News · 2026-08-04

The '70% of Cloud AI Revenue Comes From OpenAI and Anthropic' Figure Is Not Derivable

The claim that more than 70% of Amazon, Microsoft and Google's AI revenue comes from OpenAI and Anthropic is not derivable from public information. It originates in a single newsletter, Ed Zitron's Where's Your Ed At, whose own method is an anonymous-source estimate of OpenAI's Azure spending annualised against a Microsoft run rate from a different fiscal quarter, an older anonymous-source Anthropic AWS bill doubled by assumption, and - for Google - an explicit statement that the figure cannot be calculated because Alphabet does not disclose AI revenue.

Key facts

• Amazon's latest release puts its AI business above a $25 billion annual run rate, superseding the $15 billion denominator the calculation used.
• Microsoft last disclosed a $37 billion AI-business run rate in April and did not update the total on its latest call; it does separately isolate OpenAI in bookings and remaining-performance-obligation reporting.
• Alphabet reports Google Cloud as a blend of AI infrastructure, core platform, Workspace and TPU sales, with no AI-revenue line and no customer breakdown.
• Source of the claim: Where's Your Ed At, an independent newsletter.

What the arithmetic actually rests on

Take the Microsoft leg. You need OpenAI's Azure spending, which is not public, so you use an anonymous-source figure. You annualise it. Then you need Microsoft's AI revenue as a denominator, which was last disclosed in a different fiscal quarter, so you infer it. Two estimates, one from unnamed sources and one from a stale disclosure, divided by each other.

The Amazon leg is worse: an old anonymous-source figure for Anthropic's AWS bill, doubled by assumption to account for growth, over a run rate Amazon has since revised upward by more than half.

And Google simply is not in the calculation. The analysis says so.

None of this makes the underlying suspicion unreasonable. It makes the number unusable - and a number is exactly what gets screenshotted and repeated. That is the difference between an argument and a statistic.

What is verifiable, and it is not nothing

The narrower story stands up on primary sources, and it is genuinely strange. Strategic investors in the AI labs are also the labs' largest compute vendors.

Microsoft holds roughly 27% of OpenAI, remains its primary cloud partner, and receives revenue-share payments through 2030. Amazon says Anthropic committed over $100 billion of AWS technology spending across ten years, while Amazon invested $5 billion immediately with up to $20 billion more on milestones, on top of a prior $8 billion. Anthropic confirms large capacity agreements with both Amazon and Google.

So money moves from a cloud provider into a lab, and from the lab back to the cloud provider as compute spending. That is a real structural feature worth watching. It is also not the same as saying most of the provider's AI revenue is that loop, and the forward commitments in particular are pledges, not booked revenue.

Amazon's own filing makes a related distinction cleanly: its $53.4 billion Anthropic investment gain sits in non-operating other income, separate from net sales and AWS operating income. This site covered that when it landed - Amazon booked $53 billion on Anthropic and it is not revenue.

Why it matters

Circularity is the single most load-bearing worry in the AI trade right now. If demand for AI compute is substantially demand from companies the compute providers have themselves funded, the growth curve everyone is extrapolating means something different than it appears to.

That is precisely why the claim deserves better evidence than it has. A story this consequential either gets checked or it becomes a thing everyone knows and nobody can source. And the checkable version is available: watch the RPO and bookings disclosures where Microsoft does isolate OpenAI, and watch whether Amazon's $25 billion run rate holds its growth as Anthropic's committed spending ramps.

The shakeout is separately visible

One first-party data point needs no estimation. Flowise, a low-code builder for LLM workflows, announced its sunset: development froze July 29, the repository is archived August 10, official support ends August 31. Its stated reason is that coding agents have become capable enough that rigid low-code workflow builders no longer earn their place.

That is a cleaner signal about the shape of this market than any inferred percentage. The tools being squeezed are the ones whose value was making a capable model easier to wire up - a job the model increasingly does itself.

The honest caveat

Rejecting the 70% figure is not a defence of the cloud providers' disclosures, which are the actual problem. None of the three breaks out AI revenue in a way that would let anyone check this. Alphabet has no AI line at all. Microsoft's last AI run-rate number is months old. Amazon gives a run rate with no definition and no customer shares. The reason a shaky estimate travelled this far is that nothing better exists, and that is a choice the companies made.

Primary source, verified: read the paper →

Ground Truth. (2026, August 4). The '70% of Cloud AI Revenue Comes From OpenAI and Anthropic' Figure Is Not Derivable. Ground Truth. https://groundtruth.day/news/the-70-percent-circular-ai-revenue-figure-is-not-derivable.html

@misc{groundtruth:the-70-percent-circular-ai-revenue-figure-is-not-derivable,
  title  = {The '70% of Cloud AI Revenue Comes From OpenAI and Anthropic' Figure Is Not Derivable},
  author = {{Ground Truth}},
  year   = {2026},
  month  = {aug},
  url    = {https://groundtruth.day/news/the-70-percent-circular-ai-revenue-figure-is-not-derivable.html}
}

Topics: business · finance · openai · anthropic · cloud · circular-deals · accountability · microsoft