Meituan, the Chinese food delivery and services giant, has thrown its hat into the country's intensifying artificial intelligence race. The company announced the release of LongCat-2.0, a massive large language model (LLM) with over one trillion parameters, which it plans to make open-source. What sets this announcement apart is that Meituan says the model was trained from scratch on a cluster of 50,000 chips using domestically produced processors, a direct response to tightening US export controls on advanced semiconductors.
What is LongCat-2.0?
Large language models are the technology behind AI chatbots and generative AI tools. The 'parameters' in a model are essentially the connections that help it learn and generate text, code, or other content. A trillion-parameter model is exceptionally large, putting LongCat-2.0 in the same league as some of the most advanced models from companies like OpenAI and Google. By open-sourcing the model, Meituan is allowing developers and researchers to freely use, modify, and build upon it, a strategy that can accelerate adoption and innovation.
The training process for such a model is enormously demanding. It requires vast amounts of computing power, typically supplied by clusters of specialized chips. Meituan's use of a 50,000-chip cluster built entirely with domestic processors is a significant technical achievement and a clear signal of China's push for self-sufficiency in AI infrastructure.
The US Export Control Context
The announcement comes against the backdrop of US government restrictions that limit the sale of advanced AI chips, such as those made by Nvidia, to Chinese companies. These controls have forced Chinese tech firms to seek alternatives, either by stockpiling chips before the restrictions took effect or by developing their own hardware. Meituan's claim that LongCat-2.0 was trained on a domestic cluster suggests that Chinese-made processors are becoming a viable option for even the most demanding AI workloads.
This development is part of a broader trend. Other Chinese tech giants, including Baidu, Alibaba, and Tencent, have also been developing their own LLMs, often using a mix of domestic and imported chips. The ability to train a trillion-parameter model on homegrown hardware could reduce China's reliance on foreign technology and potentially accelerate the country's AI capabilities.
What It Means for Investors
For everyday investors, Meituan's move has several implications. First, it underscores the intense competition in China's AI sector. Companies are racing to develop the most powerful models, and open-sourcing is a common tactic to attract talent and build ecosystems. This competition could lead to rapid innovation but also high spending, which may pressure profit margins in the near term.
Second, the reliance on domestic chips highlights the resilience of China's tech supply chain. While US export controls were intended to slow China's AI progress, companies like Meituan are finding workarounds. This could support the case for investing in Chinese AI and chip stocks, which have already seen rallies on optimism about domestic alternatives. For context, China AI and chip stocks rallied in June as factory activity returned to growth, reflecting broader economic support for the sector.
However, investors should also be cautious. The massive spending on AI infrastructure by hyperscalers—large cloud and tech companies—has raised concerns about a potential investment bust. The Bank for International Settlements recently warned that the $1 trillion hyperscaler AI spending spree risks an investment bust, suggesting that not all AI investments will pay off. Meituan, primarily a food delivery and local services company, will need to demonstrate that its AI investments translate into real business value, such as improved delivery logistics, better customer recommendations, or new revenue streams.
Looking Ahead
Meituan's entry into the AI race with a domestically trained model is a notable milestone, but the real test will be adoption and application. The company plans to open-source LongCat-2.0, which could help it gain traction among developers. Investors will be watching for signs of how the model is used, whether it leads to partnerships, and how it affects Meituan's core business.
In the broader context, China's AI sector is showing resilience despite external pressures. The country's services sector recently edged back into growth in June, and factory activity also returned to expansion, providing a supportive economic backdrop for tech investments. Meanwhile, the central bank has been doubling overnight cash injections to ease funding squeezes, which helps maintain liquidity for companies investing in long-term projects like AI.
For now, Meituan's LongCat-2.0 is a reminder that the AI race is global, and that Chinese companies are finding ways to compete even with restricted access to cutting-edge hardware. Investors should keep an eye on how this open-source model is received and whether it helps Meituan diversify beyond its core food delivery business.
