Moonshot's Kimi K3 AI Soars Past Capacity, Halts New Sign-ups

Chinese AI startup Moonshot AI has hit a major speed bump, temporarily halting new paid subscriptions for its groundbreaking Kimi K3 AI model due to overwhelming demand that pushed its computing capacity to the absolute limit. Launched just days ago, the 2.8 trillion-parameter large language model quickly captivated users, forcing the company to prioritize existing subscribers and manage its stretched resources. This isn't just a minor blip; it underscores the intense global AI arms race and the critical 'compute bottleneck' facing even the most promising models, especially in China where US export controls restrict access to advanced chips. Kimi K3 has already outperformed top US rivals like Anthropic Claude Fable 5 and OpenAI's GPT-5.6 Sol on key benchmarks, adding to anxieties about China's rapidly closing AI gap. For Moonshot AI, which is eyeing a major Hong Kong Initial Public Offering (IPO) at a potential $30 billion valuation, this capacity crunch is a double-edged sword: proof of immense market traction, but also a stark reminder of the massive infrastructure investment required to sustain growth. Looking ahead, Moonshot AI has pledged to expand its Graphics Processing Unit (GPU) capacity 'as fast as we can,' planning to reopen subscriptions in batches and even split its service plans to allocate resources more efficiently, particularly for demanding coding workflows. The incident also fuels a broader debate on Wall Street, suggesting that sheer computing scale alone might not guarantee AI dominance, with architectural innovation and training efficiency gaining importance. Meanwhile, Chinese tech giant Alibaba has also just launched its Qwen3.8 model, intensifying the competition in the rapidly evolving global AI landscape.