Anthropic buys software development kit generator Stainless, shuts hosted compiler same day

The hosted compiler that every major AI lab uses to maintain its developer libraries went dark on May 18th. Testing by type0 confirms api.stainless.com returned 404 errors the same morning Anthropic announced it had acquired Stainless — the four-person New York startup behind the pipeline — for $300 million-plus, as TechCrunch reported. The trade press described the deal as a developer experience play. The structural read is different: Anthropic just bought the maintenance infrastructure for the SDKs that OpenAI, Google, Meta, Cloudflare, Runway, Groq, and Cerebras distribute to millions of developers — and switched it off the same day.

Stainless makes software development kits. Not the APIs themselves, but the client libraries that engineers install when they want to call an AI model. When you run pip install openai or pull the Gemini Python package, the package on the other side was generated by Stainless. The company took an API specification and automatically produced production-ready SDKs across Python, TypeScript, Go, Java, and Kotlin, handling retries, pagination, streaming, and authentication without manual library maintenance.

The dependency is visible in the source. The top of [OpenAI's Python SDK __init__.py](https://raw.githubusercontent.com/openai/openai-python/main/src/openai/__init__.py) reads: "File generated from our OpenAPI spec by Stainless." The same attribution appears in Anthropic's own Python SDK. What the labs handed off to Stainless was not just tooling — it was the build pipeline for the libraries their customers depend on.

OpenAI tried building its own SDK generator in the early days of the API, according to people familiar with the project, and abandoned it when the maintenance burden exceeded what one engineer could sustain. It adopted Stainless instead and stayed. The same was true for Google, Meta, Cloudflare, Runway, Groq, and Cerebras. Forbes reported millions of downloads per week across all Stainless-powered SDKs.

Existing customers retain the SDKs they have already generated and can modify them. What is going away is the hosted compiler that produced new builds whenever an API specification changed. That automation was the entire value proposition for labs that stopped maintaining their own SDK infrastructure. Anthropic says the compiler is now switched off; testing by type0 confirms the endpoint at api.stainless.com returns 404 errors.

For OpenAI, the timing is awkward. The company just shipped GPT-5 with a restructured API surface. Developers integrating GPT-5 through the official Python SDK would, under the old arrangement, have received automated updates as the API spec evolved. Per Anthropic's announcement, the compiler producing those updates is now switched off. OpenAI can rebuild its own SDK generation infrastructure, the way it did before Stainless existed, or migrate to one of the funded alternatives that have emerged in the last eighteen months: Speakeasy, LibLab, Konfig, and Fern all raised specifically to serve this category. The OpenAPI Generator open-source project also remains available. None of these currently match Stainless on polish, but the gap is closable in engineering quarters, not years.

Google faces a similar calculation across parts of its Gemini API surface. Meta is rebuilding Llama Stack integration on its own timeline. Enterprise platform teams at nine-figure AI customers will need to evaluate whether their existing SDK integrations require active maintenance they no longer have, and whether to build internally or migrate to an alternative.

There is a second-order question that enterprise procurement and security teams will ask before the engineers do. API specification changes from every major foundation model lab flowed through the Stainless compiler before those changes went public, according to Forbes. That meant Stainless saw new model capabilities, pricing changes, and API parameter modifications from OpenAI and Google before the industry knew them. Anthropic says any such signal is firewalled. That assurance is now Anthropic's promise to keep and the rest of the industry to trust. The perception risk alone will factor into contract renewals.

This is Anthropic's fourth acquisition in roughly six months. Entrepreneur Loop reported Bun in December, Vercept in February, Coefficient Bio in April. Each sits on a different layer of the developer stack, and each was used by Claude customers or by Claude itself before the acquisition. The pattern underneath is not talent acquisition. It is vertical positioning.

The company crossed a $30 billion annualized revenue run rate in April, ahead of OpenAI's roughly $24 billion, according to people familiar with the numbers. More than 1,000 enterprise customers now spend over $1 million per year with Anthropic, doubling from 500 in February. It is no longer building just a better model. It is assembling a platform, and the Stainless deal is its most strategically loaded move yet.