Alex Rattray spent four years building a tool that quietly became plumbing for every major AI lab. Monday morning, he sold it to one of them. By Monday afternoon, that one was the only one left using it.

Anthropic announced it bought Stainless, the New York-based startup Rattray founded in 2022 after leaving Stripe. The price wasn't disclosed. The Information says north of $300 million. The deal itself takes ten minutes to summarize. The implications take longer.

Stainless does one thing well: it turns API specifications into production-ready SDKs—the libraries developers use to actually call those APIs—across Python, TypeScript, Go, Java, and Kotlin. Every time OpenAI shipped a new endpoint, Stainless kept the Python client in sync. Same for Anthropic. Same for Google. Same for Cloudflare, Replicate, Runway, and Meta. Backed by Sequoia and Andreessen Horowitz, the company became infrastructure—the kind nobody notices until it's gone.

Now it's gone. Or rather, gone for everyone except Anthropic. The company confirmed it will wind down all hosted Stainless products, including the SDK generator. Existing customers keep the code already generated. They lose access to the platform that keeps it current. As APIs evolve, those SDKs rot. Quickly.

Rattray called the move easy in his press release: "Anthropic was one of the first teams to bet on this with us." That part is true. Stainless software has powered every official Anthropic SDK since the company started shipping APIs. What he didn't say is that he just took years of accumulated infrastructure work and handed it to one customer while pulling it away from five others, including the customer most associated with the modern AI industry.

This is not a normal acquisition. Anthropic didn't buy Stainless to scale Stainless. It bought it to remove it from the market. There's a name for that: defensive acquisition, or, in less polite jurisdictions, anticompetitive consolidation. The FTC has been quietly examining exactly this pattern in tech mergers for the last three years. Whether they look at this one depends on whether anyone files a complaint.

OpenAI almost certainly won't. They have better engineers than Stainless and the headcount to rebuild internally. Cloudflare won't either—they own enough developer mindshare to ship their own generator in a quarter. The interesting question is what happens to Replicate, Runway, and the long tail of mid-sized AI companies whose SDK maintenance just became someone else's part-time job.

My Opinion

I'll be blunt. This is a smart, ruthless move, and I respect it. Anthropic just paid roughly one-third of one percent of its rumored $900 billion valuation to make every competitor's developer experience marginally worse. That's the best return on $300 million I've seen this year.

Here's what bugs me, though. Anthropic spent the last three years telling everyone it's the responsible AI lab—the one that publishes safety research, the one that refused to sell to the Pentagon for "any lawful purpose," the one that calls itself an "AI safety company." This acquisition isn't safety. It's a moat. There's nothing wrong with building a moat, but the company positioning itself as the principled alternative just pulled the most cynical M&A play of 2026 so far. The cognitive dissonance is going to leak out at the next All Hands.

The bigger story is what this signals. We're now in the phase where AI labs are buying adjacent infrastructure not to use it, but to deny it to rivals. Expect more of these. Cursor, Pinecone, LangChain, every plumbing company the big four depend on—you're now an acquisition target whether you want to be or not. The valuation premium isn't for your product. It's for taking your product off the table.

Rattray and his team get to keep doing the work they love. OpenAI gets to write the same SDK code Stainless was writing for them, internally, at much higher cost. Anthropic gets a small, durable advantage and a recruiting tool that says "join the winner." The losers are the developers at every other AI lab who now have one more permanent thing on their plate. Welcome to the consolidation everyone said wouldn't happen because the AI labs were too principled for it.

Author: Yahor Kamarou (Mark) / www.humai.blog / 19 May 2026