On October 9, 2026, one report put OpenAI’s annualised revenue run rate around $50 billion in September, while Bloomberg’s sources expected the company to reach or top $70 billion by the end of 2026 — a $20 billion disagreement about the same private company’s books, large enough to move public chip stocks.
The Financial Times’ account of the discrepancy is the interesting document. The gap, the FT detailed, stems from attempts to compare OpenAI’s figures directly with Anthropic’s, including how sales routed through cloud providers are accounted: revenue recognised gross or net, committed or consumed, the bookkeeper’s choices that become billions when the underlying machine is this large. Two honest spreadsheets can produce both numbers; that is precisely the problem. A private company with no filing obligations has become systemically important to the valuations of public ones — every GPU order forecast, every data-centre lease, every chip stock’s multiple leans on figures the market cannot audit.
Investors are learning that AI growth is real and still hard to measure on a single line. The demand is not in question; every measure of usage, from enterprise deployments to the utilities interconnecting new data centres, confirms it. What is unmeasurable, from outside, is the margin structure of the demand — whether $50 or $70 billion of revenue is being bought with $80 billion of compute, and whose balance sheet absorbs the difference.
The number is a market now
Until OpenAI files like a public company, its run rate will trade like a rumour with a stock ticker attached. NewsWibe’s Business & Technology Desk will follow the reported figures and the accounting debate they have exposed.
There is precedent for this condition: the railroads of the 1860s, the telecoms of the 1990s, each becoming too important to value and too private to verify at the same moment. The numbers eventually became public. So will these — the only question is whether the revelation arrives as an IPO prospectus or as an archaeology project.
