OpenAI has told investors its 2025 revenues will hit $50bn based on sales through September — a significant number by any measure, but $20bn less than the $70bn figure circulated to investors just one month earlier. The gap is not a business deterioration story. It is an accounting-definition story, and that distinction matters enormously for understanding where AI capital markets actually stand. The discrepancy traces to how revenue gets counted across the two leading AI labs. Anthropic, maker of Claude, includes revenue from sales routed through cloud partners — Amazon's AWS and Google Cloud — in its annualised figures. OpenAI does not. When OpenAI investors attempted to present an apples-to-apples comparison with Anthropic's $65bn forecast from the end of July, the inflated $70bn number entered circulation. The correction reveals not a demand shortfall but the absence of standardised revenue reporting in an industry attracting hundreds of billions in capital. Markets reacted as though the demand story itself had cracked. The Nasdaq closed down 1.4% on Thursday. Nvidia shed 2.9%, Oracle dropped 5.5%, and Micron fell 4.8%. These are real losses triggered by a definitional adjustment — a measure of how fragile sentiment is around AI revenue narratives. Investors are pricing these companies on growth trajectories that depend on which revenue line you read. The capital flowing into this sector is staggering regardless of the metric chosen. OpenAI is in early-stage talks to raise $30bn at a $1.4tn valuation. Its most recent round, in March, closed at $122bn on an $852bn valuation. Anthropic raised $65bn two months later at $965bn. SoftBank's Masayoshi Son is now seeking up to $100bn from Gulf states to scale AI investments further, having already committed $65bn to OpenAI alone. SoftBank raised $11.1bn in the largest high-yield corporate bond sale in history — at yields as high as 9.75% — to fund these bets. Sam Altman announced OpenAI will not pursue its expected IPO this year, citing AI safety concerns. That decision followed reports of AI agents going rogue to hack external systems and safety researchers departing Anthropic and Google over the technology's risks. Two Anthropic researchers warned that AI development without safeguards could lead to human extinction. Anthropic, meanwhile, is expected to push ahead with an IPO as soon as next month — safety warnings from its own researchers notwithstanding. The political backdrop is shifting. Democrat and Republican politicians are calling for new rules to govern AI systems. But the regulatory conversation is running far behind the capital allocation. Hundreds of billions are being committed on revenue projections that lack standardised definitions, by companies whose own researchers warn the technology may be existentially dangerous, funded by high-yield debt issued at near-junk yields. The question is not whether AI will generate value — $50bn in annualised revenue is real. The question is whether the valuation superstructure built on top of that revenue can survive the gap between what investors are told and what the numbers actually mean. What emerges is a familiar pattern from previous technology cycles: genuine underlying capability used to justify capital deployment at multiples that require perfect execution and continuous acceleration. The $20bn discrepancy is trivial as a business fact. As a signal about the information environment investors are navigating, it is anything but.
