What Moonshot changed
Moonshot also made technical changes. It said it adjusted parameters to reduce unnecessary subagent calls and output length. In simple terms, subagent calls happen when one AI agent hands parts of a task to others, while output length is the number of tokens, or small chunks of text, the model produces in reply. Both can use up a subscriber's allowance quickly.
About Moonshot AI
Moonshot AI was founded in March 2023 in China, and its Kimi chatbot was released to the general public in November 2023. In July 2025, the company released Kimi K2, a one-trillion-parameter model that was open-sourced. Parameters are the internal settings a model learns during training, and a higher count usually means a larger system.
The company unveiled its Kimi K3 AI with 2.8 trillion parameters in July 2026. According to the company, it is the largest open-weight AI model in the world. Since the trained file is public, developers can use it and make applications on top of it.
Investors have backed the company strongly. TechCrunch reported in May that Moonshot raised about $2 billion at a $20 billion valuation. Moonshot has also begun seeking investor approval for a possible initial public offering in Hong Kong.
A pattern of capacity strain
This is not the first time a Kimi launch has put pressure on the company's systems. In July, Moonshot paused new subscriptions after demand for Kimi K3 pushed its GPU capacity close to the limit within 48 hours. Reuters also reported that the company planned to split future memberships into two plans, including a coding-specific option, to better match computing power to demand.
The wider industry is under similar strain. Chinese AI labs are suffering from computing shortages following US chip export controls.
What it means
The quota reset shows how hard it is for fast-growing AI firms to turn strong demand into a smooth service. Launching a popular model is one challenge, and serving it reliably to paying customers is another. For Moonshot, which is reportedly eyeing a stock market listing, keeping subscribers happy will matter as much as building bigger models.