China’s Meituan says its new AI model was trained on domestic chips

What happened

Chinese tech giant Meituan announced its new AI model, LongCat-2.0, contains 1.6 trillion parameters and was trained entirely on domestically made chips. This is the first AI model of this scale in China to be developed end-to-end using home-grown silicon, bypassing reliance on foreign hardware. The move clearly responds to current US export controls restricting AI chip access for Chinese companies.

Why it matters

Training a massive AI model on Chinese-made chips breaks a critical dependency on US technology. This reduces the risk of supply disruptions caused by export restrictions and allows China’s AI ecosystem more control over key infrastructure. For businesses and investors, Meituan’s approach pressures chipmakers globally and signals China’s determination to build a self-sufficient AI hardware stack. It also underscores the growing complexity and cost of training trillion-parameter models outside traditional Western cloud providers.

What to watch next

Track how the performance and efficiency of LongCat-2.0 compares to models trained on US or Nvidia GPUs. Watch if this pushes other Chinese firms to accelerate their chip development or training on domestic hardware. The industry should also monitor any shifts in global AI collaboration or supply chain realignments caused by such hardware independence moves. Meituan’s next steps in scaling or commercializing this model will reveal whether this architecture can sustain competitive advantage.

AI Quick Briefs Editorial Desk