Google Isn't Betting on Anthropic. Google Is Its Landlord.

What if the biggest AI investment in history is not really an investment at all?

Within the space of a single week, Amazon committed $25 billion to Anthropic and Google followed with another $40 billion — a combined $75 billion from two cloud giants flowing into a single AI lab. The headlines declared it a historic vote of confidence in AI's future. I read it differently. What Google and Amazon have quietly structured is something far more unusual: a mechanism to lend Anthropic money that will, in large part, be repaid through Anthropic's own cloud bills.

The structure of Google's deal makes this hard to miss. Anthropic's Claude models are trained predominantly on Google TPU pods; Amazon Trainium handles the secondary workload. Google's $40 billion is split into $10 billion upfront and $30 billion contingent on unspecified "performance targets" — the kind of language that, in deals of this shape, almost invariably encodes compute spend milestones. Amazon's $25 billion follows the same playbook. Anthropic's annual run rate has now hit $30 billion; a substantial portion of that revenue flows directly back to the two companies writing the biggest cheques. The money goes in. Much of it comes back.

The real issue: When your two largest investors are also your two primary infrastructure suppliers, the line between "funding" and "a very expensive long-term supply contract" quietly disappears.

The obvious counterargument: equity is still equity. Google reportedly holds around 14% of Anthropic. If the IPO — being explored for as early as October 2026 — prices anywhere near the rumored $800 billion valuation, that stake is worth north of $110 billion on a $40 billion outlay. That is a genuine return, and I am not dismissing it.

But here is what that counterargument misses: Google profits from Anthropic whether or not there is ever an IPO, whether or not Claude remains the leading model, whether or not Anthropic's safety-first thesis proves correct. Every TPU cycle Anthropic burns is Google Cloud revenue. The equity upside is a bonus layered on top of a guaranteed revenue stream. That is not how investors typically behave. That is how landlords behave.

The deeper question this raises is not financial but structural. Anthropic's credibility rests on its independence — its willingness to slow down, to publish safety research that makes the industry uncomfortable, to say no to capabilities the market demands. That independence is harder to maintain when $75 billion in operating capital is this tightly coupled to the commercial interests of two specific infrastructure vendors. A lab that cannot afford to anger its compute suppliers is not truly independent, whatever its charter says.

Watch for the moment Anthropic's next major training run demands infrastructure that neither Google's TPUs nor Amazon's Trainium can serve efficiently. That will tell us whether these deals represent genuine partnership — or whether the exits have been quietly locked from the outside.