Meituan’s LongCat-2.0 shows China can train massive AI models without Nvidia

What happened

Meituan developed and trained LongCat-2.0, an AI model with 1.6 trillion parameters, using only Chinese-made chips. The process did not rely on Nvidia hardware, marking a significant shift in how large AI models can be built outside of Nvidia’s ecosystem.

Why it matters

Large AI training traditionally depends heavily on Nvidia GPUs. Meituan’s achievement shows Chinese companies can build and train massive models without Nvidia’s technology, weakening Nvidia’s previously near-monopoly status in this space. This effort pressures suppliers to compete on hardware or risk losing ground in the world’s fastest-growing AI market. For businesses, this means alternative hardware platforms trained for giant AI models are maturing, which could lower costs or hedge supply chain risks. Investors and founders should track this as a signal that China is investing heavily in independent AI stack development, which will shape global AI competition.

What to watch next

The next question is whether LongCat-2.0 competes with global AI leaders in performance or mainly serves domestic use cases. Watch if Meituan or other Chinese players expand these hardware innovations into AI cloud services or developer tools that can attract global clients. Also, keep an eye on how Nvidia responds, whether by innovating hardware, adjusting pricing, or pushing software ecosystem lock-in to fend off these rising alternatives. Finally, look for other companies in China attempting similar feats, boosting local AI independence and cloud infrastructure options.

AI Quick Briefs Editorial Desk