In Meituan’s second quarter financial report conference call held on August 28, Meituan CEO Wang Xing made a clear statement on the company’s AI strategy: Meituan will not become a “Token factory” and all AI models and products will focus on supporting the development of the main business. This statement is seen as an important signal for Meituan to stick to the boundaries of its core business in the AI wave.
AI is a comprehensive upgrade of organizations, products and work processes
Wang Xing said at the meeting that AI is a comprehensive upgrade of organization, products and work processes for Meituan. He emphasized: "We always believe that the long-term value of AI capabilities lies in whether it can truly enter real workflows, solve real problems, and form long-term stable infrastructure capabilities."
According to Wang Xing, Meituan’s overall AI strategy currently focuses on three directions: Building LLM (building a large language model), AI at Work (AI empowerment work) and AI in Products (AI integration into products). The company is launching dedicated AI agents in multiple vertical fields such as catering, services, retail, medicine and health, and hotel travel to help merchants optimize daily operations and improve operating efficiency. Wang Xing also said that as AI capabilities continue to penetrate into merchant operations, the relationship between Meituan and merchants is gradually shifting from traditional online channels to AI business partners. The company's long-term goal is to become the "digital co-pilot" of platform merchants.
With the blessing of domestic computing power, the trillion-dollar model is already in place
It is worth noting that Meituan has made substantial breakthroughs in AI infrastructure. At the end of June this year, Meituan officially released a new generation of trillion-parameter large model LongCat-2.0. The model adopts MoE (Mixed Expert) architecture, with a total parameter size of 1.6 trillion, and each Token activates about 48 billion parameters on average. It natively supports 1M ultra-long context and can process millions of word-level inputs at a time.
More importantly, LongCat-2.0 is the industry's first trillion-parameter large model that relies entirely on domestic computing power to complete the entire training and inference process. The model completed full-process training on a cluster of more than 50,000 domestic computing cards and was pre-trained from scratch. Wang Xing specifically mentioned at the financial report meeting: "Nationalized training and reasoning capabilities have also helped us establish stronger advantages in long-term costs and infrastructure controllability."
R&D investment is significantly increased, and AI strategy is pragmatically advanced
Financial report data shows that Meituan’s R&D investment in the second quarter increased by 22.5% year-on-year to 7.7 billion yuan, accounting for 7.3% of total revenue, and the increase ranked first among all expense items. However, Wang Xing also emphasized that Meituan will carefully consider the rate of return when investing in AI and will not blindly invest beyond the company’s financial capabilities.
In the context of increasingly fierce competition in the AI large model track, Wang Xing’s statement clearly delineates Meituan’s AI boundaries - AI is a tool, not an end; it is a means to serve the main business, not a new business.
Comments