Anthropic cuts Windsurf off from Claude over OpenAI acquisition rumors

There is nothing quite like watching an AI platform feud break out in real time.

Windsurf CEO Varun Mohan took to X on June 3 to share some bad news. With less than five days of notice, Anthropic suddenly severed Windsurf's direct first-party access to its Claude 3.x models, including 3.5 Sonnet and 3.7 Sonnet.

Mohan said Windsurf was forced to scramble onto third-party inference providers to keep service running, warning users they might experience short-term instability. He noted they wanted to pay Anthropic for full capacity, but Anthropic simply pulled the plug. Meanwhile, other models like Gemini 2.5 Pro and GPT 4.1 remain unaffected.

The $3B elephant in the room

Anthropic did not stay quiet for long. Speaking at TechCrunch Sessions on June 5, co-founder and Chief Science Officer Jared Kaplan laid out the exact reason: Bloomberg recently reported that OpenAI is in talks to acquire Windsurf for $3 billion.

Kaplan was unusually candid on stage, explaining that it would be odd for Anthropic to sell Claude to OpenAI. He emphasized that Anthropic is currently compute-constrained and needs to prioritize lasting partnerships.

Interestingly, Kaplan said Anthropic is still happily collaborating with Cursor, another popular AI coding assistant, despite Cursor building its own models. But Anthropic is also pushing hard into its own agentic products like Claude Code, betting that autonomous coding agents will quickly replace static chatbots.

Why it matters

We think Anthropic's swift cutoff is a massive wake-up call for the entire AI app ecosystem.

If you are a developer using Windsurf primarily for Claude 3.7 Sonnet, you should expect potential hiccups in latency and availability over the coming days. Windsurf is pushing Gemini 2.5 Pro with promotional pricing for a reason, and you may want to test alternative workflows while the dust settles.

For the industry, this highlights the razor-thin line between platform partner and mortal enemy. Anthropic simply cannot justify spending its scarce Amazon compute clusters fueling a product that may belong to OpenAI next week.

It also reveals a growing conflict of interest. As Anthropic builds its own agentic coding tools like Claude Code, independent wrapper products will find it increasingly dangerous to rely on a single foundation model provider. Having multi-model fallbacks is no longer an optimization; it is an existential survival strategy.