Meituan trains 1.6-trillion-parameter AI model LongCat-2.0 entirely on Chinese chips

Chinese company Meituan has announced that its new AI model, LongCat-2.0, with 1.6 trillion parameters, is the first model of this scale trained entirely on domestic Chinese chips. The company openly frames the move as a response to US export restrictions.

What we know about LongCat-2.0

The most notable thing about Meituan's new model isn't its size, though it is genuinely large. It's the hardware it was trained on. The company says LongCat-2.0 is the first model of this scale trained end-to-end on domestic Chinese chips rather than on processors from Western vendors.

• The model is called LongCat-2.0 and has 1.6 trillion parameters.
• According to the company, it is the first model of this size to complete full training on domestic Chinese chips.
• Meituan explicitly ties the move to US export restrictions on advanced chips.

Why this is a response to US export restrictions

Training a 1.6-trillion-parameter model entirely on domestic hardware is a hard technical problem. It typically requires chips optimized for the specific architecture and compute volume. Meituan ran the whole process end-to-end without Western chips, and the company presents that directly as proof that US export restrictions are not stopping the development of large Chinese AI models. The framing matters too: Meituan positioned the announcement not as a routine technical release but as a "pointed answer" — a direct, targeted response to export control policy. That shows the company sees this news primarily in a geopolitical context, not just a product one.

What sets LongCat-2.0 apart from other large models

Most models of comparable scale worldwide are still trained on chips from Western vendors, which remain the industry standard for training large language models. Meituan's claim that LongCat-2.0 is the first model of its size trained entirely on domestic chips is what makes it stand out against that backdrop. The company emphasizes not the model's architecture but the infrastructure independence of its training process.

What it means

Meituan's announcement goes beyond the launch of yet another large language model. It signals that Chinese tech companies can scale trillion-parameter AI models on their own chip base. If other players confirm and repeat such claims, US export restrictions will gradually lose weight as a tool for containing Chinese AI, and competition in building the largest language models will stop depending on a single chip ecosystem altogether.