On Monday, Jassy and Dario stood on a stage in Seattle and shook hands over a $125 billion agreement. Amazon writes Anthropic a check. Anthropic writes it right back.

Well, most of it.

Here's the actual math: Amazon commits up to $25 billion in fresh investment — $5 billion now, another $20 billion tied to "commercial milestones." Anthropic commits to spending more than $100 billion on Amazon Web Services over the next decade. The cash Amazon is putting in is less than the cash Anthropic is promising to send back.

The Numbers Don't Lie

Anthropic's valuation landed at $380 billion in this round. That's up from $183 billion eight weeks ago. The company now claims $30 billion in annualized revenue, which would edge it past OpenAI. Claude runs on AWS. AWS runs on Nvidia and Amazon's own Trainium chips. Anthropic pledged to bring nearly 1 gigawatt of Trainium capacity online by year-end, with up to 5 gigawatts secured for training.

Two months ago, Amazon inked a similar deal with OpenAI — up to $50 billion in investment, matched by a multi-year OpenAI commitment to use AWS compute. The pattern is unmistakable: hyperscalers invest in AI labs, AI labs spend that money back on hyperscaler compute, both sides book the flow as growth.

Nvidia did it first with CoreWeave. Microsoft did it with OpenAI. Now Amazon has done it twice in sixty days.

Why This Is Financial Engineering, Not Investment

If I give you $25 and you commit to buying $100 of my lemonade over the next ten years, did I invest in your business? Or did I just pre-purchase my own revenue?

The SEC calls this "round-tripping" when telcos do it. The accounting rules make it hard to book both sides as genuine. Yet in AI it's become the default operating model. The Financial Times flagged this exact pattern as a systemic risk two months ago. Nobody in the industry listened because nobody has an incentive to.

The part that should worry everyone: Anthropic said this week that enterprise demand has caused "inevitable strain" on its infrastructure, impacting reliability. The fix? More AWS. The revenue target? Keep climbing. The valuation? Up another $200 billion in two months. None of these numbers were set by a market — they were set by two companies negotiating across a table.

My Opinion

I'll be blunt. This is not an investment. This is Amazon buying future AWS revenue and getting equity as a bonus for showing up. If Anthropic's Claude product stays strong, Amazon wins twice. If Claude slows down, Amazon still ate its commitment. The only way Amazon loses is if Anthropic goes bankrupt before the $100 billion clock runs out — and when you're the one providing the compute that keeps the company alive, that outcome is remarkably within your control.

What bugs me is the narrative everyone is reinforcing. Headlines call it "Amazon expanding its AI partnership." It is not a partnership. It's a distribution agreement dressed up as venture capital, and it inflates the valuations everyone else is benchmarking against. Every founder raising a "Series C at $5 billion on $30 million ARR" is pricing themselves off numbers that were manufactured in a boardroom, not the market.

By this time next year, Amazon will have committed $75 billion across OpenAI and Anthropic. AWS will have booked $150 billion in forward commitments. Both labs will claim near-vertical growth. And when someone finally audits the flows, the story will be boring: money went in a circle, and everyone called it a boom.

Author: Yahor Kamarou (Mark) / www.humai.blog / 21 Apr 2026