Zitron: 70% of AI revenue traces to OpenAI and Anthropic

Ed Zitron's core claim, laid out in a recent YouTube interview and expanded in his newsletter, is that the AI revenue picture at the hyperscalers is far more concentrated than the equity market seems to price in. Microsoft's own FY26 filing that it "recorded revenue from commercial arrangements with OpenAI, inclusive of revenue sharing payments, of $24.1 billion" pegs OpenAI at roughly 70% of Microsoft's AI sales, and around 7% of total company revenue.

The pattern shows up across the other clouds too. UBS estimates cited by Zitron put OpenAI and Anthropic at 27% of Google Cloud revenue in 2026 and more than 48% in 2027, with Anthropic alone modelled at $76 billion. Barclays research on AWS puts the OpenAI plus Anthropic share at 13% this year, rising to 18% next. Aggregate across Microsoft, Google and Amazon and the reporting says analyst estimates have the two labs at over 70% of hyperscaler AI revenue.

For readers outside the trading desks, the demand story underneath the biggest CapEx cycle in tech rests on two customers that lose money at extraordinary rates. Zitron cites audited disclosures showing OpenAI lost $20.9 billion in 2025, and his framing to The Next Web is that much of Microsoft's AI revenue is "OpenAI's compute bill dressed as growth." If that framing holds, the hyperscalers are subsidising their own top line by funding rounds that flow back to them as Azure or GCP spend.

The specific figures come with caveats. The 70% figure at Microsoft is a Bloomberg estimate derived from disclosed revenue and prior growth rates, not an official split; the UBS and Barclays numbers are analyst models, not company guidance. Missing from the coverage is a clean dollar breakdown of AI revenue from customers outside these two labs, or how much of Anthropic's projected Google Cloud spend is contractually committed versus extrapolated from announced CapEx.

The useful thing for an operator to watch is not the next model release, it is the next funding round at either lab, and whether hyperscaler AI segment growth continues to decompose into "OpenAI and Anthropic" plus a much smaller everything-else line. If it does, the base case for the current AI trade is narrower than the headline numbers suggest.