Moonshot AI has suspended new consumer subscriptions for Kimi K3, saying that demand over a 48 hour stretch pushed the service close to the limits of its computing capacity. The pause came roughly three days after launch. Existing subscribers are unaffected, and the company says it is adding capacity as fast as it can and will reopen new subscription spots in batches.
Kimi K3 is the model that made the stampede predictable. Released on July 17, it is a 2.8 trillion parameter open weight system with a million token context window and native multimodal capabilities, the largest open model ever published, and Moonshot paired it with aggressive pricing. Within days it was the most talked about model in the world, with benchmark results the company says put it at the frontier.
The constraint is GPUs. Moonshot has world class research output and nowhere near the serving fleet of a United States hyperscaler, and United States export controls make expanding that fleet harder. It is a now familiar pattern for Chinese labs, where a model good enough to grab global headlines immediately outruns the servers available to run it. The suspension covers new consumer signups, while the API and existing plans continue.
The business stakes are larger than a waitlist. Reuters reports that Moonshot is preparing an initial public offering, and a viral flagship that cannot take new customers is both the best and worst possible headline for that process. It proves demand in the most concrete way available and it advertises the capacity ceiling in the same breath.
The wider signal from this week is consistent. Kimi K3 arrived, Alibaba answered within two days with its Qwen3.8 Max preview, and the bottleneck that actually bit first was not model quality but compute to serve it. Demand for open weight frontier models is real, immediate and paying, and for this wave of the race, capacity is the scarce resource.