The AI trade is so crowded that an accounting clarification can tank the Nasdaq. On October 8, a Financial Times report noting that OpenAI told investors its annualized revenue was “close to $50 billion” — not the roughly $70 billion figure that had been circulating — sent Oracle down more than 5%, AMD off over 4.6%, NVIDIA down nearly 2.8%, and dragged the Nasdaq to a 1.4% afternoon loss. Microsoft closed lower too. That’s a lot of damage from a number that, by most reasonable reads, was never quite right to begin with.
As reported, the confusion traces back to how investors were benchmarking OpenAI against Anthropic. Anthropic counts all sales processed through cloud partners like Amazon and Google in its revenue figures. OpenAI reports on a net basis, recording only what it actually keeps. To make the two companies comparable, investors started adjusting OpenAI’s numbers upward. When OpenAI confirmed revenue growth had exceeded 70% over a recent period, analysts applied that growth rate to an already-inflated August baseline of around $40 billion, and the math produced a $70 billion figure. OpenAI’s own internal numbers, though, have consistently sat near $50 billion.
So where does the gap come from? According to CNBC, the $70 billion figure includes revenue-sharing payments from major cloud partners, primarily Microsoft. OpenAI reportedly pays Microsoft around 20% of revenue under an agreement that runs through 2030. Strip that out, and you get to $50 billion. Some analysts would argue that gross revenue is actually the better lens for evaluating an AI model company, because that revenue-sharing agreement has an expiration date. Once it lapses, gross and net figures converge, and the whole discrepancy disappears. Jim Cramer, who has been bullish on AI infrastructure demand, called the report a “nothing burger” and said the market would shake it off quickly.
He may be right on the fundamentals. But the market reaction still tells you something important. OpenAI has become the central load-bearing beam of the AI capital spending narrative. Oracle’s cloud contracts, NVIDIA’s GPU pipeline, AMD’s data center ambitions, Microsoft’s AI monetization story — all of it is tied, to varying degrees, to how fast OpenAI grows. When that anchor sends any signal that looks like a downward revision, even a technical accounting one, money moves fast. That’s not irrational exactly, but it does reflect how much of current AI stock valuations are built on forward projections rather than verified cash flows.
The valuation math gets uncomfortable when you look closely. OpenAI is reportedly seeking at least $30 billion in new funding at a valuation as high as $1.4 trillion. At $50 billion in annualized revenue, that’s a price-to-sales multiple of roughly 28x. At $70 billion, it’s about 20x. Neither is cheap. Broadcom is also reportedly arranging more than $50 billion to develop custom silicon for OpenAI, which is a meaningful signal of where the infrastructure spending is headed regardless of which revenue number you use.
The IPO picture has also shifted. On prediction market Polymarket, traders now put roughly a 58% probability on OpenAI going public before the end of 2027 with a first-day market cap of at least $1.5 trillion. That’s down from around 80% in September. The probability of no IPO before 2028 has climbed to about 23%.
The underlying demand for AI compute, storage, and optical components hasn’t disappeared. But the market’s willingness to price in best-case scenarios without scrutiny is clearly shrinking. The $50 billion versus $70 billion debate was always partly a distraction. The real question is whether OpenAI can justify any of these multiples with durable, profitable growth — and that answer isn’t in the ARR figure either way.
