Kimi's K3 demand just crashed its own GPU. New subscriptions are on hold.

Kimi.ai paused new K3 subscriptions after demand over 48 hours maxed out its GPU capacity. The company posted the decision on X on July 19, 2026, saying it would give existing subscribers first access to compute. The halt came weeks after Kimi K3 became the largest open model, and that attention outstripped available hardware.

The K3 model drew more interest than expected and strained capacity close to its limits. To keep service quality up, Kimi.ai will open new spots gradually as it adds infrastructure. The spike followed benchmark results where K3 tied at 92% on Next.js code tasks, a score that pulled in developers beyond the usual base.

The company plans to split memberships into two tiers: Kimi Membership for web, app, and work, and Kimi Code Membership for coding. The goal is to match compute loads to usage patterns and avoid outages. That approach is similar to Kimi Claw's one-click cloud deployment, which sidesteps hardware management.

The post did not say when subscriptions would reopen or how big the GPU fleet is. Similar capacity pressures have hit other Chinese labs, as the competitive landscape shifts. What is clear: K3, at 2.8T parameters, is the largest open-weight model around, and the surge was a predictable outcome of its sudden popularity.