Meituan unveils LongCat-2.0 AI, the first large-scale language model trained with Chinese-made chips

Chinese technology giant Meituan presented this Tuesday (30) a new large-scale artificial intelligence (AI) system, which the company describes as the pioneer to be completely developed with microprocessors produced within the Asian country itself. The information was reported by the AFP agency.
This development comes after China launched DeepSeek AI in January 2025, notable for its operating costs that were around 75% lower compared to solutions from other companies on the market.
The current technological scenario is marked by an intense geopolitical dispute between China and the United States in the sector of semiconductors and chips essential for artificial intelligence.
Tensions increased after the United States government, under the leadership of Donald Trump, imposed restrictions on the import of high-performance chips from Nvidia (NVDA), the global leader in the segment, which motivated China to intensify its own investments and efforts to create advanced chips.
In this context, Meituan, which has a diverse portfolio that ranges from food delivery services and accommodation reservations to the development of cutting-edge technologies, announced LongCat-2.0, a large language model that the company positions as equivalent to Gemini 3.1, launched by Google in February.
Large language models (LLMs) constitute the fundamental technological basis for the functioning of chatbots and several other innovative artificial intelligence applications.

Meituan detailed in a statement that LongCat-2.0 is “the industry’s first trillion-parameter model to complete end-to-end training and inference on a national computing infrastructure consisting of 50,000 chips.”
Despite the relevance of the announcement, the company did not disclose the name of the Chinese group responsible for manufacturing the chips used in the development and training process of the new AI.
This launch represents a watershed for China’s AI industry, especially because the training of competitive models, which require massive digital data processing, requires very high capacity and performance chips.
Meituan’s research team began investigating and applying domestically manufactured chips as early as 2023, and, according to the company, “has demonstrated that it is possible to carry out large-scale model training using national computing clusters.”






