China’s Meituan open-sources massive LongCat-2.0 AI model, saying it was trained on domestic chips

What happened

Meituan, a Beijing-based tech giant known for services like food delivery, released LongCat-2.0, a large language model with 1.6 trillion parameters. Notably, Meituan trained this AI entirely on Chinese-made chips and domestic computing clusters, signaling a push for self-reliance in AI hardware. The company open-sourced the model, making it accessible to developers and researchers.

Why it matters

LongCat-2.0 shows China is building massive AI models without relying on Western or global semiconductor suppliers. This reduces dependence on US or foreign hardware and circumvents export restrictions or supply chain bottlenecks. For builders and investors, it signals acceleration in China’s AI ecosystem and chip development capabilities. Meituan’s move also challenges assumptions that only hyperscalers and chip leaders in the US dominate large-scale AI innovation. Open sourcing a model of this scale could lower barriers for Chinese startups and research groups, increasing competition and innovation in local AI applications.

What to watch next

The real test will be LongCat-2.0’s performance and adoption beyond China, especially against more established models trained on global infrastructure. Watch how developers leverage the model for commercial use cases and whether similar initiatives emerge from other Chinese companies, furthering domestic AI independence. Also, keep an eye on Chinese chipmakers to see if this AI demand fuels their technological advances or volume growth. Lastly, observe if Western companies adjust AI strategies in response to this self-contained ecosystem growth within China.

AI Quick Briefs Editorial Desk