A bookkeeping rule buried in US generally accepted accounting principles — one that governs whether a company records a transaction at its full value or only its retained share — produced a $20 billion discrepancy in OpenAI’s reported revenue and triggered a broad selloff across AI-adjacent equities Thursday.
The Financial Times reported Thursday that OpenAI’s actual annualized revenue run rate approached $50 billion at the end of September 2026 — a figure $20 billion below the $70 billion estimate that had circulated among investors and driven a fresh wave of AI market optimism. Nvidia fell nearly 3 percent, Oracle shed close to 6 percent, and Arm Holdings dropped more than 6 percent on the news.
The gap was not the product of fraud, aggressive accounting, or company misdirection. It emerged from a structural incompatibility between how OpenAI and Anthropic — the two leading frontier AI labs — each recognize revenue from sales made through cloud partners like Microsoft Azure, Amazon Web Services, and Google Cloud. That incompatibility, rooted in ASC 606’s principal-versus-agent framework, means the two companies’ headline numbers are measuring fundamentally different things. Investors who tried to bridge that gap made a compounding arithmetic error that inflated OpenAI’s estimated revenue by roughly $20 billion.
OpenAI Records 20 Cents Where Anthropic Records a Dollar
The technical accounting distinction is precise. Under ASC 606 — the United States accounting standard that governs when and how revenue is recognized — a company records the gross value of a transaction when it acts as the “principal”: the party primarily responsible for delivering the product and bearing the risk of the transaction. It records only its net share — the amount it actually retains after paying partners — when it acts as an “agent.”
OpenAI and Anthropic take opposite positions on this question when customers buy AI access through cloud marketplaces.
When a customer purchases $1 worth of tokens through Microsoft Azure, OpenAI records only its roughly 20-cent share of that transaction as revenue — Microsoft keeps the remainder. Anthropic, by contrast, records the full $1 as top-line revenue, then lists the cloud partner’s cut as a cost of sales, reasoning that Anthropic is the principal in the transaction regardless of which marketplace delivered it.
The difference is not cosmetic. Semafor reported in April 2026 that this accounting divergence alone — before any growth-rate miscalculation — accounts for a gap of “up to $8 billion in annualized revenue” between the two companies. Two companies with identical underlying businesses could show a reported revenue difference of up to $8 billion simply because they drew the principal-versus-agent line in different places.
How Investors Turned an $8 Billion Accounting Gap Into a $20 Billion Error
The $8 billion structural gap was not the endpoint — it was the starting point for a compounding error.
OpenAI communicated to investors in late summer 2026 that its annualized revenue had grown more than 70 percent. That figure was accurate and applied to a roughly $40 billion baseline confirmed by Bloomberg in August 2026. But some investors, attempting to make OpenAI’s net-revenue figure comparable to Anthropic’s gross-revenue figure, applied Anthropic’s methodology when extrapolating the growth rate. The result was a calculation that arrived at approximately $70 billion.
Axios reported that $70 billion figure on September 29, 2026, citing an anonymous source. It subsequently circulated widely among investors and across financial media. A source familiar with the investor documents confirmed to multiple outlets — including CNBC — that the $70 billion figure never originated from OpenAI itself. The Financial Times report, citing documents shared directly with OpenAI’s financial backers, established that the actual September figure approached $50 billion.
At $50 Billion, OpenAI’s Revenue Trajectory Remains Extraordinary
Strip away the accounting noise and OpenAI’s underlying growth figures remain remarkable by any conventional measure.
OpenAI CFO Sarah Friar confirmed in a mid-January 2026 blog post that the company’s annualized recurring revenue topped $20 billion at the end of 2025. By February or March 2026, that run rate had climbed to approximately $25 billion. Bloomberg reported in August 2026 that it had surpassed $40 billion — roughly doubling in eight months — driven by growth in coding products, enterprise subscriptions, and an expanding advertising business. The corrected September figure of approximately $50 billion represents meaningful continued growth from that $40 billion baseline.
The concern Thursday’s report surfaced is not about absolute performance. It is about the gap between what the market had priced in and what the underlying figure actually supports. OpenAI’s valuation soared to $852 billion following a $122 billion fundraise in March 2026. The company is currently negotiating a fresh private round that could value it at roughly $1.4 trillion. At those figures, every data point in the revenue trajectory carries outsized weight.
Why Anthropic Now Appears to Lead in Revenue Despite Smaller Real Market Position
One of the more striking consequences of the correction: on reported headline figures, OpenAI now appears to trail Anthropic. Anthropic reported an annualized run rate of $65 billion at the end of July 2026. At OpenAI’s corrected $50 billion net-revenue figure, that headline gap appears significant.
The irony is direct. The very attempt by investors to close the apples-to-oranges comparison — by restating OpenAI’s net revenue using Anthropic’s gross methodology — is precisely what produced the inflated $70 billion estimate that later required correction.
When analysts apply consistent accounting assumptions across both companies, the picture shifts materially. Adjusting Anthropic’s $65 billion figure to net-revenue treatment — netting out cloud-partner revenue shares that Anthropic books at gross — reduces Anthropic’s reported lead considerably. The two companies may be far closer in underlying commercial performance than either set of headline figures suggests. What is not in dispute is that using one company’s gross figure alongside the other’s net figure as a direct comparison was, from the outset, a methodological error.
Does Revenue Correction Change Math on OpenAI’s $1.4 Trillion Valuation?
OpenAI had been widely expected to file for an IPO by the end of 2026. A confidential S-1 was submitted to the SEC in late May 2026, with news of the filing breaking publicly in early June. But CEO Sam Altman has since publicly ruled out a 2026 listing, citing concerns about the pace of AI safety research and the company’s readiness for public-market scrutiny. A 2027 listing is now the prevailing expectation, with OpenAI reportedly targeting a valuation above $1 trillion for its public debut.
A public offering is typically priced as a multiple of revenue. When the revenue denominator in that equation has just been marked down by $20 billion, justifying the upper end of a trillion-dollar-plus valuation range requires either a higher revenue multiple or accelerating evidence of re-growth. OpenAI has reportedly told potential investors it expects to generate $280 billion in annual revenue by 2030 — a projection that requires compounding at rates most analysts consider optimistic even under the most favorable AI adoption scenarios. Those long-range numbers now enter the conversation against the backdrop of a near-term miss.
OpenAI has not issued a detailed public response to the Financial Times reporting. Its current fundraising discussions at a $1.4 trillion valuation appear to be continuing, suggesting that private-market investors are treating the revenue correction as a recalibration rather than a fundamental reassessment of the company’s long-term trajectory. The distinction matters: recalibrations affect near-term comps. Reassessments affect the discount rate applied to future cash flows.
Sam Altman Has Called for International AI Oversight While Negotiating Record Valuations
The revenue correction arrives at a moment when OpenAI’s own CEO has been among the most prominent voices calling for external constraints on the industry he leads.
In May 2023, Altman testified before Congress on AI licensing, urging lawmakers to establish a federal licensing regime for advanced AI systems before a Senate Judiciary Subcommittee hearing. In February 2026, Altman called for urgent global AI oversight, proposing an international oversight body with meaningful enforcement capability — a framework explicitly modeled on existing nuclear and biosafety institutions — at the AI Impact Summit in New Delhi.
OpenAI CFO Sarah Friar has focused her public commentary primarily on financial performance and operational metrics. No documented technology-policy positions have been identified for Friar in public reporting — a notable absence given her senior role at a company that has become a focal point of global AI governance debate.
The juxtaposition carries significance for a TechTimes audience: the company simultaneously negotiating a $1.4 trillion private valuation and presenting a genuinely accurate but widely misunderstood revenue figure is led by an executive who has argued, on the record and before Congress, that no single company — his own included — should develop powerful AI without external oversight.
What Markets Repriced When OpenAI’s Revenue Fell $20 Billion Short
Thursday’s selloff extended well beyond OpenAI, illustrating how concentrated the AI growth narrative has become around a single company’s revenue trajectory.
Nvidia fell nearly 3 percent. Oracle, which holds a reported $300 billion cloud services agreement with OpenAI spread over five years, shed close to 6 percent. Oracle carries roughly $130 billion in total debt as of its most recent disclosures — a debt load that makes investors acutely sensitive to any signal that OpenAI’s ability to absorb that compute capacity at the projected pace may be softer than expected.
Arm Holdings — a chip designer whose business model is tied broadly to AI inference demand — dropped more than 6 percent despite the company’s customer base extending well beyond OpenAI. SpaceX saw its shares fall roughly 4 percent in private-market trading.
The depth of the reaction illustrates the structural problem underlying the current AI investment cycle: trillions of dollars in data center construction, chip procurement, and energy infrastructure have been committed on the premise that frontier AI demand — anchored primarily by OpenAI — will compound fast enough to service those capital commitments. Every time a top-line figure comes in below what the market had priced, the question resurfaces of whether the buildout is running ahead of the monetizable demand curve. The $20 billion correction does not answer that question. It intensifies it.
Frequently Asked Questions
Why was the $70 billion revenue figure wrong if OpenAI communicated 70 percent growth?
OpenAI communicated accurately that its revenue had grown more than 70 percent from a roughly $40 billion August 2026 baseline. The $70 billion figure was not reported by OpenAI — it was first reported by Axios on September 29, 2026, citing an anonymous source. It emerged from investors attempting to restate OpenAI’s net-revenue growth rate using Anthropic’s gross-revenue methodology, which counts the full value of cloud-partner sales rather than only OpenAI’s retained share. When the Financial Times obtained actual investor documents, the September figure was approaching $50 billion — accurate under OpenAI’s own accounting approach, which was never the basis for the $70 billion projection.
What is the practical difference between gross and net revenue accounting for AI companies?
ASC 606, the US revenue recognition standard, requires companies to determine whether they act as a “principal” or “agent” in each transaction. A principal records the full transaction value; an agent records only its retained share. For OpenAI, a $1 cloud-partner sale produces approximately 20 cents of recorded revenue; for Anthropic, the same $1 sale produces $1 of recorded revenue, with the cloud partner’s fee reported separately as a cost. Semafor has confirmed this produces a structural gap of up to $8 billion in annualized revenue between the two companies even before growth rates enter the calculation. Neither treatment is wrong — they reflect genuine disagreements about each company’s role in the transaction — but they cannot be compared directly without adjustment.
Is OpenAI’s actual revenue understated, or is the $50 billion figure accurate?
Both companies are reporting accurately under their respective accounting choices. OpenAI’s $50 billion represents what the company actually retains from its sales — its net economic share. Anthropic’s $65 billion represents the gross transaction value flowing through its products before cloud-partner costs. Both numbers are internally consistent and prepared under recognized accounting frameworks. The problem is not misreporting; it is the comparison. Investors who equated the two figures without applying an accounting adjustment introduced a structural error that compounded with the growth-rate calculation to produce the $70 billion estimate. The $20 billion gap is a measurement-methodology artifact, not hidden revenue.
What does the revenue correction mean for OpenAI’s IPO timeline and valuation?
OpenAI CEO Sam Altman has already ruled out a 2026 IPO, with a 2027 listing now the prevailing expectation. An IPO priced as a revenue multiple becomes more difficult to defend at the upper end of a $1 trillion-plus range when the revenue denominator has been marked down by $20 billion. OpenAI’s $280 billion revenue projection for 2030 requires compounding well beyond analyst consensus even on the original figures; the correction narrows the range of defensible multiples for the current $1.4 trillion private-round valuation. Private-market investors appear, for now, to be treating the correction as a recalibration rather than a reassessment — but the IPO math has become harder.
