OpenAI's $20 Billion Revenue Gap Is a Counting Dispute That Cost Nvidia $169 Billion

Nobody lost a dollar of OpenAI revenue this week. The number that moved was a label on it. That was enough to take roughly $169 billion off Nvidia's market value in one session, which tells you how much of the AI supply chain is sized to figures nobody outside the company can audit.

What Actually Happened

On Thursday, Reuters relayed a Financial Times report, citing financial documents shared with investors, that OpenAI's annualized revenue is about $20 billion less than previously signaled. Per Reuters, OpenAI recently told investors the figure was approaching $50 billion at the end of September, far short of the $70 billion that media outlets, including Reuters, reported late that month. Reuters said it could not independently verify the report, and OpenAI did not immediately respond to a request for comment.

The FT's explanation, as summarized by Reuters and Mobile World Live, is a definitions problem. The discrepancy arose from attempts by OpenAI's own investors to produce a direct comparison with Anthropic. The two labs calculate the figure differently: Anthropic includes revenue from sales through cloud partners such as AWS and Google Cloud, and OpenAI does not. Mobile World Live adds, citing the FT, that OpenAI told investors its direct revenue still grew by more than 70% in the third quarter alone.

On Friday, Bloomberg reported, per PYMNTS, that OpenAI expects annualized revenue of at least $70 billion by the end of the year, up from $50 billion at the end of September. Bloomberg's sources were unnamed, the growth was attributed to the enterprise business, and the figures were shared with investors during fundraising discussions. OpenAI declined to comment.

The sequence, as the reporting lays it out:

• September 29: Reuters, citing a source, and Axios reported OpenAI's annualized recurring revenue was approaching $70 billion, driven by a more than twofold increase in enterprise sales since July.
• October 8: The FT reports the investor documents show about $50 billion, on OpenAI's own counting basis.
• October 8, market close: Chip and AI-infrastructure stocks fall. Kiplinger's market wrap headlines the Nasdaq dropping 345 points as OpenAI revenue concerns weigh on tech.
• October 9: Bloomberg reports the $70 billion target for year-end.

The Market Priced a Definition as a Demand Signal

According to 24/7 Wall St., Nvidia fell 2.94% to $230.48 on October 8, erasing roughly $169 billion of market value after the report. Other suppliers fell harder: CoreWeave dropped 7.77%, Oracle lost 5.48%, Broadcom fell 4.35% and AMD declined 3.90%. A gauge of chip stocks sank 3.4%, while the S&P 500 lost only 0.47%. The gap between those two numbers is the point. This was a sector-specific reaction, not a broad risk-off day.

Nvidia's one-day loss was larger than the entire revenue gap by a factor of eight, which is a clumsy comparison, because market value and annual revenue are different things. But it shows what investors were really testing. They were not asking whether OpenAI has $50 billion or $70 billion in revenue. They were asking whether the buyers behind multi-year chip, cloud and power commitments are growing as fast as the headline said.

24/7 Wall St. notes that the higher figure came from OpenAI's own investors, that OpenAI never releases the number itself, and that no sales were actually lost. Its read matches the FT's: the same demand produces different totals depending on whether partner-channel sales are counted.

One caution on the growth claims. Axios, per PYMNTS, put the late-September pace at a 70% increase since the start of the third quarter, with B2B revenue up more than 100% over the same period. 24/7 Wall St. cites an unnamed source describing an investor presentation showing 77% run-rate growth in the quarter, and says nobody has confirmed it. The FT's version is that direct revenue grew by more than 70%. Those figures are in the same range but come from different sources and different bases. Treat the range as reported, not as an audited fact.

Who Is Most Exposed to the Number

24/7 Wall St. ranks the damage by how much each company's story depends on a few buyers:

• CoreWeave fell the most. Its value rests on a $104 billion backlog funded largely with debt, and second-quarter interest expense reached $640 million, per the same report.
• Oracle holds $664 billion in remaining performance obligations, meaning contracted revenue it has not yet recognized. It has not disclosed how much comes from OpenAI.
• Broadcom shipped OpenAI's first custom accelerator and sees a path to more than 5 gigawatts of OpenAI installations in 2028, per 24/7 Wall St. Management also said OpenAI's financing position is less clear than that of its other large lab customer.
• Nvidia has OpenAI commitments of about 12 gigawatts, against $96.221 billion of revenue in its latest quarter, according to the same source.

This is the structure we flagged earlier in the week in our look at SpaceX's chip financing: the buyer carries the obligation and the vendor books the revenue. When the buyer's revenue is the unaudited variable, the vendor's valuation inherits that uncertainty. Nvidia has also been a backer of OpenAI's own build-out, as in the Ohio data center campus, which makes it both supplier and stakeholder in the same number.

Why the Anthropic Comparison Is the Real Story

The reason the gap exists is that two private companies are now being benchmarked against each other by their investors. Mobile World Live cites CNBC reporting that Anthropic's annualized run rate hit $65 billion by late July 2026. If that figure includes AWS and Google Cloud partner sales, as the FT says Anthropic's does, then a side-by-side with OpenAI's $50 billion is not like for like. Neither company has published audited statements, so no outside party can restate one lab's number on the other's basis.

Reuters adds a useful caveat on the metric itself: annualized run rate is sometimes a misleading sales measure, often calculated by multiplying one month's revenue by 12, though it has become a popular yardstick among fast-growing Silicon Valley startups. A strong month, a seasonal spike or a large one-off enterprise contract can all inflate it. Counting rules for partner revenue add another layer.

Both companies may soon face a stricter standard. Reuters notes that OpenAI and Anthropic are potentially preparing to go public, which would give Wall Street a clearer view of their finances. Mobile World Live reports Anthropic is targeting an IPO in November, after the US midterm elections, and OpenAI a listing next year. PYMNTS recalls that Sam Altman told Fortune on September 11 that the IPO will not happen in 2026. Until a filing lands, the audited number does not exist.

The Year-End Target Does Not Settle It

Bloomberg's $70 billion year-end figure sounds like a recovery. It may not be one. If the $70 billion is on OpenAI's own counting basis, the jump from $50 billion in September is 40% in one quarter, which is real growth but slower than the 70%-plus pace reported for the third quarter. If it is on a different basis, the two numbers are not comparable at all. The PYMNTS summary of the Bloomberg report does not say which, and we have not seen the original.

PYMNTS also notes Bloomberg's earlier report that OpenAI hopes to raise $30 billion in new funding at a valuation of roughly $1.4 trillion, up from the $852 billion valuation set in a March round that raised $122 billion. A revenue number shared with investors during a raise is a negotiating document. It is also the number the market will now ask OpenAI to define.

What Builders Should Take From It

• Check which definition a revenue number uses before you cite it. Run rate, ARR, direct revenue and partner-inclusive revenue are different measures. Put the basis in the footnote whenever you quote a competitor's figure.
• Treat private-lab revenue as a reported estimate. The $70 billion and the $50 billion both came from unaudited sources. Do not build a financial model on a single run-rate headline.
• Map your vendor concentration. If your roadmap depends on one lab's pricing, capacity or solvency, a one-day supplier selloff shows how fast sentiment can move. Keep a second model provider tested and ready.
• Watch the infrastructure names for stress. CoreWeave's debt-funded backlog and Oracle's contracted obligations are where a demand disappointment would show up first. If you buy GPU capacity from either, check contract flexibility.
• Wait for audited numbers on IPO decisions. An S-1 from either lab will settle the definitions. Until then, make commitments that can be resized.

Developer312 covers the AI business signals builders actually need to act on. Get the weekday briefing at developer312.com.