China's Kimi AI Model Pauses Subscriptions Due to Overwhelming Demand
The powerful Kimi K3 model, a significant player in the U.S.-China AI race, faced capacity issues just days after its launch.
The new Chinese artificial intelligence model Kimi K3 has suspended new subscriptions after a surge in demand overwhelmed its capacity within days of its release. The development highlights the challenges Chinese AI models face in meeting growing user demand, both domestically and internationally, amid an intensifying U.S.-China technology competition.
Moonshot AI, the Beijing-based developer of Kimi K3, announced the pause in new subscriptions on Sunday, stating that demand had pushed capacity to its limits over the previous 48 hours. The company indicated it is prioritizing existing subscribers and is working to increase capacity, with new subscription spots to be reopened in batches. A similar message was posted on Chinese social media platforms.
The public rollout of Kimi K3 late last week aimed to challenge established AI models from companies like Anthropic and OpenAI. With 2.8 trillion parameters, Kimi K3 is considered the world's largest open-source AI model. Its performance in front-end coding capabilities has ranked it at the top of major advanced AI models on the Arena evaluation platform, according to reports.
Analysts noted that new model releases commonly generate significant interest that can strain compute infrastructure. Lian Jye Su, a chief analyst at Omdia, suggested that Moonshot AI may not have fully anticipated Kimi K3's popularity, leading to challenges in allocating its compute resources, which are both expensive and demanding.
The release of Kimi K3 has also raised concerns among U.S. technology stocks. Worries persist that more affordable Chinese AI models could impact the pricing power and demand for U.S. AI companies, even as U.S.-led restrictions limit China's access to advanced technologies, including cutting-edge chips.
Kimi K3 is among several recent Chinese AI models drawing global attention. In early 2025, DeepSeek's latest V4 model gained international recognition. Additionally, Alibaba previewed its Qwen3.8 Max model, reportedly with 2.4 trillion parameters, and the Chinese startup Zhipu (Z.ai) released its GLM-5.2 model, which has seen rapid adoption worldwide.