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

2026-06-30 · Source: The Decoder · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Intermediate, quick

Summary

Meituan has successfully trained LongCat-2.0, a 1.6 trillion parameter AI model, entirely on a cluster of over 50,000 domestically made AI ASICs in China, processing more than 35 trillion tokens. This achievement, by a team established in 2023, demonstrates China's capability to develop large-scale AI models without relying on Nvidia hardware, despite US export controls implemented since 2022. LongCat-2.0 shows competitive performance on certain benchmarks, topping Gemini 3.1 Pro and GPT-5.5 on SWE-bench Pro (59.5) and SWE-bench Multilingual (77.3). However, it trails leading Western models like Claude Opus 4.7 and 4.8 on these, and significantly lags Gemini and GPT-5.5 on tests such as IFEval (90.0), IMO-AnswerBench (81.8), and GPQA-diamond (88.9). The specific chip maker was not disclosed, and the model is not yet publicly available for independent verification.

Key takeaway

For AI policy makers and strategists assessing global AI capabilities, Meituan's LongCat-2.0 signals a significant shift. You should recognize that China's domestic chip industry can now support the training of massive, competitive AI models, potentially reducing the impact of export controls. This development necessitates a re-evaluation of current technology restriction strategies and their long-term effectiveness on global AI leadership.

Key insights

China has demonstrated the ability to train trillion-parameter AI models using only domestic hardware, challenging US export controls.

Principles

• Domestic AI ASIC clusters can support large-scale model training.
• Competitive AI models can emerge despite hardware restrictions.
• Benchmark performance varies across different evaluation suites.

Method

The article describes training a 1.6 trillion parameter model on a cluster of over 50,000 domestic AI ASICs, processing 35 trillion tokens.

In practice

• Evaluate model performance across diverse benchmarks.
• Investigate alternative hardware for AI development.
• Monitor geopolitical impacts on AI supply chains.

Topics

• Large Language Models
• AI Hardware
• China AI
• Export Controls
• AI Benchmarking
• Domestic Chip Production

Best for: Research Scientist, AI Scientist, Director of AI/ML, Policy Maker

Related on AIssential

• China boosts 'AI for Science' computing in 2 months without US chips - Business Standard — China rapidly deployed a large, domestically-chipped AI computing cluster for scientific research, reducing reliance on US technology.
• Meituan Releases New AI Model Trained Entirely on Chinese Chips — Meituan's LongCat-2.0, trained on Chinese chips, signals growing domestic AI hardware independence.
• 🔮 Exponential View #570: Inside Jensen Huang’s worldview — Nvidia's market dominance stems from strategic supply chain orchestration and a deeply embedded software ecosystem.
• Huawei Ascend Production Ramp: Die Banks, TSMC Continued Production, HBM is The Bottleneck — China's drive for AI compute self-sufficiency is bottlenecked by HBM supply, despite significant domestic chip production efforts.
• Meta to Spend Billions on AMD Gear, AI Scare Trade Continues — Diversification in AI chip procurement and specialized AI agents are reshaping market dynamics and software valuations.
• ChinAI #361: What can CANN do for China's independent compute capacity? — Huawei's CANN is maturing, fostering an independent chip-model synergy ecosystem challenging Nvidia's CUDA dominance.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Decoder.