On May 18, Anthropic announced that it had acquired a company called Stainless. The purchase price wasn't disclosed, but The Information had reported a week earlier that the deal was worth more than $300 million.
There's a reason this is more than a routine M&A headline. Stainless is the company that has built all of Anthropic's official SDKs—but it has also built the SDKs for OpenAI, Google, Cloudflare, Replicate, and Runway. In other words, Anthropic just bought, wholesale, the core infrastructure its own competitors were running on.
And with the acquisition, Anthropic made one thing clear: it would shut down Stainless's hosted product and SDK generator entirely. Existing customers can keep using the SDKs they've already generated, but new generation and automatic updates are gone. From here on, the tool runs only inside Anthropic.
It's a signal that the competition in the AI industry has dropped one level deeper.
What Is an SDK?
To understand this news, you first need to know what an SDK is.
If an API is a restaurant's menu, an SDK is the manual that tells you how to order. When an AI model company opens up its API, developers need to be able to call it from their own programs. But hand-crafting every HTTP request, parsing JSON, and handling authentication each time is tedious. So companies provide a "library that makes all of this easier to use." That's an SDK.
OpenAI's Python library, Anthropic's TypeScript library, Google's Java library—these are all SDKs. They're the bridge that lets a developer pull a model into their own code with a single line: import openai or import anthropic.
What Stainless did is interesting. It built a tool that takes an API specification as input and automatically generates SDKs in multiple languages—Python, TypeScript, Kotlin, Go, Java, and more. When the API changes, the SDK updates automatically, so no one has to maintain it by hand.
This matters because, for an AI company, SDK maintenance is a headache. Models update every week, and if the SDK can't keep up, developers can't use the new features. That's why OpenAI, Anthropic, and Google all used Stainless. For automatically building and maintaining their own SDKs, nothing else came close.
The Battlefield Just Got One Layer Deeper
Competition among AI companies has so far moved through three stages.
Stage 1: Model quality. Who can build the smarter model? GPT-4 vs. Claude 3 vs. Gemini. The competition ran on benchmark scores, reasoning ability, and coding ability.
Stage 2: Developer experience. Who can build the more usable tools? Claude Code, Codex, Cursor. This was the stage of pulling developers in through CLIs and IDE integrations.
Stage 3: Infrastructure ownership. Who controls the infrastructure those tools run on? The Stainless acquisition is the signal of this stage.
Each stage determines the one above it. A great model goes unused if the developer experience is bad. And even with a great developer experience, if the infrastructure sits in a competitor's hands, it can be cut off at any time. So Anthropic bought the very infrastructure it had been depending on.
"Which SDK Is the Default" Decides Everything
Imagine a junior developer using an AI model for the first time. Search for it, and the first thing that comes up is the official SDK. pip install anthropic, pip install openai, pip install google-generativeai. That single command determines which AI company's client gets installed first.
Here's the decisive part. Once you start writing code with one SDK, moving to another isn't easy. Function signatures differ, error handling differs, streaming works differently. The SDK you learned first determines which model that developer uses afterward. It amounts to a lock-in strategy.
Zoom out, and "which SDK is in the production code" determines "which model is tied to that company's revenue." An SDK isn't just a client library; it's the starting point of a revenue stream.
This is where Anthropic's reason for buying Stainless becomes clear. If it can raise the quality of its own SDK faster than its rivals while keeping those rivals from using the same tool, the friction of choosing Claude drops—and the friction of using a different model rises. The SDK itself becomes the competitive advantage.
The Next Move After Overtaking on Enterprise Customers
What makes this acquisition even more significant is its timing. Last month, reports emerged that Anthropic had overtaken OpenAI in the number of enterprise customers. It was the first reversal in an area where OpenAI had long held the lead.
Right after that overtake, Anthropic's next move was the Stainless acquisition. It aimed for two effects at once: cementing the lead it had gained while weakening a competitor's infrastructure.
Once Stainless's hosted product shuts down, OpenAI and Google have to go back to maintaining their SDKs themselves or find another tool. In the short term, that means higher costs and slower speed. This isn't a matter of days or weeks, but given how fast AI models update—every week—the accumulated gap could widen meaningfully.
Anthropic's Vertical Integration Strategy
The Stainless acquisition signals that Anthropic's vertical integration strategy is kicking into gear. It's building a structure where one company controls everything from top to bottom.
The model layer. Claude Opus 4.7, Sonnet 4.6, Haiku 4.5. It builds the models itself.
The developer-tool layer. Claude Code, Claude Desktop. It builds the CLI and the IDE integration itself.
The SDK layer. With the Stainless acquisition, it now owns the SDK generation tool itself.
The consulting layer. The $1.5 billion joint venture announced on May 4 with Blackstone and Goldman Sachs. It's building a consulting arm that installs Claude directly into private-equity portfolio companies.
The infrastructure layer. Through partnerships with Amazon and Google, it's securing cloud infrastructure.
All five of these layers are coming under Anthropic's control. It's a structure where an advantage in one layer carries into an advantage in the next. The model is good, so the tools are good; the tools are good, so the SDK feels natural; the SDK feels natural, so the consulting is fast; the consulting is fast, so revenue grows.
OpenAI Won't Sit Still Either
Anthropic's moves put the same pressure on OpenAI, because they signal that model quality alone is no longer enough.
On May 4, OpenAI announced a $10 billion deployment company. It was an aggressive move to secure enterprise channels—going so far as to offer private equity a guaranteed 17.5% return. The Stainless episode is likely to be yet another jolt for OpenAI. Strengthen its own SDK infrastructure, acquire another tool company, or build a new tool itself—some response will come, one way or another.
Google faces similar pressure; it needs to strengthen its own SDK infrastructure. And smaller companies like Cloudflare, Replicate, and Runway have suddenly lost a core tool. A market has opened for someone to build a new tool to fill that gap.
What Korean Developers and Companies Should Watch
This shift carries direct meaning for Korean developers and companies.
SDK choice becomes model choice. If you're starting a project that bolts on AI features, the SDK you choose determines that project's dependencies for the next six months. You have to think not just about "which model is good," but "which SDK is going to get embedded in our code."
Spreading your dependencies matters. Rely on a single company's SDK, and you become vulnerable to that company's policy changes. You need a design that uses Claude, GPT, and Gemini differently depending on the nature of the task, and that abstracts the interface so swapping SDKs is easy.
Korean AI companies have to invest in SDK quality too. There are companies in Korea building their own AI models. Compete on model quality alone, and they'll be outmuscled by Big Tech. An SDK specialized for Korean-language processing, and tools suited to Korean development environments, could become a point of differentiation.
Developer experience is revenue. Korean companies selling APIs or SaaS B2B should take the same lesson. Building a good product isn't enough. Polishing the SDK and documentation so a developer can make their first successful call within five minutes is the starting point of revenue.
Who Owns the Developer?
The real message of this acquisition is this. AI companies no longer compete on "who builds the smarter model." They compete on "who owns the developer's convenience."
The tools a developer uses every day, the library they call every day, the docs they read every day—when all of it sits under one company's control, that company's model gets embedded in every product the developer builds. Revenue is the natural result.
Stainless was one piece of that control. It looks small, but it was central. Anthropic paid more than $300 million for that piece because it understands the size of the cumulative effect a single piece like this produces.
Going forward, when you see acquisition news from AI companies, watch for one thing: "Which tool in a developer's daily life does this acquisition control?" That tool is becoming a more telling indicator than the model itself.
From model-quality competition, to developer-experience competition, and on to infrastructure-ownership competition. The battlefield is deepening one layer at a time. And the deeper it goes, the harder it becomes for small companies to catch up. The integration game among Big Tech has entered its real opening stage.
Which SDK becomes the default decides which model goes into production. And who owns that SDK decides the next five years of the AI market.
