On Thursday afternoon, Google stamped a check for $40 billion. The recipient: Anthropic — the same company whose Claude has been quietly winning the enterprise AI market Google's own Gemini was supposed to dominate.

Four days earlier, Amazon had announced its own $25 billion pact with Anthropic. Add the two and Anthropic just collected sixty-five billion dollars in fresh pledges in less than a hundred hours.

That's not a funding round. That's a pile-on.

The Actual Numbers

Google's commitment breaks into two pieces. Ten billion in immediate cash at Anthropic's existing $380 billion post-money valuation. Then thirty billion more — but only if Anthropic hits performance milestones neither side will disclose. There's also five gigawatts of compute capacity coming online in 2027, jointly with Broadcom.

Before this deal, Google had already poured roughly $3 billion into Anthropic, holding a 14% stake. Now it's tripling down on a company whose annualized revenue went from $1 billion in January 2025 to $30 billion this month — a 30x climb in fifteen months, mostly powered by Claude Code adoption inside Fortune 500 engineering teams.

Amazon's parallel deal looks suspiciously similar. Five billion immediately, twenty billion tied to "commercial milestones," plus another five gigawatts of infrastructure. Anthropic, in turn, has committed to spending more than $100 billion on AWS over the next decade.

The Loop Nobody Mentions

The shape of these deals is identical, and the shape is the point. Big Tech announces a headline number. A fraction is real cash. The rest is undisclosed milestones. The startup spends the cash on the investor's own compute. The investor books revenue. The valuation goes up. The stock follows.

Google Cloud has spent two years explaining to Wall Street why TPU buildout is justified. Anthropic, conveniently, is one of the largest single consumers of TPUs on earth. The $40 billion is, in practical terms, a very expensive customer acquisition cost — paid in advance, recorded as an investment, and recouped through cloud bills nobody outside the deal will ever audit.

This is the third such structure in eight weeks. Two months ago, Amazon backstopped up to $50 billion of OpenAI's $110 billion round, also tied to a $100 billion cloud commitment. Same Amazon. Same arithmetic. Circular finance dressed as an investment thesis.

Why Google Specifically

Here's the part Google's PR will not say out loud: Claude is beating Gemini in the only segment that pays the bills. Enterprise. Anthropic's commercial product line is generating real, recurring revenue from customers who chose it over Google's offering. Forty billion dollars buys a meaningful stake in your competitor's success — and the right to keep selling them GPUs, TPUs, and cloud infrastructure regardless of whose model wins the benchmarks.

Anthropic, meanwhile, gets to scale without ever proving it can sell to customers outside the Big Tech investor circle. Most of its compute is funded by the people it competes with. The company is essentially incubated by the three hyperscalers — Google, Amazon, and the absent-but-watching Microsoft — at $380 billion of paper valuation.

My Opinion

I'll be blunt. This is not investment. This is laundering compute spend through a startup balance sheet so that two of the four hyperscalers can pretend their AI revenue is growing organically. It works because the press copies the headline number, the analysts model the contingent portion as "downside-protected upside," and the regulators don't yet have a mental model for a market where six companies pass the same forty billion dollars between them in a triangle.

What bugs me most is the contingent structure. Thirty out of forty billion is conditional on performance targets nobody will see. That's not a deal — that's an option. Google gets the upside if Anthropic moons, the headline if it doesn't, and the ability to quietly defer the payment if the math gets ugly. Amazon's $20 billion tranche works the same way. We're going to find out, two or three years from now, that "performance milestones" was always code for "we'll see."

The real story isn't $40 billion. It's that the AI market in 2026 has stopped being a market and started being a cartel. A handful of capital allocators are funding a handful of model labs are funding a handful of cloud providers are the same handful of capital allocators. When that loop finally hits a real customer-acquisition wall — and it will, because the enterprise software market is not infinite — the unwind will be ugly, fast, and very public.

Until then, expect at least two more of these deals before summer. Microsoft hasn't placed its second bet yet. It will.

Author: Yahor Kamarou (Mark) / www.humai.blog / 25 Apr 2026