LongCat-2.0: China's Largest AI Model Without NVIDIA

China has unveiled its largest artificial intelligence model trained entirely without NVIDIA chips. Meituan, a food delivery giant, debuted LongCat-2.0, an open-source large language model with 1.6 trillion parameters. This project ran completely on domestic hardware. Here's the thing: LongCat-2.0 is the first trillion-parameter model to complete both training and inference on domestic hardware. This is a significant technical milestone. The company used specialized chips and Huawei's Collective Communication Library for coordination. What's interesting is that this model outperformed Google's older Gemini 3.1 Pro on some benchmarks, though it still trails systems like OpenAI's GPT-5.5. Analyst Yuchen Jin notes that export controls on NVIDIA GPUs may accelerate the development of AI on Chinese chips. The bottom line: This development reshapes global views on China's tech self-reliance and intensifies the global AI compute arms race.