Skip to content
Markets & Trends

OpenAI's Revenue Was $20bn Smaller Than The Market Thought, And Chip Stocks Lost 3.4% In An Afternoon. The Gap Is Accounting, Not Demand - Which Is Exactly Why It Matters

On 8 October the Financial Times reported that OpenAI had told investors its annualised revenue was approaching $50bn - not the roughly $70bn that had circulated in the press and in investors' models. Within hours the Nasdaq was down 1.25%, a gauge of chip giants had fallen 3.4%, Oracle had lost more than 5%, CoreWeave about 8%, and Nvidia and AMD 3-4%. The twist is that very little changed about OpenAI's business. The $70bn figure appears to have come from investors comparing OpenAI with Anthropic, which books the full value of sales made through cloud partners as revenue, while OpenAI books only its share. One company reports gross, the other net. This is the story of how a definitional difference moved hundreds of billions of dollars of market value, what it says about how fragile the AI trade has become, and what every firm that feeds AI metrics into models, ratings and signals should do about it.

AlchmAI Editorial13 min read

~$50bn

OpenAI's annualised revenue at the end of September, as it told investors - against roughly $70bn previously reported (FT, 8 October)

-3.4%

A gauge of chip giants on 8 October, with the Nasdaq down 1.25% to 27,193 and the S&P 500 down 0.47%

-8%

CoreWeave on the day; Oracle fell more than 5% and Nvidia and AMD roughly 3% and 4%

Gross vs net

The accounting difference behind most of the gap: Anthropic books cloud-partner sales gross, OpenAI books only its share

The headline was simple enough to move a market. On Thursday 8 October the Financial Times reported that OpenAI had told investors its annualised revenue was approaching $50bn at the end of September. The number most investors had been carrying - including in earlier reporting - was about $70bn. A $20bn hole in the revenue of the company at the centre of the AI build-out landed in an already nervous market, with the 10-year Treasury yield at 5.23%, Brent crude above $100 and the AI complex priced for perfection after Nvidia's record $5.7 trillion valuation the week before.

The reaction was immediate. The Nasdaq Composite fell 1.25% to 27,193.34 and the S&P 500 0.47% to 7,765.36, while the Dow edged up. A gauge of chip giants dropped 3.4%, having been down as much as 4% intraday. Oracle - whose data-centre backlog is heavily tied to OpenAI - lost more than 5%. CoreWeave, the GPU cloud, fell about 8%. Nvidia and AMD lost close to 3% and 4%; Micron, Intel, Arm and Marvell each fell 5-6% or more. Palantir, upgraded by Goldman Sachs, was one of the few names to close higher.

Gross Versus Net, In Plain English

Accounting standards on both sides of the Atlantic - IFRS 15 and the US equivalent, ASC 606 - ask one question when a sale involves a middleman: who is the principal? If a company controls the product before it reaches the customer and is primarily responsible for delivering it, it is the principal and reports the full sale as revenue, with the middleman's fee as a cost. If it is effectively an agent arranging for someone else to provide the product, it reports only its net fee. Judgement is involved, and two companies selling similar things through similar channels can reasonably land on different sides.

  • Gross reporting makes revenue look bigger and margins look smaller. Net reporting does the opposite. Neither changes the cash the company keeps.
  • Comparisons across companies are only meaningful on the same basis. Comparing a gross number with a net number - as the $70bn estimate effectively did - produces a gap that is entirely definitional.
  • Growth rates are less affected than levels. Reports put OpenAI's total revenue growth at 77% and enterprise revenue growth at 107% in the third quarter - figures that do not depend on which convention is used.

So Why Did The Market Care So Much?

Because the AI trade is no longer priced on AI companies' revenue. It is priced on their commitments. OpenAI raised $122bn in March and has signed enormous multi-year compute agreements; Anthropic disclosed $518bn of commitments in its prospectus last week; Oracle, CoreWeave and the chipmakers have built their forecasts around those commitments being honoured. Bain estimated last month that the industry needs $6 trillion of annual revenue by 2031 to justify the data-centre build-out. Every one of those numbers is ultimately underwritten by a handful of customers' ability to grow revenue fast enough. A $20bn surprise in the denominator - even a definitional one - is a reminder of how concentrated that bet has become.

  1. 01Concentration: a large share of AI infrastructure revenue depends on very few tenants. A perception shift about one of them moves the whole complex.
  2. 02Leverage: suppliers are lending to customers to buy their products, as Broadcom's $42bn facility for Anthropic showed. That raises the stakes on every customer revenue figure.
  3. 03Rates: with the 10-year near 5.2%, long-duration growth stocks have less margin for error than at any point in two decades.
  4. 04Timing: OpenAI's IPO has reportedly slipped to early 2027 and Anthropic's roadshow is expected this month. Investors are about to get audited numbers for the first time, and are re-pricing the unaudited ones in advance.

“Nothing about OpenAI's customers changed on Thursday. What changed was the market's confidence that it knew what the numbers meant - and in a trade built on commitments, that is the thing that moves prices.”

The Lesson For Anyone Who Feeds AI Metrics Into Models

Every sell-side model, quant signal and stock-rating system that ingested '$70bn' was wrong in the same direction at the same time - not because anyone made a calculation error, but because a reported figure travelled without its definition. For firms that integrate private-company metrics, alternative data or AI-generated summaries into ratings and trading signals, that is the operational lesson of the week.

The Bottom Line

OpenAI's annualised revenue of about $50bn rather than the roughly $70bn the market had assumed knocked 3.4% off a gauge of chip giants, more than 5% off Oracle and about 8% off CoreWeave in an afternoon - yet most of the gap is the difference between Anthropic's gross reporting of cloud-partner sales and OpenAI's net reporting, with underlying growth still reported at 77% overall and 107% in enterprise. The episode shows how concentrated and commitment-driven the AI trade has become, and how fast a figure without its definition can move it. For investors and the firms that build their signals, the response is to track the basis and lineage of every AI metric as carefully as the metric itself. That is the real-time data and AI signal integration work we do as a fintech AI agency in London, and this week showed what it costs to skip it.

References & Further Reading

Markets & TrendsOpenAIAI stocksAI Signal & Stock Rating IntegrationFintech AI Agency LondonReal-Time Data Ingestion & Plottingrevenue recognition
Share Email
AI

AlchmAI Editorial

Research and analysis, London

The AlchmAI team writes about the markets, technology and regulation we work with every day. We build trading platforms, real-time charts and AI analysis tools for brokers, prop firms and fintech teams from our office in Mayfair, London. Every article lists its sources. Nothing we publish is investment advice.

This article is general information and commentary. It is not investment advice or a recommendation to buy or sell any investment. Important information

Keep reading

All articles