Moonshot AI stopped accepting new subscribers for its Kimi K3 model on July 19, 2026, just days after launch, because the company's GPU infrastructure couldn't handle the demand. The pause protects existing users from service degradation, the company said. Active subscriptions remain unaffected.
Why the pause was needed
The open-weight model, with 2.8 trillion parameters, launched around July 16. Within 48 hours, the surge in usage strained available GPU capacity. Moonshot AI is racing to scale its infrastructure but hasn't given a specific date for reopening new sign-ups. The situation reflects a broader industry problem: AI companies are struggling to keep up with rapid usage spikes amid fierce competition for scarce computing resources.
What Kimi K3 offers
Kimi K3 features a 1-million-token context window and native multimodal capabilities. Its full weights are scheduled for public release on July 27. The Arena benchmark ranked K3 first for building web interfaces, ahead of models from several leading American and Chinese labs. Moonshot AI restructured its membership into two tiers: Kimi Web/App/Work and Kimi Code Membership.
The cost advantage
Chinese models are 112 times cheaper than Anthropic per million tokens. Pricing comparisons: Anthropic $56, OpenAI $26, Meta $1.50, xAI and Google $1, Chinese models $0.50. That cost edge has driven strong API demand, contributing to Moonshot AI's reported $300 million annual recurring revenue in June.
The company's valuation surpassed $20 billion in May, and it is negotiating a new investment round that could push it beyond $30 billion. Moonshot AI is also considering a Hong Kong IPO within roughly six months. Founded in 2023 by former Tsinghua University professor Yang Zhilin, the company now faces the challenge of scaling fast enough to meet demand without alienating users. An exact reopening timeline for K3 subscriptions has not been provided.
