Moonshot AI halts Kimi K3 sign-ups after demand maxes out its GPUs

An AI model so popular it sold out: Moonshot AI stopped selling Kimi K3 subscriptions 48 hours after launch.

Full capacity in two days

The Beijing startup paused all new sign-ups on July 19, 2026, after demand for its new flagship model pushed its GPU clusters to full capacity. Existing subscribers keep access; everyone else waits for capacity to be added in batches.

What Kimi K3 is

Kimi K3 is a 2.8 trillion parameter mixture-of-experts model, and its benchmark run explains the stampede: it has beaten GPT-5.6 Sol and Claude Fable 5 on several benchmarks, including Arena AI’s front-end coding ranking. Users do note a tradeoff: it runs noticeably slower than the top US models.

A Chinese lab fielding a model that trades benchmark wins with the American frontier is itself news. A Chinese lab that cannot serve the resulting demand is the next chapter of the same story, and it points straight at the chip supply chain.

The open-source land grab

Here is the bigger play. Moonshot says it will release the model weights on July 27, making K3 the largest open-weight frontier model in the world.

The logic turns a weakness into a strategy. Moonshot’s hardware constraint means it cannot monetize all the demand it created, so instead it is betting the global developer community will absorb the serving load that China’s chip supply chain cannot. Anyone with the infrastructure to run a 2.8 trillion parameter model gets a frontier-class system for free, and Moonshot gets an ecosystem it could never have served itself.

Consider the situation from the frontier lab’s perspective: it literally cannot sell more of its product, and its answer is to give the model away. Come July 27, the most interesting question in open-source AI becomes very practical: who has the hardware to actually run it?