Meituan has successfully trained LongCat-2.0, a 1.6 trillion parameter AI model, entirely on domestic Chinese processors without relying on Nvidia hardware. The achievement demonstrates China's capability to develop large-scale AI models using homegrown semiconductor technology.
Meituan's latest AI model represents a significant milestone for China's semiconductor independence and AI development capabilities. The company trained LongCat-2.0—a massive language model with 1.6 trillion parameters—exclusively using Chinese-manufactured chips, bypassing the need for Nvidia GPUs that have traditionally dominated AI training workloads.
The development comes amid ongoing US export restrictions on advanced semiconductors to China, which have limited access to cutting-edge Nvidia processors. By successfully training such a large model on domestic hardware, Meituan has demonstrated that Chinese tech companies can work around these constraints while maintaining competitive AI capabilities.
LongCat-2.0's scale places it among the largest AI models globally. The model's successful training on Chinese chips validates the performance and reliability of domestic semiconductor solutions for complex AI workloads. This could accelerate adoption of homegrown processors across China's tech sector.
Meituan, best known for its food delivery and local services platform, has increasingly invested in AI infrastructure and model development. The company's ability to train such a massive model suggests it has built substantial computational capacity and expertise in optimizing AI training processes.
The achievement holds broader implications for China's tech industry. As US restrictions continue, Chinese companies are developing alternative supply chains and proving they can build world-class AI systems without Western components. This development could reshape the global AI chip market and reduce dependency on single suppliers.
However, questions remain about comparative performance metrics, training efficiency, and real-world deployment scenarios. The demonstration establishes technical feasibility but doesn't necessarily indicate parity with equivalent models trained on premium Nvidia hardware in all operational contexts.
Anthropic has released a "reflect" feature for Claude that shows users an analysis of their chatbot usage patterns over customizable time periods. The feature joins a growing trend of year-in-review analytics across tech platforms.
The internet, smartphones, social media, and generative AI are collectively driving a sharp drop in how much people read longer works like books, according to reporting from The Atlantic.
Character.AI has debuted three human-written, AI-generated microdramas that users can interact with by chatting with characters. The platform plans to eventually enable users to create their own shows.