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

Kimi.ai, the company behind the K3 large language model, announced it is temporarily suspending new subscriptions after demand over the past 48 hours strained its GPU infrastructure. The decision, shared via a post on X on July 19, 2026, prioritizes compute resources for existing subscribers over new sign-ups. The move comes just weeks after Kimi K3 usage on Opencode doubled in one week, signaling a broader demand wave that the company's hardware wasn't ready to absorb.

According to the post, the K3 model received more interest than anticipated, pushing close to the limits of current capacity. To protect the experience of existing users, Kimi.ai will roll out new subscription spots in batches as it expands capacity. The demand spike follows the release of benchmarks where Kimi K3 tied for first at 92% on Next.js code tasks, which likely drove attention from developers beyond usual usage patterns.

The company also outlined plans to split its membership into two focused plans: Kimi Membership, covering the web, app, and work use cases, and Kimi Code Membership, tailored for coding workflows. The move is intended to match compute resources more precisely to usage and keep the service stable for different user segments. That mirrors the strategy behind Kimi Claw's one-click cloud deployment, which also aims to simplify access by removing hardware overhead.

The post did not specify when new subscriptions would reopen or provide details on the current scale of Kimi's GPU fleet. What is known is that Kimi K3, at 2.8 trillion parameters, is the largest open-weight model available, and it beats frontier models on the benchmarks that matter for certain workloads, making the capacity crunch a natural consequence of its sudden popularity.