Making AI understand the physical world is Meituan's (美团) new competitive battleground

Meituan's offensive strategy: turn local services into an AI-native front

Meituan (美团) is sharpening its AI playbook to win the mental real estate of local life services. CEO Wang Xing (王兴) has framed AI as an offensive opportunity rather than a defensive fix — AI at work, AI in products, and homegrown foundational models. In practice that has meant building its LongCat model, deploying consumer-facing assistants such as XiaoMei (小美) and Tabbit, embedding “XiaoTuan” (小团) in the Meituan app, and rolling out B2B tools like the chain-restaurant assistant “智能掌柜·品牌顾问.”

The contest is no longer just about clicks — it's about decision efficiency

The rivalry has moved beyond coupons and delivery speed. It is evolving into a fight over which platform can make better, faster life-service decisions on behalf of users. It has been reported that Alibaba (阿里) spent roughly ¥3 billion over the Lunar New Year seeding Qianwen (千问) as a life‑service entry, while Tencent (腾讯) is nudging WeChat toward more active Agent capabilities. What matters now is not who drives the most traffic, but who can convert messy, real‑world signals into precise choices for users.

The technical hinge: let AI read the physical world

To outcompete, Meituan insists AI must understand the physical world — maps, POI details, real‑time capacity, menu specialties, and traffic. That requires a digital‑twin style data layer linking online intent to offline reality. Meituan has launched a LongCat‑based Deep Research agent and, it has been reported that, expanded investments across AI chips, compute, embodied intelligence and autonomous delivery (drones and vehicles) to close that loop. Internally the company is pushing merchants from simple digitization to “AI‑era” digitalization so models can act on richer, real‑world context.

Stakes and strategic implications

This is a systems fight: foundational models, new digital infrastructure, agent UX and execution (human or robotic) all matter. Add geopolitics — U.S. export controls on advanced chips and broader tech rivalry — and the incentive to build domestic, end‑to‑end stacks intensifies. Can Meituan keep the user’s living‑services mindshare and translate it into an AI super‑entry, or will it be relegated to a skilled executor within someone else’s assistant? The answer will depend on who best marries scene‑level data with decisioning and fulfilment in the messy real world.