Every major chip order forecast for 2027 was anchored, at least in part, to OpenAI generating roughly $68 to $70 billion in annualized revenue. That number is wrong. OpenAI told investors it hit roughly $50 billion in annualized revenue at the end of September, lower than the $68 billion figure that was widely reported late last month. The Financial Times broke the story Thursday; CNBC confirmed it. By the close, the Nasdaq Composite fell 1.3% and the Philadelphia Semiconductor Index tumbled about 3.4%.
The real question sitting in every investment committee meeting this morning is not whether OpenAI lied. It did not. The roughly $20 billion difference comes down to how two rival AI labs count their money. OpenAI includes only its share of sales made through partners. Anthropic, by contrast, counts gross revenue from cloud partners such as AWS and Google Cloud. To compare the two companies on equal footing, investors had been normalizing OpenAI’s figures to match Anthropic’s method. That adjustment produced the approximately $70 billion estimates that floated around in late September. The question is what the correct number implies for the hardware orders sitting behind it.
The Bull Case
Start with what OpenAI actually reported alongside the revenue figure. In addition to the $50 billion in annualized revenue, OpenAI touted 77% total run rate growth during its third quarter, as well as 107% run rate growth for its enterprise business during the same period. That is not a company in retreat. As of late September 2026, OpenAI’s annualized revenue run rate sat at approximately $50 billion, up from over $40 billion in August. The trajectory still points sharply upward.
More importantly, OpenAI is one customer. The hyperscalers collectively are a much larger one. Spending on AI infrastructure by the five largest US hyperscalers is set to increase by more than half next year to $1.2 trillion, according to a Goldman Sachs note reported by Bloomberg in late September. Nvidia’s own order book, as CEO Jensen Huang stated at GTC in March, reflects $1 trillion in orders for the company’s most advanced processors through 2027. That backlog is not a projection built from OpenAI’s revenue. It is built from Microsoft, Amazon, Google, and Meta committing capital they have already announced.
The Bear Case
The problem is not one accounting line. It is what the $20 billion gap reveals about how loosely the market was modeling AI demand. If sophisticated investors misread OpenAI’s revenue by nearly 30% for weeks, the same methodological looseness has almost certainly infected broader demand forecasts. Despite projected multi-year expansions in global AI capital expenditures, slowing growth rates and accumulated sector gains mean robust demand alone can no longer sustain upward stock momentum.
There is a longer-term wrinkle. Analysts are already predicting that 2027 will be the year when AI capex starts to decelerate. A $20 billion revenue miss at the world’s most visible AI company, even if explained by accounting, hands that camp fresh ammunition heading into earnings season.
What Investors Are Missing
Thursday’s sell-off hit the entire chip complex, but the damage was not uniform, and that spread matters. Nvidia fell about 3% Thursday, Oracle fell about 5.5%, and CoreWeave slipped nearly 8%. Additionally, AMD fell about 4%, Broadcom fell about 4%, Intel fell about 5%, and Super Micro Computer fell nearly 5%. The steepest declines were concentrated not in the chip designers, but in the companies that borrowed heavily to build AI capacity and priced that debt against a demand curve drawn from the $70 billion figure.
The overlooked implication: the accounting gap that produced Thursday’s move is also the gap that will matter when OpenAI files for its IPO. OpenAI is under pressure to justify an $852 billion valuation. It confidentially submitted draft IPO documents to regulators on June 8, 2026, and executives have suggested a listing in 2027. Every infrastructure company priced as an OpenAI proxy, CoreWeave most visibly, will be priced again once public investors get a clean, audited revenue number.
Stocks to Watch
Nvidia (NVDA) fell about 3% Thursday but retains the most defensible position. Its order book spans all five hyperscalers. A single customer’s accounting method does not alter that. The stock is worth monitoring for entry if the broader chip complex continues to reset on sentiment rather than fundamentals.
CoreWeave (CRWV) is the highest-risk name in this debate. CoreWeave shares climbed above $90 earlier this week after JPMorgan upgraded the stock and raised its price target, citing stronger compute pricing and expectations for margin expansion. The stock then reversed sharply. Its debt load, insider selling, and dependency on a handful of large compute customers make it acutely exposed to any further downward revision in AI demand assumptions.
Broadcom (AVGO) carries a different risk profile. Its custom silicon business is tied to hyperscaler spending plans that were locked in before Thursday’s news. A sustained selloff driven by OpenAI sentiment, rather than any change in Alphabet’s or Meta’s capex commitments, could create a gap between price and underlying contract value.
Marvell (MRVL) and Corning (GLW) both fell more than 5% Thursday. Both serve the optical networking layer of AI infrastructure, a segment where demand is driven by data center build-out volumes, not by which AI lab has the higher revenue run rate. If the sell-off persists into next week, these are the names where the market is likely conflating correlation with causation.
