Anthropic acquires Stainless for $300M+: the strategic move that cuts off OpenAI and Google's SDK access

Anthropic acquires Stainless for $300M+: the strategic move that cuts off OpenAI and Google's SDK access

🔎 A takeover that redefines the AI war

On May 18, 2026, Anthropic announced the acquisition of Stainless, the startup that builds the official SDKs for OpenAI, Google, and Cloudflare. The amount: more than $300 million according to Forbes. In the wake of the deal, Stainless's hosted products for Anthropic's competitors were shut down. No transition period, no grace period. A surgical scalpel strike.

This acquisition is not just a simple consolidation move. It is a declaration of war in the infrastructure layer that most developers took for granted. AI agents are only as valuable as the systems they can reach, and Anthropic just sat on the main pipe.

The timing is not coincidental. Agentic AI governance is taking on a new dimension with recent industry alliances, and every point of contact with developers becomes a strategic battlefield. Stainless was the most invisible and most critical point of contact.

The Essentials

• Anthropic acquires Stainless for $300M+ and immediately shuts down the hosted products intended for OpenAI, Google, and Cloudflare.
• Stainless generated the official SDKs for almost all major AI providers — it is the invisible infrastructure behind every API call.
• This move is part of an AI cold war where control of the lower layers (SDKs, runtime, orchestrator) determines who wins the agentic layer.
• Open source alternatives like Smithy (AWS) and Kiota (Microsoft) suddenly become strategic, but require months of migration.
• Serious antitrust questions are emerging, especially since Anthropic has already shown a desire for geopolitical control of AI flows, as illustrated by the recent OpenAI demonstration of a geometry theorem that had resisted for 80 years, reminding us that the race for technical supremacy is limitless.

Recommended Tools

What Stainless was actually doing — and why everyone depended on it

Stainless wasn't a consumer product. It was developer infrastructure: the startup automatically generated SDKs in a dozen languages (Python, TypeScript, Go, Java, Rust, etc.) from OpenAPI specifications.

Concretely, when a developer ran pip install openai or npm install @google/generative-ai, they were downloading code generated by Stainless. The startup earned recurring revenue from each provider for this hosted service.

The problem became obvious when TechCrunch revealed the extent of the dependency: OpenAI, Google, Cloudflare, and dozens of other companies relied on Stainless for their developer layer. A single point of failure, a single 40-person company, controlled the code gateway of the AI industry.

The situation illustrates a simple principle: seemingly minor variations in initial conditions can produce disproportionate macroscopic effects. Stainless was one of those minor variations — until it became a $300 million strategic lever.

The details of the acquisition and the immediate shutdown

The official announcement from Anthropic presents the buyout as "vertical integration to improve the Claude developer experience." The press release does not mention the shutdown of competing services. However, that is exactly what happened in the hours that followed.

Stainless' hosted products — automatic generation, self-maintained documentation, versioning systems — were cut off for customers who were not Anthropic. OpenAI and Google woke up overnight with frozen SDKs, with no possibility to push updates for new API endpoints.

The internal reaction at OpenAI was brutal. The developer team had to urgently set up an in-house SDK generation pipeline. Google, with its broader ecosystem, was able to redirect to its internal tools more quickly, but the disruption was real for several days.

The amount of $300M+ is significant. Stainless had an estimated post-money valuation of around $200M during its last funding round. Anthropic paid a minimum 50% premium — which indicates that the strategic value (cutting off competitors) counted just as much as the startup's intrinsic value.

Why SDKs are the real battlefield of agentic AI

The AI war has gone through several phases. First, the model war (who has the best benchmark). Then the price war (who is the cheapest per token). Now, it's the agentic infrastructure war — and SDKs are the nerve center of this war.

An AI agent doesn't work alone. It calls tools, external APIs, databases, web services. Every call goes through an SDK. If your SDK is slow, poorly typed, or broken, your agent is useless — regardless of the quality of the underlying model.

This is where the comparison Google Gemini vs ChatGPT vs Claude : lequel pour quel usage ? takes on a new dimension. The quality of a model is no longer enough. You need a complete developer ecosystem, and the SDK is the foundational building block of this ecosystem.

Anthropic understood that in a world where GPT-5.5 scores 98.2 in agentic, Gemini 3 Pro Deep Think 95.4, and Claude Opus 4.7 (Adaptive) 94.3 on reference benchmarks, differentiation can no longer be based solely on scores. It is achieved through the integrated developer experience.

By owning Stainless, Anthropic controls:
- The speed of releasing new features (new Claude endpoints deployed across all languages in hours, not weeks).
- The technical quality of the SDKs (strict TypeScript types, consistent error handling, reliable streaming).
- The information advantage (Stainless sees developer usage patterns across all providers).

The impact on model rankings — what actually changes

The acquisition of Stainless does not directly change benchmark scores. But it shifts the competitive dynamic in a measurable way. Let's look at the current agentic ranking:

The overall ranking confirms the trend:

The lesson is clear: providers that relied on Stainless are now in a position of weakness. Those who had invested in their own pipelines (DeepSeek, xAI, Kimi) are suddenly at an advantage.

Open source alternatives: Smithy, Kiota and the race for independence

The closure of Stainless services triggered a rush toward open source alternatives. Two projects stand out.

Smithy (AWS) is an interface definition language developed internally by Amazon. It generates SDKs, documentation, and test artifacts from a single model. AWS has been using it for all its official SDKs since 2023. The problem: the learning curve is steep, and migrating from Stainless OpenAPI schemas requires significant rewriting work.

Kiota (Microsoft) is closer in spirit to Stainless. It is an open source SDK generator that takes OpenAPI descriptions as input and produces code in 12 languages. Developed by the Microsoft Graph team, it is mature and well-documented. But it does not offer a hosted layer — each company must deploy its own CI/CD pipeline.

The practical reality is that migration will take 2 to 6 months for the most exposed companies. During this period, Anthropic benefits from a silent but powerful competitive advantage: its SDKs continue to evolve while those of competitors are frozen.

For developers who want to experiment without relying on a specific SDK, free AI APIs like Groq, Google or OpenRouter offer an accessible entry point. But for production, the SDK question always comes back.

Antitrust implications: a minefield

Anthropic's move raises serious antitrust questions, and not just in the United States.

In the United States, the FTC has already been examining acquisition practices in AI since 2024. The acquisition of Stainless presents a problematic profile: a company acquires a critical vendor shared by all its competitors, then closes access to those competitors. In essence, this is acquisition-based exclusion.

In Europe, the Digital Markets Act (DMA) could apply if Anthropic is designated as a dominant player. The criterion is not revenue but the control of infrastructure that other companies "cannot realistically circumvent." An official SDK fits this definition precisely.

The legal framework exists, but the outcome is unpredictable. Past precedents (Microsoft in the 1990s, Google with Android) suggest that regulators are slow but eventually act.

The risk for Anthropic is twofold. On the one hand, an injunction could force them to reopen Stainless's services to competitors, nullifying the strategic advantage. On the other hand, even an investigation without sanctions creates uncertainty that deters investors and business partners.

The geopolitical dimension: control of AI flows

This acquisition is part of a broader trend of geopolitical control over AI infrastructure. Anthropic has already shown its willingness to restrict geographic access, such as when the startup denied China access to the Mythos model.

With Stainless, Anthropic now has an additional layer of control. The SDKs generated by Stainless include telemetry and version verification mechanisms. In theory, Anthropic could use these mechanisms to track or restrict the use of the SDKs — including those already deployed at competitors before the acquisition.

This possibility has alarmed security teams at OpenAI and Google. Internal audits are underway to verify whether existing SDKs contain callback mechanisms to Stainless's servers (and therefore Anthropic's).

Developers are discovering that their SDKs could contain invisible data flows they were never aware of — built-in telemetry whose impact on security and sovereignty is potentially massive.

What this means for independent developers

For the solo developer or small team, the lesson is brutal: never depend on infrastructure you don't control.

If you're building agents with Claude Opus 4.7 or Claude Sonnet 4.6, you're in a comfortable position in the short term. The SDK integration will improve. But if your architecture is coupled to the Anthropic SDK, you're creating a new dependency — different from the one you had with Stainless, but real.

The best practice is to abstract away from the specific SDK. Use an internal abstraction layer that normalizes API calls, then plug that layer into the SDK of your choice. This way, migrating from Claude to Gemini 3.1 Pro or GPT-5.4 Pro becomes a configuration change, not a rewrite.

To deploy your agents, reliable hosting like Hostinger at $2.99/month gets the job done for projects in the growth phase. The important thing is to keep the freedom to change the underlying model without breaking everything.

Self-hosted models like Kimi K2.6 Moonshot AI (88.1 agentic) or GLM-5 Reasoning from Z.AI (82) become more attractive in this context. Their SDK is clean, self-hosted, and no one can cut it off from the outside. The raw score is lower than GPT-5.5, but sovereignty has a value that benchmarks don't measure.

Competitors' reactions — what OpenAI and Google are doing now

OpenAI activated its contingency plan in less than 48 hours. The company had anticipated this scenario since acquisition rumors circulated in March 2026. A team of 15 engineers was mobilized to accelerate the deployment of an internal SDK generation pipeline based on Kiota.

Google took a different path. The company now uses its internal tools (including gRPC and protobuf components) to generate the Gemini SDKs. The advantage: total control. The disadvantage: a potentially less polished developer experience than what Stainless offered.

Cloudflare, smaller in the AI ecosystem, is the most significant collateral victim. The company does not have the resources of OpenAI or Google to build an equivalent pipeline urgently. The Workers AI SDKs are currently in maintenance mode, and some recent endpoints are simply not accessible via the existing SDKs.

xAI and DeepSeek, which did not use Stainless, took advantage of the situation to send aggressive messages to developers, inviting them to migrate to Grok 4.1 (90 in general, 79 in agentic) or DeepSeek V4 Pro Max (88 in general) where the SDKs are stable and sovereign.

❌ Common mistakes

Mistake 1: Thinking SDKs are a trivial problem

What's wrong: underestimating the complexity of a production SDK (streaming handling, retry logic, strict typing, pagination). The solution: treat your SDK as a critical component of your stack, not as a secondary dependency.

Mistake 2: Directly coupling your code to the Anthropic SDK after this acquisition

What's wrong: replacing a Stainless dependency with a direct Anthropic dependency, creating the same vendor lock-in problem. The solution: implement an abstraction layer that allows you to swap providers by modifying a single file.

Mistake 3: Ignoring existing SDK telemetry

What's wrong: continuing to use SDKs generated by Stainless without auditing them. These SDKs could contain network calls to Stainless servers. The solution: scrutinize your dependencies with a local network proxy (mitmproxy, Charles) and verify every outgoing call.

Mistake 4: Believing open source models are immune

What's wrong: thinking that self-host = risk-free. Z.AI's GLM-5 or Kimi K2.6 rely on ecosystems that can also be targeted. The solution: true independence comes from controlling the infrastructure layer, not just the model.

❓ Frequently Asked Questions

Will Anthropic open-source Stainless ?

Nothing indicates so. The official statement mentions "deep integration" with the Claude ecosystem. Opening the source code would negate the competitive advantage of the acquisition. Regulators could mandate it, but that is not the current direction.

Can I continue using the existing OpenAI SDKs ?

Yes, but they are frozen. New OpenAI API features will not be reflected in these SDKs until OpenAI deploys its own replacement pipeline. Plan for a migration within 3 to 6 months.

Is Claude Opus 4.7 now the best choice for agents ?

In terms of pure score, GPT-5.5 remains ahead (98.2 vs 94.3). But Anthropic's SDK advantage narrows the practical gap. For projects where the developer experience matters as much as the raw score, Claude becomes a defensible choice.

What if my company is locked into the Cloudflare SDKs ?

Contact Cloudflare support to get a migration timeline. In the meantime, use the REST API directly with a standard HTTP client. It's less elegant but functional and has no Stainless dependency.

Is this move legal ?

Probably, at this stage. Acquiring a company and stopping services for competitors is not illegal in itself. The antitrust question hinges on Anthropic's potential dominant position in the AI SDK market, which will take months to establish legally.

✅ Conclusion

Anthropic didn't just buy a tool — it bought the pipeline through which the entire AI industry's developers pass, then turned off the tap for competitors. At $300M+, it's one of the most aggressive strategic moves in the AI war. If you're building with agents today, your priority is no longer choosing the best model: it's ensuring you can switch tomorrow. Start by testing the free AI APIs available to avoid any exclusive dependency right from the prototyping stage.