AI Reality — Human in the Loop

Can They Ever Pay It Back?

The two labs at the heart of the bet, their own revenue forecasts set against what they've already promised to spend. These are the companies' optimistic numbers — and one of them still doesn't add up.

OpenAI

its own revenue forecast
Projected revenue · scale to $350B
2025 actual$13B
2026~$36B
2030 target~$350B
What it has committed to spend
>$1.15 trillion

in hardware & compute commitments (2025–35), plus ~$856B of infrastructure spend and ~$278B of cash burn through 2030. Best-case cumulative revenue over that window: ~$840B.

Underwater on its own math. Even if every optimistic forecast lands, five years of revenue don't cover what it has already promised to spend — and it only turns cash-positive, maybe, in 2029–30.
OpenAI internal docs via Fortune (Nov 2025) & FT/Bloomberg (Sep 2026); The Information

Anthropic

its own revenue forecast
Annualised run-rate · scale to $70B
2024~$1B
2025~$9B
Q2 2026 run-rate~$46B
Q3 2026 run-rate~$65B
What it has committed to spend
~$901 billion

in total committed spend disclosed in its S-1/A no. 2 (10 Nov 2026), about $452B of it due before the end of 2030 — far beyond the ~$80B of AWS and Google commitments it carried a year ago. Its IPO is now priced at $70–78 a share, a ~$1.4T fully diluted valuation, with a first trade expected around 19 November.

Closer than OpenAI — but no longer a clean close. Anthropic's revenue is scaling fast: the S-1/A shows about $17B in the third quarter alone, up more than elevenfold on a year earlier. But the commitments it has now signed — roughly $901B — dwarf the ~$80B it carried a year ago, so the comfortable "its own numbers balance" read no longer holds. Still far healthier than OpenAI's math, but the gap it has to grow into just got much larger.
Anthropic S-1/A no. 2 (SEC, 10 Nov 2026); earlier projections via The Information / TechCrunch (Nov 2025)
The catch — read this before you trust either bar

Every number above is the company's own forecast, made while raising money on it.

Independent voices doubt even these. Sequoia's David Cahn estimates the industry needs on the order of $4 trillion in lifetime revenue just to pay back the compute it's already buying, and analysts have called OpenAI's targets flatly "impossible." Believe the forecasts, and OpenAI is still underwater. Doubt them, and both are. The one thing not in dispute: the money owed is real and signed; the money to pay it back is a projection.