China’s LongCat-2.0 Becomes the Biggest AI Model Without Nvidia Chips
China simply debuted its greatest AI mannequin skilled completely with out NVIDIA chips. Meituan unveiled LongCat-2.0, a 1.6 trillion parameter open-source massive language mannequin. The Beijing-based meals supply big ran the venture on home {hardware} end-to-end.
The breakthrough now reshapes how the international AI business views China’s tech self-reliance push.
What Meituan’s LongCat-2.0 Brings to the AI Race
A big language mannequin is an AI system skilled on huge datasets. These techniques perceive, generate, and motive in human language throughout many domains. LongCat-2.0 ranks amongst the largest ever, with 1.6 trillion parameters and a 1-million-token context window.
The launch comes as China continues to push for full self-reliance in critical compute infrastructure. Furthermore, Meituan mentioned LongCat-2.0 is the business’s first trillion-parameter mannequin to finish each coaching and inference on home {hardware}. As a consequence, the venture marks a significant technical milestone.
Follow us on X to get the newest information because it occurs.
The key distinction issues. DeepSeek’s V4-pro relied on home chips just for inference. That is the lighter job of answering consumer questions.
By distinction, LongCat-2.0 used home-grown {hardware} for each inference and the way more demanding pre-training stage.
Meituan said the cluster was constructed round large-scale ASIC superpods. These are chips personalized for particular workloads. Moreover, the firm used Huawei’s Collective Communication Library (HCCL) to handle chip-to-chip coordination at scale. The setup mirrors how NVIDIA’s NCCL coordinates its personal GPU clusters.
“…It jogs my memory of Jensen Huang’s level on the Dwarkesh podcast: export controls on Nvidia GPUs received’t cease China. They’ll simply speed up the improvement of AI that runs on Chinese chips,” analyst Yuchen Jin said on X.
Why the LongCat-2.0 Launch Matters Globally
LongCat-2.0 confirmed robust efficiency throughout a number of benchmarks. It outperformed Google’s older Gemini 3.1 Pro on Terminal-Bench 2.1 and SWE-Bench Pro.
However, the mannequin nonetheless trails international frontier techniques. These embrace OpenAI’s GPT-5.5 and Anthropic’s Opus 4.8 throughout the most demanding agentic and reasoning duties.
Industry observers reacted instantly. Tech analyst TP Huang said the launch places to relaxation considerations about Huawei’s Atlas-950 SuperPoDs. Furthermore, Lehigh University researcher Hanchi Sun called it the first mannequin ever skilled to near-frontier efficiency on 50,000 Chinese home accelerators.
“…If China can scale frontier coaching on native silicon at this degree, the compute arms race is wider open than ever,” enterprise accomplice Alvin Foo noted.
Significant hurdles stay throughout the broader Chinese AI stack. Meituan acknowledged that its software ecosystem still lags NVIDIA’s mature GPU community. Moreover, reminiscence limits had been the major bottleneck throughout pre-training. As a consequence, home accelerators carry much less reminiscence per gadget than NVIDIA’s banned H800 chip.
The broader sign is structural. Meituan’s success proves that frontier-scale coaching is now technically viable on Chinese {hardware}.
Consequently, the hole between Chinese open-source fashions and the prime closed Western techniques could shrink sooner than latest forecasts projected.
The put up China’s LongCat-2.0 Becomes the Biggest AI Model Without Nvidia Chips appeared first on BeInCrypto.
