Demand for Moonshot AI's Kimi K3 overwhelmed the Beijing startup's infrastructure within 48 hours of launch last week, forcing it to freeze new subscriptions and exposing a compute-chip shortage that threatens China's AI ambitions.
"Kimi K3 has received far more love than we expected," Moonshot wrote on X late Sunday. "Over the past 48 hours, demand has pushed close to the limits of our current capacity." The company said it is prioritizing existing subscribers and will reopen registration in batches as it adds capacity.
A similar message was posted on Chinese social media. At 2.8 trillion parameters, Kimi K3 is the world's largest open-source AI model. It topped the Arena evaluation platform's front-end coding rankings immediately after its public release, rattling U.S. rivals including Anthropic and OpenAI.
But the subscription freeze tells a different story about China's AI supply chain. Lian Jye Su, a chief analyst at Omdia, said the outage shows Moonshot lacks the compute chips needed to handle sudden demand surges.
"New model releases generally trigger massive interest, which can strain existing compute infrastructure," Su said. "This does show Moonshot AI does not have sufficient compute chips to serve the current surge in demand."
K3 is compute-intensive, Su added, making capacity allocation expensive and difficult. The crunch stems partly from Moonshot underestimating K3's popularity, but also from U.S. export restrictions that block China from accessing the most advanced chips. The launch has already pressured American tech stocks, with investors worried that cheaper Chinese AI models could undercut U.S. pricing power.
This dynamic mirrors the DeepSeek shock of early 2025, when a Chinese model triggered a global market selloff.
K3 joins a wave of Chinese AI releases drawing global attention. Over the weekend, Alibaba previewed its Qwen3.8 Max model with 2.4 trillion parameters.
Last month, startup Zhipu rolled out GLM-5.2, which saw rapid worldwide adoption. The pattern is clear: Chinese AI labs can build world-class models, but they still can't buy enough chips to run them at scale.
