AI-linked equities were back in favor on Friday morning after a bruising Thursday session, with Nasdaq futures up about 0.8% and chipmakers including Nvidia, AMD and hyperscalers like Oracle and Microsoft recovering part of the ground they lost on an alarming window into OpenAI’s real revenue. Nvidia was up roughly 1.6% in premarket action, having fallen almost 3% the previous day to $230.48 as the Financial Times reported OpenAI’s annualized revenue was closer to $50 billion at the end of September than the $70 billion figure investors had been circulating a month earlier.
That $20 billion gap, on paper, is a big number for a single company with no public financials. What the FT story actually describes, on closer reading, is an accounting difference more than a demand signal. The $70 billion figure that circulated in late September counted gross revenue OpenAI generates through cloud partners, including what end-customer spending flows through partners like Nvidia on chips and infrastructure costs. The $50 billion figure strips those partner-linked amounts out, giving a cleaner picture of OpenAI’s direct commercial revenue.
“Markets have been given another AI relief impulse, but the underlying structure remains fragile,” TickMill Group strategist Patrick Munnelly said as futures rallied and investors worked through the accounting explanation.
The numbers on both sides
OpenAI’s annualized revenue, meaning recent sales projected forward across a full year, hit nearly $50 billion at the end of September on the FT’s reading. That is still an enormous growth figure for a company that registered roughly $12 billion in revenue a year earlier, in calendar 2025. It also puts OpenAI ahead of Anthropic, the rival lab that reached $65 billion in annualized revenue at the end of July on a comparable direct basis, and roughly matches or beats Stripe, Salesforce and a handful of other mature software companies in annual scale.
The problem is that investors track OpenAI’s revenue as the clearest single figure validating a build-out of data centers, chips and infrastructure running into the hundreds of billions of dollars. When the headline moves down by $20 billion in a single week, circuits short. On Thursday the Philadelphia Semiconductor Index dropped 3.4%, with Nvidia down 2.9%, AMD off 3.9%, Broadcom losing 4.35% and Oracle falling 5.5%. CoreWeave, the most OpenAI-levered of the AI cloud providers, fell with them. Even Microsoft, whose exposure is wider and whoseOpenAI deal provides contract stability rather than upside optionality, lost 1.35%.
| Company | Thursday move | Friday premarket | Why the reaction |
|---|---|---|---|
| Nvidia (NVDA) | Nearly -3%, to $230.48 | Up about 1.6% | Biggest single shareholder of the AI buildout; on-recordanger over slowdown fears |
| Oracle (ORCL) | -5.5% | Recovering | Holds a $300 billion OpenAI cloud contract and $664 billion total backlog; most directly exposed to OpenAI’s spend |
| Broadcom (AVGO) | -4.35% | Recovering | Building custom AI chips with OpenAI and reportedly arranging over $50 billion in financing to fund it |
| AMD | -3.9% | Recovering | Six-gigawatt infrastructure deal with OpenAI announced this autumn |
| CoreWeave (CRWV) | Down roughly 8% | Recovering | Backlog depends on OpenAI cloud demand; carries the most leverage of the AI infrastructure names |
Accounting gap vs. demand drop
What actually differentiated Thursday’s selloff from a routine AI wobble was the speed with which analysts pushed back. The popular read on social media was that OpenAI had somehow lost $20 billion in weeks. Within hours of the FT story, data points circulated showing the earlier $70 billion figure was a different way of counting the same business, not a step down in economic substance. CNBC reported that a source familiar with the matter said the higher number counted gross revenue from OpenAI’s corporate partnerships, including sales through Nvidia-linked commercial arrangements.
Wall Street’s own research desks picked this up quickly. The comparison to Anthropic is instructive: Anthropic’s $65 billion figure tracks direct commercial revenue, not grosspartner-billed infrastructure spend. Once OpenAI’s $50 billion is sourced the same way, the two labs are closer in scale than the raw numbers suggest, which is the opposite of the conclusion the raw gap pointed toward.
OpenAI’s growth path also matters. The company has been raising at a $1.4 trillion valuation, and the FT story noted that any new round could rely on the strength of the revenue figure. A downgrade in the headline matters more to that raise than it does to the underlying business, and the fundraising process still has months to run, so Friday was never going to be the last word on the story.
What the AI buildout still needs to prove
Friday’s rebound does not erase the question the FT report brought back into frame: whether massive infrastructure spending can keep clearing its own hurdle rate. OpenAI, Anthropic, Google, Meta and a handful of other AI leaders have signed hundreds of billions of dollars of compute and infrastructure contracts. Some of those are obligations, some options, some non-binding letters of intent. Whatever the exact legal form, someone has to pay for the buildout when it comes.
Nvidia and the AI hardware ecosystem sit at the center of that bet. Nvidia’s own valuation, which peaked near $243.37 in October before Thursday’s pullback, embeds assumptions about continued hyperscaler demand and sustained revenue growth straight off Blackwell and Rubin architectures. Micron and HBM suppliers are part of the same chain. So are the data center REITs and power providers feeding those campuses. A wobble in OpenAI’s trajectory, even one driven mostly by accounting, reassures nobody in that chain that the demand curve is guaranteed to hold.
For OpenAI itself, the disclosure race with Anthropic continues. Anthropic reportedly arrived at $65 billion annualized by July and is preparing for a public listing, possibly in November, that would force hard auditing of its own financials. OpenAI has said it has no plans to go public this year, but the Securities and Exchange Commission filing pressure and the mechanics of raising at a trillion-dollar-plus valuation mean the numbers will get pinned down one way or another.
Where the trade went next
Futures recovered on Friday on the argument that the revenue gap was accounting, not economics. Dip buyers have a reasonable case if the first reading was wrong. But the underlying question, whether AI infrastructure obligations can generate a return on the capital deployed, is the same question the market was asking last week and will ask again next week. One wobbly revenue figure did not answer it, and neither did a Friday bounce.
The Nvidia chart itself is a useful shorthand. The stock traded at $230.48 after Thursday’s drop, roughly 5% below its $243.37 all-time high, and remains ahead of the broader market on the year. Its forward P/E of around 29 times, with revenue growth running north of 100%, is not stretched by historical dot-com standards. But the concentrated exposure of the AI trade, with five or six names driving most of the S&P 500’s return this year, means every single-day move in one of them carries an outsized index impact. That is, in short, why one OpenAI number could swing $170 billion out of Nvidia’s market cap in a single session without Nvidia itself shedding any underlying business.