OpenAI’s $20 Billion Vanishing Act, and the Afternoon Wall Street Panicked

Thursday afternoon started like any other trading day, and then a single number from the most important private company on Earth knocked the legs out of the market. The Nasdaq fell 1.4%. The S&P 500 dropped 0.7%. The Dow dipped. Billions of dollars of stock market value evaporated in a few hours, and the trigger was not a bank, a central bank, or a war. It was a revenue figure from OpenAI that came in about $20 billion short of what investors had been led to expect.

That sentence sent stocks tumbling for a simple reason.

The number that broke the afternoon

The Financial Times, citing financial documents shared with investors, reported that OpenAI’s annualized revenue is roughly $20 billion below what had previously been signaled. To understand why that sentence sent stocks tumbling, you need to know how fast the signal itself has been moving.

Late last month, reports based on information given to investors put OpenAI’s annualized revenue run rate near $70 billion. Then the company itself reportedly told investors the figure was closer to $50 billion at the end of September. Now the FT says the actual documented number is about $20 billion below what was signaled, a gap it described as likely to damp optimism about the growth of AI demand.

Note the careful language here. Nobody said OpenAI’s revenue is falling. A run rate is a snapshot: recent monthly revenue multiplied to look like a full year. The problem is not the direction. It is the distance between what the market believed and what the paperwork apparently says.

When you are valued at $852 billion in a secondary share sale, as OpenAI was in August when $7 billion of shares changed hands, your growth story is not allowed to have plot holes. And when you are reportedly talking to investors about a fresh funding round near a $1.2 trillion valuation, a $20 billion question mark is a loud question mark.

Why Wall Street took it personally

The reason the Nasdaq cared so much about one private company’s paperwork is simple: the AI trade is no longer a trade. It is the market. Trillions of dollars of public company value now rest on the assumption that AI demand is real, growing, and profitable enough to justify the most aggressive capital spending program in corporate history.

So when doubt appears at the top of the chain, it travels fast. Here is what Thursday afternoon looked like, by the numbers reported in the afternoon coverage:

• The tech-heavy Nasdaq Composite fell 1.4%, with the S&P 500 down 0.7% and the Dow slipping modestly.
• The information technology sector led the decline inside the S&P 500, falling 2%.
• The PHLX Semiconductor Index, the chip sector’s report card, dropped 4.2%.
• AI infrastructure names like Coherent, Bloom Energy, Vistra, Lumentum, Super Micro Computer and Intel were among the worst performers in the S&P 500.
• The Roundhill Memory ETF, already weak after Samsung fell in South Korean trading, dropped 4.5%. The iShares Semiconductor ETF fell 3.5%. Even the Magnificent Seven ETF lost about 1%. Arm, Intel and Marvell were among the biggest Nasdaq decliners.

That list tells its own story. This was not a broad risk-off afternoon. It was a referendum on the AI demand narrative, and the vote was no confidence.

The burn rate nobody wants to talk about

Here is the uncomfortable math that has been sitting in plain sight for months. Leaked financials reported earlier this year showed OpenAI losing about $20.9 billion from operations in 2025 on roughly $13.07 billion of revenue. The company lost more money than it made, and its valuation kept climbing toward a trillion anyway.

It gets spicier. The Information has reported that OpenAI’s losses could reach $14 billion in 2026, with cumulative losses from 2023 through 2028 projected around $44 billion, and profitability not expected until 2029. The company has raised its projected cash burn through 2029 to about $115 billion. These are staggering numbers for a company whose entire valuation depends on the belief that revenue will eventually catch up with spending.

That belief is not crazy. OpenAI grew from around $2 billion in 2023 revenue to more than $20 billion in annualized run rate by the end of 2025. ChatGPT has more than a billion users by some estimates. The growth is real. But growth at this price is not measured in the billions. It is measured in the gap between the story and the paperwork, and Thursday made that gap visible.

Sam Altman has confirmed the company will not go public this year. CFO Sarah Friar has told employees the company will be a public company in 2027, possibly sooner. That timeline matters, because a public offering means audited financials, quarterly earnings calls, and the kind of scrutiny that makes a $20 billion expectations gap impossible to hand-wave away. The clock is ticking.

The Anthropic problem nobody at OpenAI mentions

The revenue story has a rival-shaped hole in it. Back in May, Anthropic said its annualized revenue run rate had surpassed $47 billion. That is a remarkable number, and here is the part that should worry OpenAI’s bankers: roughly 80% of Anthropic’s revenue reportedly comes from business customers, from companies running Claude Code, enterprise agents, and API-based automation.

Business customers are the good kind of revenue. They pay more, sign longer contracts, and churn less. OpenAI’s revenue, by contrast, is still dominated by humans paying for chat: ChatGPT subscriptions are reportedly the largest single bucket. There is nothing wrong with consumer subscriptions. But a trillion-dollar valuation built on $20-a-month subscriptions needs either an enormous number of subscribers or a faster pivot to enterprise, and the competition is not waiting.

Google’s Gemini surged in late 2025 with everyday users. Anthropic’s coding tools have grown at a pace that surprised even industry veterans. And the market just got a reminder that when it comes to OpenAI’s actual numbers, everyone outside the company is essentially reading tea leaves from investor documents and media leaks.

This market has been here before

Thursday was not the first time OpenAI’s numbers have shaken the tape. Back in May, the Wall Street Journal reported that CFO Sarah Friar had raised concerns inside the company about its ability to pay for future computing contracts if revenue did not grow fast enough, and that OpenAI had missed some internal targets for new users and revenue.

The market’s reaction then was almost identical to now. Shares of OpenAI partners fell: Oracle dropped 4% on a reported five-year, $300 billion cloud computing arrangement, and CoreWeave fell nearly 6% on its $11.9 billion OpenAI contract. Friar and Altman jointly rebutted the report, saying claims of division or pullbacks in computing investment were ridiculous and that they were aligned on buying as much compute as possible.

That last detail is the real fault line. OpenAI’s strategy is to spend its way to the future: more compute, bigger models, more data centers, more chips. The company needs something like $115 billion of cash through 2029 to keep doing it. That strategy works brilliantly as long as revenue keeps the story credible. The moment the story wobbles, every chip stock, every data center REIT, and every AI infrastructure name inherits the wobble.

Read our earlier coverage of how the AI infrastructure boom is reshaping the tech world in our Technology section, including Google’s billion-dollar nuclear bet on AI data centers and SpaceX’s hunt for $40 billion to buy Nvidia chips.

What actually happens next

A single afternoon selloff does not end the AI boom, and anyone telling you otherwise is selling something. Run rates bounce around. Investor documents get misread. And OpenAI has real, enormous, growing revenue by any historical standard. The company has gone from $2 billion to tens of billions in annualized revenue in about three years. Most startups would kill for that trajectory.

But that is not the game being played anymore. The game is a private company valued near a trillion dollars asking the world to finance the biggest capital expenditure program in history on the strength of numbers that just turned out to be $20 billion softer than expected. In that game, perception is the product, and Thursday afternoon was a perception event.

Watch what happens next: whether OpenAI disputes the FT’s reading of the documents, whether any of the dozens of AI infrastructure names that fell recover, and whether the company’s rumored $1.2 trillion funding round materializes at the price being discussed. And watch 2027. That is when OpenAI is expected to go public, when the audited numbers finally land, and when the expectations game gets replaced by earnings calls.

Twenty billion dollars is not a rounding error when you are asking the world to fund the next industrial revolution. It is, at minimum, a reminder that the AI boom runs on two commodities: compute and confidence. Thursday showed which one is harder to manufacture.