• 2 min read
Moonshot AI halts Kimi K3 signups after demand spike
Moonshot AI paused new Kimi K3 consumer subscriptions just three days after launch as demand outpaced compute capacity.
Image: ITzine
Just three days after launching Kimi K3, Moonshot AI stopped taking new consumer subscriptions. The company said demand rose faster than it could expand compute capacity, and it chose to protect access for existing paid users rather than let service quality slip.
K3 drew attention for more than launch hype. The model has 2.8 trillion parameters, uses a Mixture-of-Experts (MoE) architecture, and supports context windows of up to 100 million tokens. That combination makes it notable for long prompts and more complex applied workloads. The release also drew attention outside China, including from Elon Musk.
Now Moonshot AI’s problem is less about visibility and more about infrastructure. The company says it is redirecting available resources to current subscribers and speeding up deployment of new hardware. It has not said when new subscriptions will reopen.
Recommended reading
Jaron Lanier says 'AI' is the wrong story
Why Kimi K3 hit an infrastructure limit
An MoE design can reduce compute use for a single request, but large-scale product usage still drives load up quickly. With K3, that pressure is especially visible on long-context tasks: the more tokens involved and the more complex the prompt, the more expensive each response becomes for the provider.
Independent testing added to the surge in interest. According to Guillermo Rauch, CEO of Vercel, K3 delivered strong results in interface code generation and ranked first in his comparison. For China’s AI market, that is a meaningful signal: local models are increasingly getting noticed by developers outside the country, not just at home.
This is also not an isolated case. In generative AI, compute shortages have become routine. OpenAI, Anthropic, and Google regularly restrict access to their heaviest model modes, while Chinese companies are also competing directly with DeepSeek, Alibaba, and ByteDance.
Moonshot AI says it will also reorganize its subscription lineup, separating Kimi Web, Kimi App, and Kimi Work from Kimi Code to spread demand across different types of tasks. If that works, the company should get a clearer view of where capacity is tightest — coding, document analysis, or general model use.
AI Editor
Ava covers the rapidly evolving world of artificial intelligence, from foundational models and research labs to the real-world economics of intelligence. With a background in computational linguistics, she cuts through the hype to find out what actually works. She firmly believes that benchmarks are just marketing until reproduced in the wild.
via ITzine
