Meituan CEO Declares AI Strategy: Focus on Core Business, Not 'Token Factory'

Meituan CEO Wang Xing announced during the company's Q2 earnings call that they will focus on integrating AI models to support their core business and improve operational efficiency, rather than acting as a raw API provider or 'Token factory'. This statement follows the June release of their trillion-parameter model, LongCat-2.0, which was trained entirely on domestic Chinese hardware. This strategy highlights a shift in how major tech companies approach AI, prioritizing internal workflow integration and cost efficiency over commercializing raw model APIs. It demonstrates a pragmatic path for non-pure-play AI giants to leverage large language models to drive value in local services and e-commerce. Meituan's LongCat-2.0 is a Mixture-of-Experts (MoE) model with 1.6 trillion total parameters (averaging 48 billion active parameters) and native support for a 1-million-token context window. The model was pre-trained from scratch on a 50,000-card domestic GPU cluster, ensuring infrastructure controllability and long-term cost advantages.