversation. Because the data on how most people use the function suggests the scope has been dramatically under-claimed.

The 80 / 15 / 5 problem

Dominique told us that Gartner, with Booz and Company, has been tracking the root causes of enterprise value destruction among Fortune 100 companies on rolling ten-year periods since the late 1990s. What the numbers show, consistently across decades, is something like this:

- 80 percent of lost enterprise value traces to mismanagement of strategic risks.
- 15 percent to mismanagement of operational risks.
- 3–4 percent to regulatory and compliance issues.
- 1–2 percent to accounting or financial reporting.

Even in decades that include a major debacle like Enron or WorldCom, the accounting share rises only to about 5 percent.

Dominique’s framing of what this means is unusually sharp:

“Governance and risk management is a vast territory. Accounting and financial reporting absolutely matters, but it is not the lion’s share of what can go wrong with your organization. If your internal audit function is primarily a financial-controls shop, it is focused on the tail, not the trunk.”

Read that twice. The function that most directors and CFOs mentally file under “the people who check our SOX controls” is, by the profession’s own standards and by the empirical record of what actually destroys companies, meant to be pointed at the 95 percent — not the 1 to 2.

---

How SOX hijacked the profession — and why Dominique tried to stop it in 2003

The story of how we ended up here is instructive. Dominique was in the room when the profession negotiated with the SEC’s Chief Accountant over how Sarbanes–Oxley would be executed. She and her colleagues could already see what was going to happen.

“I told the Chief Accountant directly: financial reporting is the thermometer the doctor uses to take the patient’s temperature. If all your energy goes into making that thermometer perfectly accurate, you will eventually have a patient dying with a perfect temperature reading. And that is exactly what happened in 2008. The financial crisis had nothing to do with financial reporting.”

The law itself, she points out, is clear on the division of labor. Sections 302 and 404 don’t say internal audit should own financial controls. They say it is the responsibility of “management.” Internal control over financial reporting belongs to the controller. Internal audit’s mandate is governance and risk management as a whole.

“Any CFO who thinks their audit function is primarily there to certify financial controls is significantly underusing it.”

---

Where the strategy-consulting line should be drawn

If internal audit’s scope is this broad, the natural next question is whether it should be pulled into strategy itself. Dominique is adamant about the boundary, and precise about where it sits.

Setting strategy is the executive team’s job. Overseeing strategy is the board’s. That is not negotiable. But the processes through which strategy is set — are they robust? Are the right inputs considered? Are risks to execution identified? — those are legitimate audit territory. And the execution of strategy — is management actually designing and implementing the processes required to deliver on what was decided? — that is where internal audit adds the most value.

“An internal auditor who only thinks about downside risk is going to undermine the very rationale for an acquisition.”

She’ll return to that point more forcefully in Part 2, when we get to M&A.

---

The reporting line that quietly determines everything

Perhaps the most practically consequential part of our conversation was about reporting lines — because the administrative choice a CFO makes here signals to the entire organization whose function this really is.

In the US, the Chief Audit Executive most often reports administratively to the CFO. Dominique’s view: that is fine, as long as it stays “administrative.” The functional independence of the CAE — full access to every member of the executive team, to the board, to any part of the business — must be actively protected.

“I have experienced situations where, even with a superb CFO who understood the mandate, the CTO looked at me and said: ‘Why are you here? You do finance.’ That is a failure of positioning that the CFO can and should prevent.”

And there is a long-term trend worth noting: globally, the percentage of CAEs reporting directly to the CEO (rather than to the CFO) rises year on year. Dominique reads that as a growing recognition that internal audit’s scope extends well beyond finance — and that where the function sits on the org chart silently tells the rest of the company whose interests it serves.

---

Why the team has to mirror the business

The last piece of Part 1 is about talent — and it’s where Dominique’s Uber experience starts to show through. The standard debate in the profession is career auditors versus rotational programs. Her answer is that it’s the wrong debate.

“To do excellent internal audit, it takes a village with very different competencies working together.”

At PACCAR — the Seattle truck manufacturer — the audit function’s “management” is career professionals, but the audit team itself is 100 percent rotational. Senior leaders from engineering, HR, finance, and operations rotate through for two years and then return to the business with a promotion. Over decades, the program has fed PACCAR’s executive pipeline. Several members of their current executive team have done a tour in internal audit. Mark Pigott, the long-serving chairman, started his career there.

The broader principle: the audit team must look like a miniature version of the company it audits. A homogeneous team of accountants cannot credibly audit a technology company. That claim becomes concrete in Part 2, when we get to how Dominique actually staffed the Uber function — and what it meant for auditing machine learning systems.

What I took away

The argument of Part 1 is a reframe, not a recommendation. Dominique is not saying your CFO should fire your CAE or rewrite the audit charter tomorrow. She is saying something more fundamental: the mental model most executives hold about internal audit — the SOX compliance model, the financial controls model, the thermometer model — is describing 1 to 2 percent of what the function is supposed to do.

The 95 percent that strategic and operational risk represent in the empirical record of value destruction is not, in most companies, being systematically audited by anyone. Management is too close to it. The external auditor’s scope doesn’t reach it. The board can oversee it but cannot investigate it. That leaves internal audit — and it leaves the function pointed somewhere else.

In Part 2, we get into what the right answer looks like in practice: how Dominique built the function at Uber through IPO readiness, how she handled the Hong Kong versus London risk-appetite problem, what auditing AI actually looked like on the ground, and — for CFOs who don’t yet have the function at all — how to build it from scratch without getting shut down.

Part 2 drops next week.

At 31, with no finance background whatsoever, Vasant Prabhu was sent to Amsterdam by a future CEO of PepsiCo and told he was now CFO of the European business. He describes himself, in retrospect, as “genuinely dangerous” in that role.

He stayed a CFO for the next 25 years — at Safeway, Starwood Hotels, NBCUniversal, and Visa, where he retired in 2023 as Vice Chairman and CFO. He now sits on the boards of Intuit, Delta Air Lines, and Kenvue, and is an Executive in Residence at Columbia Business School.

I have brought Vasant into my classes and programs many times. What makes him unusual is not his résumé but his complete absence of professional veneer. He says what actually happened, what the fights actually looked like, and where he was wrong. Over two sessions with my online CFO program, participants asked him everything they could not ask their own CEOs: how to handle a difficult boss, when to take a conflict to the board, whether to chase the CEO title, what AI is actually worth, and why he thinks good times are harder for CFOs than crises.

The full conversation — compiled, lightly edited, and organized by theme — follows below. It is long. It earns the length.

*******

What No Textbook Will Teach You About Being a CFO- A Conversation with Vasant Prabhu

On Becoming a CFO — and Staying One

SR: Your background is about as non-traditional as it gets for a CFO. Walk us through how it happened.

Vasant Prabhu (VP): It was entirely accidental. My undergraduate degree was in engineering. I went straight to business school because I didn’t want to be an engineer, and then I went to Booz Allen & Hamilton, where I spent about eight or nine years as a consultant, mostly doing consumer packaged goods and media work. I had little interest in, and almost no exposure to, accounting or treasury or any of the functions that CFOs typically manage.

When I left consulting, I joined PepsiCo in a strategic planning role. About a year in, Roger Enrico walks in and tells me I’m going to Europe to be the CFO of the European business. I was incredulous. I said, ‘Roger, why me? I know nothing about finance.’ He said, ‘We don’t put accountants in these jobs. We put business people in them. And if you screw up, I’ll fire you.’ So I found myself in Amsterdam at 31, genuinely dangerous in that role.

But what I discovered there changed my view completely. Pepsi had redefined the CFO role — along with GE, they were pioneers. The CFO wasn’t a scorekeeper; the CFO was the business person at the general manager’s side, responsible for making sure decisions made financial sense. In fact, the Pepsi philosophy was that if anything went wrong with the business, the first person they would fire was the CFO. That suited me. The job gave me enormous scope to have impact across every dimension of the business.

I ended up staying in CFO roles for almost 25 years across supermarkets, hotels, media, and financial technology. And one of the things I came to appreciate is that the CFO role is one of the few where you can genuinely move across industries. The moment you run a business, you get labeled — you become the hotel guy, or the supermarket guy. As CFO, for whatever reason, people assume you can function across industries. That was enormously important to me. I’m curious, I get bored easily, and I had no desire to be pigeonholed.

SR: You had opportunities to become a CEO over the years. Why did you choose not to?

VP: A few reasons. The CEO job has significantly more people-management burden. Every people problem in the company ultimately lands with the CEO. And the CEO is also the public face of the company to customers, regulators, governments, communities, and all sorts of external constituencies. I was happy to handle investors. I was less interested in spending my time on government relations or community affairs or the full range of stakeholder management that CEOs have to do.

I also realized the CFO role had fewer of the things I found frustrating and more of the things I genuinely enjoyed — analytical problem-solving, capital allocation, deal-making, operating as a close strategic partner to the CEO. Being the number two on strategy while someone else handles the politics of the public-facing role turned out to suit me well.

And then there’s the industry mobility point. The moment you become a CEO, you’re in that industry. I genuinely did not want that. I wanted to keep learning new things.

On Moving Across Industries

SR: Given how different supermarkets, hotels, media, and payments are from each other, what actually transfers from one CFO job to the next, and what forces you to completely rewire your thinking?

VP: There is far more in common than different when it comes to the core CFO disciplines. Accounting rules may vary at the margins, but it is still accounting. Investor relations may mean different investors talking about different metrics, but you are still dealing with investors. Treasury is still cash flow and balance-sheet management. Planning is still planning. So the foundational toolkit transfers.

What is genuinely different is the key value drivers and risk profile of each business. A supermarket is operationally intensive, capital intensive, heavily dependent on labor productivity and inventory turns. A hotel company — at least when Starwood still owned assets — was about yield management, real estate value, global brand building. A media company is driven by content, creative talent, and audience. Visa was a technology company disguised as a financial services firm. So you have to learn the business quickly, and you have to be genuinely humble about what you don’t know.

My rule when joining a new industry was simple: use what you already know to create value and build early credibility, while paddling furiously beneath the surface to learn the specifics as fast as possible. Do not walk in and act like you know everything. You will lose people’s respect within weeks. But show that you can apply the things you already understand, and people will give you the time to learn the rest.

SR: How do you actually land those cross-industry roles in the first place? Companies say they want industry experience.

VP: It depends entirely on who the CEO is. Many CEOs will simply never hire you if you don’t have industry experience. The ones who hired me were confident enough in their own judgment to bet on raw capability over prior industry exposure. Every one of those CEOs was demanding and extraordinary at what they did — and they were willing to trust that if I had the skills they needed, I could learn the industry. Not many CEOs think that way.

The practical reality is that the best jobs come through people you know or people who know you. I was never a great networker — most CFOs are introverts — but the fact is, every significant role I landed came either from someone who knew me directly or from a referral through someone who knew and trusted me. Work in good companies with exceptional people, develop a reputation for results, and those connections will be more valuable than any headhunter. Formal references matter too. Every time I hired someone, I relied more on references than on interviews.

On Career Management

SR: What actually separated you from your peers at Pepsi? Presumably there were many talented people there when Enrico made that call.

VP: Pepsi in those days valued people who thought differently, who were willing to take risks, who stuck their necks out rather than playing it safe. That was the culture Roger Enrico came from. He was essentially throwing me into the deep end at 31 and seeing if I could swim. There were not many companies at that time — this was the early 1980s — that would have done that. Most would have promoted the person who checked every traditional box.

What I learned is that it takes two: you need someone willing to throw you in the water, and you need to be the kind of person willing to be thrown. I always liked situations where someone trusted me with more than my experience would traditionally warrant. I was willing to accept that I might fail. Better to have the chance and fail than to never have it at all.

SR: You have a rule about not staying in a job more than ten years. What’s behind that, and how do you know when it’s actually time to go?

VP: I simply think that after about a decade you go stale — and often you don’t realize it. You stop learning. The challenges feel familiar. Your sharpness erodes. You may be comfortable, but comfort is not the same as contribution.

At Starwood I hit the eleven-year mark, survived three different CEOs along the way — which is unusual, because a new CEO normally wants to bring in their own CFO — and I could see that there was nothing else I was going to do there. I chose to leave for a media role I had long been interested in. At Safeway, the CEO and I had reached a genuine divergence on strategy, so the decision made itself. At NBC I had expected to stay a long time, but personal circumstances and a compelling opportunity at Visa intervened.

The specific trigger is less important than the underlying discipline: you need to impose the constraint on yourself. Most people wait until they’re pushed out, or until they’re so comfortable they’ve stopped growing. Neither is a good outcome.

SR: How did you keep the work genuinely interesting across nearly twenty-five years in essentially the same role?

VP: You have to make sure that every year there is a meaningful new challenge. The job title stays the same; the content should not. When I could no longer identify what the new challenge of the coming year was going to be, I knew it was time to move — either to a new role within the company or to a different company altogether.

The industry switches helped enormously. Going from supermarkets to hotels to media to financial technology meant I was always a student again. That is energizing rather than threatening, if you are genuinely curious. My advice to anyone feeling stagnant is not to wait for someone to force a change. Force it yourself. Comfort is the enemy.

On the CEO-CFO Relationship

SR: The CFO-CEO relationship seems to be at the center of almost every important challenge CFOs face. How do you think about managing it, particularly when you disagree?

VP: If you don’t like conflict, don’t be a CFO — because virtually every major conflict in a company ends up with the CFO. Resource allocation disagreements, acquisition debates, budget fights, strategic direction disputes — they all land on your desk. That is not a pathology, that is the job. If everything were running smoothly and everyone agreed, they would not need you.

My approach was to lead with facts. Other functions have the luxury of strongly held opinions that are not always grounded in data. Finance does not. Your credibility depends on bringing facts to disputes and helping people reason more clearly from them. That is how you earn the right to be heard on questions where the data is less definitive.

At the same time, you have to accept that you will not win every battle. You have to decide which fights are genuinely worth having and which are not. When you have done everything you can and still cannot prevail, you have a choice: accept the decision and make the best of it, or — if the issue is fundamental enough — resign. I was close to that point a few times, but it never came to that. The key is to stay unemotional, maintain the CEO’s trust even through disagreement, and never let a losing position turn you into an obstructionist.

SR: There is also the question of how you handle a CEO who avoids conflict — who would rather keep everyone happy than make a hard call. How do you work with that type?

VP: That is actually quite common. CEOs want to be liked; it is part of the job. The CFO is often the one they rely on to be the necessary bad guy — the person who says no, asks the hard questions, or surfaces the uncomfortable facts that no one else wants to raise. In organizations that are culturally wired to avoid conflict, you have to go looking for it. You cannot wait for it to come to you.

When I identified a conflict — an opportunity the organization was gridlocked on, or a real problem nobody wanted to acknowledge — I went to it directly, armed with facts. You try to be persuasive and structured in how you frame the argument, because if pure data is unavailable, structured logic is the next best thing. You also accept that politics and emotion can never be entirely removed from any human organization, regardless of how clearly you make the case.

SR: When a decision goes against you — say, an acquisition you opposed gets approved — what do you do?

VP: Once a decision is made, you align. You made your case as well as you could, it did not prevail, and now your job is to get the most value out of the outcome you have. You do not pout, and you certainly do not try to subvert the decision. That would make you an obstructionist, and it would rightfully end your credibility.

What you can do is work to mitigate downside risk, track the assumptions that were made in approving the decision, and be prepared to flag early when reality is diverging from the original plan. If the acquisition turns out to have been a mistake — and many do — you can then help the organization pivot rather than clinging to the original plan to avoid the discomfort of admitting error. The sunk cost mindset is the enemy of good post-deal management.

SR: On the subject of boards — how do you navigate a situation where the CEO has not fully disclosed something to the board, or where you hold a meaningfully different view than the CEO?

VP: A good CEO will always cultivate his or her board. That is not a criticism — it would be foolish not to. But it means the board’s primary relationship is with the CEO, and you, as CFO, are always working in the shadow of that dynamic. The board will generally side with the CEO in a public dispute, simply because that is all they can do.

Every audit committee is required to hold executive sessions with the CFO and controller — without the CEO present. That is your formal channel, and you should use it. Be thoughtful about what you raise and how, but know that it is there precisely for situations where you need to say something you cannot comfortably say in front of management. Good audit committee chairs actively invite that kind of candor. Not all board members are equally trustworthy, so you have to read your audience. But the mechanism exists and you should not be afraid to use it appropriately.

SR: How important is physical co-location with the CEO? Does the CFO need to be where the CEO is?

VP: My career was largely pre-COVID, so I am an old-fashioned voice on this. But my honest view is that the CEO-CFO relationship depends significantly on informal, in-person interaction — the conversations in the hallway, the ability to read the room in real time, the relationship equity that accumulates from working side by side over time. I saw the quality of my own relationships deteriorate during the COVID years when we were fully remote.

If you are an effective CFO, you will be a close partner to the CEO on the most sensitive issues the company faces. That kind of trust is much harder to build across time zones and video calls. If you have any choice in the matter, locate yourself where the CEO and the core leadership team are.

On the Craft of the CFO

SR: For CFOs who came up through controllership — the traditional accountant path — what’s your advice on making the transition to a broader business leadership role?

VP: You start with a significant advantage. The controller runs the largest organization within finance, has deep process knowledge, understands external reporting, and is typically close to the technology that runs the finance function. That is a strong foundation.

What you have to shed is the image of the scorekeeper — the person who reports on what already happened. To be seen as CFO material, you need to be seen as forward-looking, as someone who anticipates rather than just records, who provides solutions rather than just problems, who brings new ideas about how the business can run better. That means becoming a change agent within your own function: applying new technology, improving productivity, leading initiatives rather than following them. If you can make the move into financial planning and analysis — running FP&A for a period — that broadens your perspective enormously.

And regardless of your background, the controlling function cannot be neglected. You can be fired for getting that wrong. You will never be promoted for getting it right, because everyone assumes you will. The key is to have a great controller you trust, and to develop enough instinct to know when something may be going wrong before it surfaces as a crisis.

SR: As a self-described introvert, how did you manage the investor-facing dimension of the role — roadshows, earnings calls, investor conferences?

VP: I actually came to genuinely enjoy the investor dimension. Investors are analytically rigorous, they have done their homework, and they ask questions that no one inside the building is asking. They hold an ownership perspective that is fundamentally different from an operating one, and they will tell you things — directly or implicitly — about what you are not seeing clearly from the inside.

The investor interactions were a source of genuine intellectual challenge for me, and a useful check on whether I was focused on the right things. I learned from many of those conversations.

More broadly, any job will have things you like and things you don’t. The rule is: never avoid the things you don’t like. You may never love them, but you have to become competent at them. If you avoid them, they will become your vulnerability. You cannot aspire to the CFO role and say you don’t want to deal with A, B, and C. The job comes as a package.

SR: What are the most critical roles to hire for on a CFO’s team?

VP: For me, the two most important were the controller and the head of FP&A.

The controller, as I mentioned, is the largest organization in finance and the place where the greatest operational and reputational risk sits. Because I did not come from that background, I was always especially deliberate about having an outstanding controller — someone who was excellent on process and technology, not just on technical accounting. You need someone you can trust completely to run that function.

The FP&A leader is the person who should think most like you. When you are not in the room, you want that person to ask the questions you would ask, challenge the assumptions you would challenge, and provide the analysis that genuinely helps management make better decisions. You want your mirror in that role. Treasurer and IR are critically important as well, but the controller and FP&A head are the two I would fill first and fill well.

SR: On finance organization design — there is often pressure to benchmark costs against industry ratios and to offshore or outsource functions. How do you think about that?

VP: Blanket benchmarking does not work. The composition of what different companies include in their finance function headcount varies enormously, so peer comparisons are often meaningless. You need to build from the bottom up: what does this company actually need from its finance function to compete effectively?

I am also skeptical of aggressive outsourcing of core finance functions. You may save money, but the question is what you lose in judgment, responsiveness, institutional knowledge, and business partnership capability. Finance is not a pure cost center; it is an influence function. If you outsource the people who provide that influence, you may be cutting more than you realize. I was conservative on this throughout my career. In retrospect, I think that was the right call.

SR: What is your philosophy on developing talent within the finance function?

VP: Honestly, I was more focused on building a performance-based culture than on the softer aspects of talent development. I cared enormously about the quality of the people I hired, held those reporting to me accountable for building strong organizations beneath them, moved high performers up quickly, and dealt with non-performers. That philosophy — hire well, hold people accountable, reward performance — was the core of what I did.

If I were doing it over, I would invest more explicitly in the broader culture-building aspects. I tended to assume that hiring the right people would take care of culture, which is partly true but not entirely. Finance organizations are often not known for strong people cultures, and that is a gap worth addressing more intentionally than I did.

On M&A, Economic Cycles, and Board Service

SR: M&A has a notoriously poor track record. What’s the right CFO posture toward deal-making?

VP: The track record is genuinely bad, across the board, and not many companies acknowledge it honestly. Most acquisitions do not create the value that was promised when the deal was done. That is simply the reality.

Does that mean you should stop doing deals? No. A diversified portfolio of smaller bets — each sized appropriately, each with defined assumptions and milestones — is probably a better approach than avoiding M&A entirely. You need to be willing to accept that some will not work. The mistake is when companies bet too heavily on a single transformative deal, or when they are slow to acknowledge when integration is failing. I personally erred on the side of caution — perhaps too much so. In retrospect, there may have been deals worth taking more risk on.

On integration specifically: the most common failure is the sunk cost problem. The original plan is not working, real facts on the ground call for a different approach, but management and the board are reluctant to deviate from what was approved because it looks like an admission of error. The CFO’s job is to push through that discomfort. Face the new reality, reset expectations, and do what the business now actually needs — even if that means taking an impairment charge. Clinging to the original plan when it is wrong compounds the damage.

SR: You have navigated both boom and bust cycles. Which is harder for a CFO?

VP: The good times. That surprises people, but it’s true. During a crisis — and I was CFO of Starwood during the 2008 financial crisis, which was devastating for the hotel industry — leadership is actually eager to hear from the CFO. People want rigorous analysis, they want options, they want someone helping them navigate. Those are the moments when the finance function adds the most visible value.

During a boom, the challenge is almost the opposite. Money is flowing, the business is growing, and anyone who pumps the brakes gets labeled as anti-growth, unimaginative, or lacking vision. But that is precisely when discipline matters most — when capital is being allocated loosely, when assumptions about future growth are dangerously optimistic, when people are spending on things that would never survive scrutiny in a tighter environment. Being the voice of caution in those moments is genuinely difficult, because you are fighting against both organizational momentum and the natural human desire to enjoy success.

My biggest challenge at Visa was not the cycles — Visa performed extremely well for nearly all of my tenure — it was the discipline of saying no to things that did not make sense at a company where saying yes to almost anything looked justifiable given the financial results.

SR: Having served as CFO for almost 25 years and now sitting on multiple public boards, how has your view of what a CFO should be evolved?

VP: It has not changed. The characteristics I most valued in a CFO when I was operating are the same ones I look for now from the board perspective: intellectual curiosity and breadth of business understanding, analytical rigor, fact-based approach to conflict, problem-solving orientation, and the courage to seek out difficult issues rather than avoid them.

What the board role has reinforced is the importance of humility. When you are on a board, you meet four times a year and receive curated information. You are not in the business. You should never mistake your past experience — however deep — for current knowledge of the company you are overseeing. The board’s most important decision is whether the CEO is the right person for the role. Beyond that, your job is stewardship, oversight, and being available if things go meaningfully wrong.

The worst board members are those who confuse their own experience or opinions with actionable knowledge of the company in front of them. As a CFO operating under board oversight, the best thing you can do is cultivate good relationships with your audit committee, be candid in executive sessions, and make sure the board has the information it needs to actually do its job — not just the information that makes management look good.

On Technology, AI, and the Evolving Role

SR: How do you see AI changing the CFO role, and what is the right posture for finance leaders navigating the current environment?

VP: AI has been significantly overhyped, and the CFO community needs to hear that clearly. The amount of capital being deployed at major technology companies on AI infrastructure has been extraordinary, and much of it has been allocated with a faith-based logic — invest now, the returns will come eventually — rather than demonstrated ROI.

If I were a CFO today, I would be an active experimenter in a controlled way. I would be piloting AI applications in the control function as a productivity tool, and exploring its potential as a decision-support capability in FP&A. I would be looking for evidence that specific applications actually generate measurable improvements in output quality, speed, or cost.

What I would not do is position myself as the company’s AI cheerleader. That role is available, and winning it generates short-term goodwill. But the CFO’s more important job right now is to be the skeptic: to ask ‘what does this actually produce, and how does it show up in revenue growth or cost reduction?’ You are the person most likely to stop your organization from wasting significant resources on initiatives with no clear business case. That is a valuable and necessary function, and it is the historically classic role of the finance function in any hype cycle.

Apple is the company I would hold up as a model in this environment. They have been notably disciplined — criticized for being behind, in fact. I would not rule out that they prove to have been right. They focus on proven applications with clear economics and do not burn capital chasing speculative infrastructure. That is a CFO mindset.

SR: Some of our participants are facing board pressure to chase AI — boards that are reading the same breathless coverage and asking why management isn’t doing more. How do you handle that?

VP: Manage your board. That is the answer. Boards can be genuinely dangerous on this topic because many board members are not deeply knowledgeable about AI, but they feel pressure to appear informed and engaged. They read the same popular coverage everyone else does and arrive at board meetings with questions that may send the management team on expensive wild goose chases.

The solution is proactive education — share what you are doing, what you have learned, and what the legitimate constraints are — combined with willingness to push back when a board suggestion would create distraction rather than value. You and the CEO need to be aligned on this. If a CEO who was trained as a consultant is inclined to give the board whatever it asks for, the CFO has to be the one ensuring that internal resources remain focused on what the business actually needs. The performance of the business is ultimately how the board will evaluate management, not whether every board member’s suggestion was implemented.

SR: How has the CFO role itself evolved over your career, and where is it heading?

VP: When I started, my background was unusual. Most CFOs came from accounting — they were controllers who became CFOs, or they had CPA backgrounds. Today, people with my kind of background are common. The role has decisively shifted from technical financial oversight to broad business leadership. That is an irreversible change, and it is accelerating.

The dimensions that are becoming more important are technology literacy and global orientation. At Visa, IT reported to the CEO rather than to me, because the technology was the business — not a support function. CFOs increasingly need to be fluent partners with their CTOs, not just consumers of technology outputs. Being an engineer by background made me naturally comfortable with technology conversations, but that comfort level needs to be developed regardless of background.

Global scope has also become the baseline, not the exception. Almost every technology company is global from very early in its development. Having worked overseas is not a prerequisite, but a genuine global orientation — comfort with different regulatory, cultural, and economic environments — is increasingly essential.

Key Takeaways for CFOs

The following themes emerged consistently across both conversations:

On Career Architecture

• The CFO role is one of the few that allows genuine cross-industry mobility. Resist the pull toward operational roles early in your career if you value continued reinvention.

• The best jobs come through people who know you or trust those who know you. Invest in relationships within good companies — the network you build there will be more valuable than any formal job search.

• Impose your own ten-year rule. After a decade in any role, you risk going stale. Force change before it is forced on you.

• Being thrown in the deep end — at 31 with no finance background — is not the exception in great CFO careers. Seek employers and CEOs who bet on potential over pedigree.

On Conflict and Influence

• If you are conflict-averse, the CFO role is the wrong choice. Almost every major organizational disagreement eventually arrives on the CFO’s desk. The best CFOs run toward conflict, not away from it.

• Lead with facts. Other functions may hold strong opinions without strong data. Finance cannot. Your credibility depends on being the most rigorously grounded voice in the room.

• When a decision goes against you, align and make the best of it. Do not pout. Do not subvert. But stay close to the assumptions underlying the decision so you can flag divergence early.

• The CEO-CFO relationship requires physical proximity and consistent informal contact. Remote working relationships are viable for many roles; this is not one of them.

On Board Dynamics

• The board is the CEO’s domain. Do not mistake your access to the board for equivalence with the CEO’s relationship with it.

• Executive sessions are not a formality — they are the mechanism through which the CFO can surface material issues without the CEO present. Use them honestly and thoughtfully.

• As a board member, your most important contribution is clarity about what you do and do not know. Never mistake past experience for current knowledge of the company in front of you.

• Boards that are applying hype-driven pressure on management need to be educated and, where necessary, pushed back against. You are running the business. They are not.

On the Finance Function

• If you came up through controllership, your path to the CFO seat runs through escaping the scorekeeper image. Become a forward-looking problem-solver and change agent within your function.

• Hire the best possible controller. You can be fired for getting the control function wrong. You will never be promoted for getting it right.

• Your FP&A head should think like you. They are your analytical proxy in every room you cannot be in.

• Do not allow benchmarking or outsourcing pressure to hollow out the finance function’s judgment capability. The value finance creates comes from influence, not from headcount reduction.

On M&A and Capital Discipline

• The M&A track record across public companies is poor. Accept that before you start. Pursue a portfolio approach to deals rather than betting heavily on individual transactions.

• The most common post-deal failure is the sunk cost trap: refusing to pivot from a failing integration plan because doing so requires admitting error. The CFO must be willing to call that out.

• Crises are the CFO’s best moments. Bad times create demand for rigorous analysis and financial discipline. Good times are harder — that is when organizational resistance to discipline is greatest.

On AI and Technology

• Be the skeptic on AI, not the cheerleader. The company needs someone asking ‘where is the revenue growth or cost reduction?’ for every initiative. That is a core CFO function.

• Pilot AI in the control function for productivity gains and in FP&A for decision support. Build gates and checkpoints before scaling spending. Show me the beef before you show me the vision.

• Technology literacy is now a baseline requirement for CFOs, not a differentiator. Invest in your ability to partner authentically with your CTO.

I’ve been writing about this for nearly a decade. The problem they’re trying to solve is real. The solution is wrong.

We actually ran this experiment. The results were not encouraging.

The UK scrapped mandatory quarterly reporting in 2014. My co-authors Suresh Nallareddy, Bob Pozen, and I studied what happened. The findings:

• Fewer than 10% of UK firms voluntarily switched to half-yearly reporting when given the option. The other 90% kept filing quarterly anyway. CEOs worried the market would read a switch as a signal they had something to hide.

• Firms that did switch lost analyst coverage. Analysts organize around earnings releases. Half the releases, half the attention.

• Most importantly: investment patterns didn’t change. R&D, capex, M&A — none of it moved. Not up, not down.

If you hoped that fewer reports would liberate management to invest more boldly in the long term, the data says: not quite.

Short-termism is real. But quarterly reports didn’t necessarily cause it.

I made this argument in HBR in 2018 when President Trump first pushed this idea, and again in Forbes last September when he pushed it again. The core hasn’t changed because the underlying economics hasn’t.

One of the potential drivers of short-termism is how US GAAP treats investment. R&D, workforce training, brand building — all expensed immediately rather than treated as the assets they are. So when a CEO cuts R&D to make quarterly earnings look better, the income statement can’t tell the difference between prudent discipline and self-defeating underinvestment. That distortion exists at any reporting frequency. (Barron’s, 2016.) And with CEO tenure now averaging just six years, managerial incentives are already structured around the short term — tinkering with reporting frequency doesn’t touch that. A wise man once told me- if you don’t understand something, consider either lobbying or career concerns as the best hypothesis to explain that mystery.

One more point nobody is saying out loud: fraud loves a blackout period. Quarterly reports are a forcing function — every 90 days, management must stand and account. Give bad actors six months instead of three and frauds run longer, get bigger, and when they collapse, the crash is more vicious.

On compliance costs: I just ran the numbers.

The compliance-cost argument is the one that sounds most reasonable. So I tested it in a piece out this week in Forbes, using ExxonMobil as the case study. The company recently told the SEC that producing its annual 10-K requires roughly 20,000 employee hours — a “considerable undertaking,” in their words. I tripled that estimate generously to account for legal, executive, and board time, applied a loaded cost of $250 an hour, and arrived at $15 million for the entire exercise.

Then I looked at the benefit side. Being a publicly listed company rather than a private one is worth a 20–25% premium on market capitalization — the liquidity premium documented extensively in the academic literature. On ExxonMobil’s $648 billion market cap, that’s $130–162 billion. S&P 500 index membership, which mandatory disclosure makes possible, adds another $160–230 billion in structural demand premium from benchmark-constrained fund managers. Total shareholder benefit from the disclosure framework: $290–390 billion.

The return on ExxonMobil’s $15 million compliance investment: 19,000 to 26,000 to one. For every dollar the company spends producing its 10-K, shareholders receive roughly $19,000 in market value from the institutional framework that disclosure makes possible.

The quarterly report is a far simpler document than the 10-K. Its ROI would be even more striking.

What should actually change?

Fix the accounting for intangible assets so investment in R&D and human capital doesn’t mechanically depress earnings. Fix compensation structures that reward 90-day EPS beats over five-year value creation. And if reporting compliance costs genuinely drive companies to stay private — which I examined in Forbes last June — the answer is simplifying what gets disclosed, not reducing how often companies are held accountable to their shareholders.

Cutting the quarterly report treats the symptom. The ExxonMobil math suggests it might even be cutting something with a 19,000-to-1 return.

The full archive on this topic:

• “Corporate Short-Termism Leads to Long-Term Pain” — Barron’s, 2016

• “These Changes to Quarterly Reports Would Benefit Companies and Investors” — Marketwatch, 2017

• “We Can’t Study Short-Termism Without the Right Metrics” — HBR, 2017

• “What Would Happen if the U.S. Stopped Requiring Quarterly Earnings Reports?” — HBR, 2018

• “Have Reporting Burdens Led to More Firms Staying Private?” — Forbes, June 2025

• “Should We Ban Quarterly Reporting?” — Forbes, September 2025

• “ExxonMobil’s 10-K Compliance Cost Delivers a 19,000-to-1 Return for Shareholders” — Forbes, April 2026

• “Consequences of More Frequent Reporting: The U.K. Experience” (with Nallareddy and Pozen) — Journal of Law, Finance and Accounting, 2021

Few people have a clearer window into what it actually takes to become — and thrive as — a CFO than Smriti Vicari. A Partner at Egon Zehnder, one of the world’s most respected executive search and leadership advisory firms, Smriti leads searches for CFOs, CEOs, and board members with a particular focus on financial services, fintech, consumer businesses, and scale-up companies. She holds a B.Sc. in Finance from the University of Southern California and an MBA from London Business School, and she began her Egon Zehnder career in London before relocating to New York. Over the past six years, Smriti has advised boards and investors — from large global banks and private equity firms to high-growth fintechs — on how to find, develop, and retain world-class financial leadership. Her firm recently published a landmark global CFO study surveying 600 chief financial officers across industries and geographies, offering a rare empirical lens on how the role is changing.

The following interview, conducted as part of the Columbia CFO Program’s Faculty Live Online series, distills the key insights Smriti shared with current and aspiring CFOs on what the market demands, how headhunters really assess candidates, and what separates good from truly great in the C-suite.

The Expanding Remit of the CFO

Shiva Rajgopal: Let’s start with the big picture. Your firm just published a major study of 600 CFOs globally. What was the headline finding?

Smriti Vicari: The role is bigger and more complex than ever before. There’s a traditional model of the CFO — FP&A, Treasury, Tax, the core technical toolkit — but over the last five years there’s been an increasingly broad and complex set of responsibilities landing in the CFO’s lap. We’re talking about ESG, operations, M&A, risk management, cybersecurity, and supply chain. Things that would have seemed outside the finance remit entirely a decade ago.

Shiva Rajgopal: As one CEO put it to me, the CFO’s job is basically whatever the CEO does but needs to put somewhere — and it’s because the CFO is so often the trusted lieutenant to both the CEO and the board. That breadth is only going to grow.

Shiva Rajgopal: Is there tension within the function about whether that’s actually the right model?

Smriti Vicari: Absolutely. Some CFOs feel strongly that their core job is the traditional one — that they shouldn’t be taking on all these extra responsibilities — while others genuinely relish the breadth. No COO role is the same across companies, and the CFO role is beginning to resemble that: context-dependent. The real debate is whether the role can be done well if it gets too broad, and what gets sacrificed when a CFO is pulled in so many directions. Boards are wrestling with that question too.

What Boards and CEOs Actually Look For

Shiva Rajgopal: When you survey CFOs about the competencies that matter most, what comes back?

Smriti Vicari: The number one competency is the ability to drive change. Not just being strategically oriented but having the network inside an organization to facilitate change, inspiring people to come along, and having the resilience and grit to push things through. The best CFOs are genuine change agents — not incremental BAU operators. That’s becoming a huge differentiator. Strategy is the other big one. The ‘strategic CFO’ has become an increasingly popular model with CEOs. But I’d push back a little on the idea that technical CFOs aren’t strategic — they are. The better question is whether someone has the breadth of experience to advise across a wider set of issues.

Shiva Rajgopal: You’ve given some vivid examples of how two banking CFOs approach this very differently. Can you walk us through that?

Smriti Vicari: Sure. Take Jeremy Barnum at JPMorgan — excellent CFO, grew up as a capital markets banker, very technically deep, and he would not place anyone on his bench who didn’t have significant depth in the finance function itself. That’s his model. Now compare that to Mike Santimassimo at Wells Fargo, who I spoke to recently — he’s open to investment bankers, strategy professionals, people who bring different experience and can help think about the broadening remit. Two world-class CFOs, fundamentally different views on how to build talent. So, when I’m doing a search, I always have to understand the CEO’s model first.

Shiva Rajgopal: What about the internal composition of the finance team? Which roles do CFOs lean on most?

Smriti Vicari: FP&A was the clear frontrunner in our survey. The head of FP&A tends to have the broadest network across the business, carries strategic firepower, and is often a strong candidate for CFO succession. The Controller is also critical — that’s a role where you simply cannot sacrifice technical expertise; you can’t put someone from strategy in a controllership and expect it to work. And in banking specifically, Treasury is hugely important — the CFO at State Street, John Woods, came up through that route from Citizens, for example. But the ranking will differ by industry.

Shiva Rajgopal: When you ask CFOs what they wish they could spend more time on, what do you hear?

Smriti Vicari: Strategy, overwhelmingly. Most feel pulled into operational firefighting day-to-day and they want to be a more central voice in shaping the business. The second thing is developing their own teams. And this is important: we always tell our clients that if the only option when a CFO leaves is to go external, then something has gone wrong. You’ve failed as a leader. Developing internal succession is a core responsibility of the role, and frankly, too many CFOs wait too long to start doing that work seriously.

How the Search Process Actually Works

Shiva Rajgopal: Let’s lift the hood on how boards do succession planning. What does that actually look like in practice?

Smriti Vicari: Best-in-class boards do annual market scans — and for CFO and CEO, only those two roles, which tells you something about how critically they view this position. The exercise is simple: if our CFO won the lottery and moved to the Maldives tomorrow, who externally and internally is ready right now, and who is ready in 18 months? We do referencing, we call around, we score candidates across criteria — readiness, team leadership, strategy, AI capability — often without even speaking to them directly, just by talking to people around them. Then they do the same exercise with internal candidates and compare the two cohorts. The preference is almost always to go internal if a strong candidate exists. But here’s the key insight: they never evaluate internals in a vacuum. They have to see how their people stack up against the external market. And CEOs and board members will often meet two or three external candidates informally each year, just to calibrate.

Shiva Rajgopal: So your external network actually has a direct bearing on whether you get a promotion inside your own company?

Smriti Vicari: Exactly — and this is something people underestimate. We’ll score external candidates based on what their network says about them. If people aren’t talking about you, you’re invisible in that process. The network you build is not just for your next external move. It’s shaping how boards see your readiness right now.

Shiva Rajgopal: What does the distinction between internal and external candidates look like from your perspective?

Smriti Vicari: In most cases, there is a clear preference for internal talent. An internal candidate knows the business, has the networks, understands the context, and can hit the ground running. When a CFO has successfully developed a strong internal successor, that’s a sign of real leadership. The external search comes in either when there’s no internal candidate ready — which is, frankly, a failure of talent development — or when a company needs a genuinely different skill set that doesn’t exist inside. For companies that are highly M&A-driven, for instance, they might prioritize deep corporate development experience and explicitly decide to supplement the CFO with strong technical number twos.

AI, Technology, and the Skills That Will Define the Next Generation

Shiva Rajgopal: Everyone is talking about AI. What does it actually mean for how you assess CFO candidates today?

Smriti Vicari: It’s changed how we do our job markedly in the last twelve months. We’ve now included a specific AI competency in our CFO assessment framework. But what we’re really measuring is not ‘do you know how to use a particular tool’ — it’s learning agility. Can you demonstrate that you’ve been dropped into very different situations and thrown yourself into the deep end and learned? We look for career variety as a proxy for that — people who’ve worked in large globals and in privates, who’ve been individual contributors and managed large teams, who’ve changed industries. That mix signals someone who can adapt. On AI specifically: a CFO should be using Co-pilot like tools on a regular basis, should have participated in or driven a finance function technology transformation, and should be actively helping educate their teams. That’s now a baseline expectation.

Shiva Rajgopal: What are CFOs telling you about the actual impact of AI on finance operations?

Smriti Vicari: This is where the conversation gets really candid. The uncomfortable truth is that the productivity changes AI enables could be implemented almost immediately, if companies wanted to act that fast. The technology exists today to remake significant parts of a finance function. One CEO I spoke with told me there is not a single question that comes into his bank’s call center that cannot be answered by an AI agent. Not one. But wiping out an entire call center has social, operational, and emotional workforce implications that go way beyond cost savings. So, the real question isn’t whether to do it — it’s what’s the responsible pace of change, and how do you prepare your workforce rather than simply displace them. CFOs are at the center of that decision, and it’s one of the biggest questions on their minds right now.

Shiva Rajgopal: What about data infrastructure? Is AI actually solving CFOs’ information problems?

Smriti Vicari: Not as cleanly as people imagine. One banking CFO told me recently that getting excellent data-driven insights still feels ‘artisanal’ — like it requires a heroic manual effort even with all the modern tools available. The issue is data infrastructure in large, complex organizations. The pipes don’t connect. It’s garbage in, garbage out, regardless of what sits on top. So, when we do CFO searches, we specifically ask candidates: what kind of data or technology transformation have you driven within your finance function? What have you done to markedly improve operations from a data perspective? The answer tells us a lot about whether someone will be effective in this environment — not just whether they understand AI conceptually.

The CFO–CEO Relationship and the Path to the Top

Shiva Rajgopal: A number of CFOs aspire to the CEO role. What separates those who make that leap from those who don’t?

Smriti Vicari: Stakeholder management. Full stop. The CEO role is almost entirely about that. Managing the board, managing external communications, managing employees through uncertainty and change — those are the skills that distinguish a CFO who can step up from one who can’t. Where it goes wrong is when a technically excellent CFO has had little exposure to owning those external communications and relationships. The board sees them as a technical archetype rather than someone with broader strategic presence. My advice: start developing those relationships with board members and senior executives beyond your own remit as early as possible. Participate in things that go beyond your lane. Build the brand awareness that there is much more to you than the numbers.

Shiva Rajgopal: How does the CEO-CFO relationship work best in practice? And what happens when it doesn’t?

Smriti Vicari: The best CFOs and CEOs are genuinely co-running the business. It’s a relationship of trust that often extends to the personal level — knowing each other’s families, multiple touchpoints per day. The CFO frequently leads investor meetings, runs external communications, and is the co-face of the business alongside the CEO. When it works well, there’s no siloing — a CFO who wants to sit in a call center and listen to customer complaints, or who goes and learns how to code with developers, should be celebrated, not questioned. A CEO who sees that as overstepping is, in my view, not someone you want to work for.

Shiva Rajgopal: Where do boards and CEOs diverge in how they assess CFO candidates, and how do you bridge that gap?

Smriti Vicari: CEOs tend to be short-term in their thinking — they’re fighting immediate fires and need someone who can solve today’s problems. Boards are constitutionally long-term — they’re thinking about CEO succession, about whether this person could ultimately step into the top role. So, boards often want a broader skill set, while CEOs want proven capabilities against immediate needs. The best way to bridge that is to be very explicit about what the new CFO is coming in to do and what support they’ll need. Someone who’s motivated to take a role often brings growth opportunities and therefore gaps — and the job is to build the right structure around them to address those gaps in the short term while they develop.

Building Your Team, Your Network, and Your Board Profile

Shiva Rajgopal: When you coach sitting CFOs on developing their teams, what are the one or two things that matter most?

Smriti Vicari: The first is giving high-potential people early board exposure. Bring select direct reports into the boardroom with you. Let them present on a topic. Let them start building those relationships directly. The second — and this one is underestimated — is customized rotation. A lot of finance organizations have a program that rotates everyone through the same six-month stints. That’s lazy. What actually develops talent is looking at a specific individual, identifying where their gaps are, and designing a 12-to-18-month path specifically for that person. And the third is breaking insularity. The biggest weakness I see in finance teams globally is that people just don’t pick their heads up. They don’t have diverse sources of learning. Creating genuine external connections — with investors, with peers in other industries, with the broader ecosystem — is something CFOs should be actively manufacturing for their people.

Shiva Rajgopal: Should sitting CFOs be pursuing non-executive board seats?

Smriti Vicari: Yes, and sooner than most people think. There’s growing demand for people with current operating experience on boards — the pace of change requires it. I’d say that once you’re well into building your CFO profile, or already in a CFO role, it’s absolutely the right time to explore. Even starting with a non-profit audit committee is valuable. Board roles are highly competitive, and prior board experience is now a genuine differentiator when boards are selecting new independent directors. Starting to build that muscle early makes a real difference.

Shiva Rajgopal: Practically speaking, how important are CVs and LinkedIn in the search process?

Smriti Vicari: More important than some think, but not in the way most assume. Our research teams are actively scraping LinkedIn alongside CVs, so having a rich, detailed LinkedIn presence is critical. On the CV itself: one to two pages of key achievements is ideal. A six-page CV will get fed into AI and summarized anyway. But here’s what I really want you to understand: at the senior level, there are two pipelines in every search. One is the research team combing through databases and LinkedIn profiles. The other is me and my colleagues calling CEOs and investors and asking ‘who’s exceptional right now?’ The people who come through that second pipeline — through referral and reputation — are often treated differently. Your CV is a ticket to the first pipeline. Your network and how people talk about you is your ticket to the second.

Deal-Breakers, Diversity, and the Realities of the Market

Shiva Rajgopal: When a candidate gets to the final stages of a process, what actually causes things to fall apart?

Smriti Vicari: By the final stage, experience is largely off the table — they’ve gotten there because they have the experience. What derails people at the end is almost always competency or relationship fit. We’ve had processes fail because a candidate and a board member had a dinner that went badly — not just poor chemistry, but values misalignment that surfaced over a meal in ways it hadn’t in formal interviews. The other thing that derails people is a lack of self-curiosity. We use a four-component model of potential: curiosity, insight, engagement, and determination. And the dimension where we see the most failure at senior levels is self-curiosity — specifically, the knowledge that you don’t know everything. When a candidate walks in and says ‘I can do this job with my eyes closed,’ that is a red flag. What we look for is someone who can articulate exactly where they can hit the ground running on day one and where they genuinely need a strong number two. That self-awareness is what differentiates good from great.

Shiva Rajgopal: Compensation — can mismatched expectations derail a process?

Smriti Vicari: Not if we’ve done our job properly, and that’s on us. Compensation alignment should happen very early in a search process — we have market data, we have visibility into your expectations, and a good recruiter should be constantly calibrating both sides throughout. If there’s a fundamental mismatch, we address it before the process even begins. Where surprises do occasionally happen is around unvested equity — some candidates don’t fully calculate what they’d be walking away from until they’re about to sign. Those situations can be tricky, but they’re not the norm.

Shiva Rajgopal: There’s clearly a diversity and gender gap in finance leadership. What are you seeing, and what’s your honest assessment of whether it’s changing?

Smriti Vicari: There is a real gap — gender, diversity, across the board. And certain markets are more conservative than others. I’m embarrassed when I hear about candidates in my own industry being asked inappropriate questions about their personal lives — that should not happen, full stop, and we have a collective responsibility to call it out. My honest view is that the disruption happening right now — the fact that historical models of CFO success are being fundamentally challenged by AI, by changing business models, by the pace of change — creates a genuine opening. The CFOs of the past are not necessarily the right ones for the future. If you have to learn everything differently, then the question of whether you learned accounting the way your predecessor did becomes less relevant. I see real momentum toward bringing in different profiles, different backgrounds, different archetypes. It hasn’t fully arrived everywhere, but it hasn’t stopped either. The change is real — it’s just uneven.

Closing Thoughts

Shiva Rajgopal: As a final thought for our students — what is the single most important thing a CFO or aspiring CFO should do differently based on everything you’ve shared today?

Smriti Vicari: Build relationships relentlessly — and make sure those relationships understand the full three-dimensional version of you, not just what’s on your CV. Just doing your job brilliantly is necessary but not sufficient. The people who advocate for you, who know your story, who will say your name when a board member calls asking who’s exceptional in this space — those relationships are what determine your trajectory. The technical skills matter enormously. But the network, the reputation, and the visibility of your whole story are what elevate you. That’s the work that happens outside the building, and most CFOs start doing it far too late.

Key Takeaways for CFOs and Aspiring CFOs

The following ten themes capture the most actionable insights from this conversation:

1. The remit is expanding — embrace it or manage it deliberately.

The CFO role now routinely encompasses ESG, cybersecurity, M&A, operations, and more. Whether you embrace that breadth or negotiate tighter boundaries with your CEO, the expansion will not reverse. Be intentional.

2. Change agility is the defining competency.

Technical depth remains essential, but the ability to drive change — with resilience, strategic clarity, and organizational influence — is what separates the best CFOs from the rest.

3. Your internal succession is a reflection of your leadership.

If a board can only go external when you leave, you have failed as a leader. Start building your bench actively and deliberately, and do it years earlier than feels necessary.

4. Boards assess you against the external market — continuously.

Even if you’re an internal candidate, best-in-class boards are annually scanning the external CFO talent pool and quietly calibrating you against it. Your visibility in that external market shapes your promotion trajectory.

5. AI is a baseline — learning agility is what gets assessed.

Using Co-pilot tools, participating in finance function transformations, and demonstrating curiosity about how technology reshapes your work are now table stakes. What recruiters really probe for is whether you can continuously adapt and learn.

6. Stakeholder management is the gateway to the CEO role.

If you aspire to the top job, the single biggest gap you likely need to close is external communications, board relationships, and the ability to inspire broad audiences. Build those skills long before you need them.

7. Self-awareness is a deal-breaker — or a deal-maker.

Candidates who cannot articulate where they need a strong number two lose searches. Knowing your gaps and showing how you’d address them is a sign of strength, not weakness.

8. Network is curriculum.

The most powerful pipeline in any executive search runs through referral networks, not databases. Invest in knowing your peers, your industry’s investors, and key recruiters — and make sure they know your full story, not just your last two roles.

9. Pursue board experience earlier than you think.

Board roles are competitive, and prior board experience now differentiates candidates. Starting on a non-profit audit committee or advisory board early builds a muscle that compounds over time.

10. HR is too important to delegate.

As a CFO, you have enormous influence over culture and talent across the enterprise — not just within finance. The best CFOs own that responsibility rather than relegating it to HR. Culture and human capital are every senior leader’s job.

About Smriti Vicari

Smriti Vicari is a Partner at Egon Zehnder, where she leads CEO, CFO, and board searches across financial services, fintech, and consumer sectors. She holds a B.Sc. in Finance from the University of Southern California and an MBA from London Business School. Egon Zehnder’s global CFO study is available at egonzehnder.com.

I have been walking past the building where this happens for ten years. The accounting professors don’t teach over there.

But I have been quietly convinced, for most of that time, that we have the audience wrong.

Undergraduates read Antigone and understand that she is brave. They don’t yet understand what it costs to be Ismene — her sister, who privately agrees that the burial is right, says nothing, survives, and has to live with that for the rest of her life. You need to have been Ismene to understand Ismene. Most undergraduates haven’t been. Most of the CFOs and board members I teach in executive education here at Columbia have been Ismene at least once. They recognized her in the first two minutes of the scene, before I said a word.

That recognition is what I am building a course around.

Here is the idea, as simply as I can put it.

There are five works of literature — I will name three today and hold two for future posts — that I believe contain the most precise and useful diagnosis of corporate governance failure ever written. Not useful in the way a compliance manual is useful. Useful in the way that a mirror is useful: they show you what you already know about yourself but have not yet had language for.

The first is Antigone, the Sophocles play from 441 BCE. It is, on the surface, about a woman who defies a king’s order to bury her brother. It is actually about the structural failure of institutions that cannot accommodate legitimate dissent. Creon — the king, the CEO, the board chair — is not a villain. He is a reasonable man who has built a system with no formal mechanism for someone to tell him he is wrong without being destroyed for it. The governance lesson is not “listen to your whistleblowers.” It is more uncomfortable: an organization where the only options are compliance or exile is not managing risk. It is accumulating it, invisibly, in the body of every Ismene who chose to stay quiet and stay employed.

The second is Macbeth — specifically Act I, Scene 7, which is one of the greatest dramatizations of professional rationalization ever written. Macbeth sits alone before the murder and works through every argument against it with complete clarity and rigor. He knows exactly what he is about to do and exactly why he shouldn’t. Then he does it anyway. What makes this the right text for a CFO course is not the murder. It is the soliloquy. Every person who has ever engaged in earnings management — even the mild, defensible kind, the kind that is technically within GAAP — has had some version of this speech running in their head. The play’s real subject is not evil. It is the architecture of the first compromise, and the specific structural condition that makes the second one easier.

The third is Tolstoy’s The Death of Ivan Ilyich, which I use not for its darkness but for its precision. Ivan Ilyich spent his entire career doing the right things — getting the promotions, making the right moves, acquiring the right credentials and the right house with the right curtains. He dies realizing none of it was real. The four chapters I assign — which take about twenty minutes to read — are the most efficient treatment of long-termism I have ever encountered. Not as an argument. As an experience.

The course I am building around these texts — which I plan to teach in Columbia’s executive education program — runs for three and a half hours. The participants are sitting CFOs, audit committee members, and board directors. They are intelligent, time-pressed, and constitutionally skeptical of anything that sounds soft.

So the class is not soft.

We read four or five pages per text — the pages where the moral weight lives — with marginal annotations that translate the archaic into the corporate. We map the characters onto real governance roles: Creon is the imperial CEO, Ismene is the colleague who privately agrees with the dissenter and says nothing, the Chorus is the institutional shareholders who observe clearly and act too slowly. Then we run the argument. I take one side, the room takes the other, and we find the fracture.

The tool I use for this — and I want to be transparent about it because I think it is genuinely interesting, not just as a gimmick — is Claude. I run a live AI session during the class. Not because I need the AI to tell me what Antigone means. I use it because the AI has no stake in flattering anyone in the room, will argue the opposing case with complete rigor, and can generate — in real time, on screen — a precise parallel between a scene written in 441 BCE and the Boeing 737 MAX certification process, or the Enron audit committee minutes, or the WorldCom fraud timeline. The literary text stops being “soft” the moment it is sitting next to the actual board minutes from the thing it describes.

I am going to serialize this on The Googly.

Over the next several months, I will work through each text the way I plan to work through it in the classroom: the governance argument, the character map, the real case parallel, and the question I will ask the room that has no comfortable answer. This is a way of building the course in public, thinking through what works, and finding out whether the people who read about corporate governance for a living find the same things in these texts that I do.

I will also, eventually, I think, write a book about this. There is something larger here about what AI makes possible in the humanities classroom — about using Claude not as a replacement for literary expertise but as a way of making literary expertise available in rooms that would never otherwise encounter it. I don’t know exactly what that book looks like yet. Writing these posts is how I will find out.

But the place to start is Antigone.

The post goes up in a few weeks. If you already know which corporate governance crisis I am thinking of as the live case parallel for a play about a woman who defied a king’s order in order to honor a higher obligation — and if that crisis involves the certification of something that people privately knew was wrong — you probably have a sense of where this is going. If you have suggestions on what else I should use, please share.

More soon.

— Shiva

I’ve spent my career throwing googlies at conventional business wisdom.

I’m Shiva Rajgopal, the Kester and Byrnes Professor of Accounting at Columbia Business School. For over two decades, I’ve studied how corporations account for what they do, how boards govern — or fail to govern — the companies they oversee, and more recently, how artificial intelligence is rewriting the rules for both. I write a fortnightly column for Forbes, and my academic research, otherwise buried in obscure journals that no one reads, has appeared in a Cliff-notes version in the Harvard Business Review, the Wall Street Journal, and Barron’s and other places.

I was raised in Bombay (now Mumbai), which means I grew up watching cricket, arguing about Bollywood, and learning early that the most interesting things in life rarely go where you expect them to.

This newsletter is the natural next step.

Why The Googly? Why now?

I am currently writing three books — and I’ll be honest, that’s a big part of why this newsletter exists.

The first, with my friend Bob Eccles who has his own Substack, is on integrated reporting and sustainability what it really means to integrate E and S and G into future cash flows and cash flow risk. Our mission is to make sustainability boring again by getting the capital markets ecosystem to focus on the basics of value creation, stripped of the emotion, rhetoric and politics.

The second, also with Bob Eccles, is on the basics of boardrooms: a plain-language guide to how boards actually work, who sits on them, what they’re supposed to do, are you ready for a board, what industries or companies might you be a good fit for, how and the yawning gap between theory and practice. I have been involved in board education at Columbia for a decade or more and have had the good fortune of learning from the best and brightest in boardrooms. Bob, is arguably, the yoda of governance and sustainability.

The third is on AI in the boardroom: how directors can use AI to become better directors. The premise of being a director is inherently fragile. Directors depend on the CEO to give them information to evaluate the CEO. Most directors, apart from the very best, are overwhelmed by the informational demands of the job and are rarely equipped to intelligently consume and critique what management tells them. The objective is to equip directors with AI tools to make them better at their jobs. For instance, help the technologist better understand the 10-K and the proxy statement if she finds herself on the audit committee. Help the strategy person evaluate what the compensation consultant is telling her if she finds herself on the comp committee. Help the chair of the Nom and Gov committee identify skill gaps the board needs to address and help identify new talent instead of fishing in the same old tired waters of friends and family.

Three big topics. One newsletter. Delivered fortnightly — because good ideas deserve more than a tweet, but your inbox deserves more respect than a weekly obligation.

What you can expect

Each issue of The Googly will take one idea — from governance, financial reporting, sustainability, AI, capital markets, or the occasional detour into cricket or Bollywood — and turn it over carefully. I’ll bring my research, my Forbes writing, and occasionally my opinions, which my Columbia students will tell you I am never short of.

I will try to be contrarian without being contrarian for its own sake. I will try to be rigorous without being unreadable. And I will always try to be honest, even when — especially when — the honest take is the uncomfortable one.

That’s the promise.

If someone forwarded this to you, you can subscribe here. It’s free. And if you think a friend, a board director, a CFO, or a sustainability officer would find this useful, please pass it along.

The first delivery is coming soon. I hope it spins the way you don’t expect.

— Shiva

Columbia Business School | Forbes Columnist | Mumbai, originally

Subscribe now