Moonshot AI has temporarily paused new subscriptions for Kimi K3 after demand for the newly launched model pushed the company’s available GPU capacity close to its limits. Existing subscribers are not affected, and the company said it is adding capacity and plans to reopen new subscription spots in batches.
The pause is a material follow-up to Kimi K3’s launch because it shows that market interest in China’s latest frontier-class open model is translating into immediate infrastructure pressure. Kimi K3 is described by the company as a 2.8-trillion-parameter model with a one-million-token context window, built for long-horizon coding, knowledge work and reasoning. The model is available through Kimi.com, Kimi Work, Kimi Code and the Kimi API, with full model weights planned for release on July 27, 2026.
Moonshot also plans to split memberships into more focused plans: one for Kimi Web, App and Work, and another for coding workflows under Kimi Code. The move is intended to allocate compute more precisely and preserve service stability. That operational decision highlights a central issue in the AI market: even when models are positioned as open or lower-cost alternatives, live usage at scale remains constrained by compute supply, inference economics and the ability to manage high-intensity user workloads.
Kimi K3 has drawn attention because it represents another step in China’s push to compete with leading U.S. AI labs on model capability and price-performance. The company claims strong performance across its internal evaluation suite, while also acknowledging that the model trails leading closed models in some areas. Independent assessments will become more meaningful after the planned release of the weights, but the demand surge already suggests that developers and enterprises are willing to test alternatives outside the dominant U.S. model ecosystem.
For enterprise AI buyers in India and APAC, the development adds another variable to model selection. Cost, coding performance, long-context capability, data governance, geopolitical exposure, latency, reliability and vendor continuity are becoming part of procurement decisions. The subscription pause also shows that capacity guarantees may matter as much as benchmark claims when enterprises build production workflows on AI platforms.
Moonshot’s next step will be watched closely: reopening subscriptions in batches while releasing weights on schedule would strengthen confidence in Kimi K3’s operating roadmap, while prolonged constraints would underline how compute availability remains one of the defining bottlenecks in frontier AI adoption.
