Meituan does not operate a "Token factory"

"We have always believed that the long-term value of AI capabilities lies in whether it can truly integrate into real workflows, solve practical problems, and form long-term and stable infrastructure capabilities."

At the Q2 2026 earnings call, Wang Xing clarified the offensive direction of Meituan's AI strategy: "We will not become a Token factory. Our models and AI products will be used to support our core business, improving user and merchant experience as well as the company's operational efficiency."

In the second quarter, Meituan continued to strengthen its AI capability building. The company's quarterly R&D investment rose 22.5% year-on-year to RMB 7.7 billion, accounting for 7.3% of total revenue.

On the AI front, Meituan's strategy has always been proactive rather than defensive. In terms of internal AI transformation, Meituan went through a trial-and-error period of rapid progress, completing the first phase of popularization of AI knowledge through encouraging all employees to build small AI applications internally, then quickly corrected its course, achieved iteration through the three-in-one model of "business design + organizational transformation + technology application", and precipitated mature experience to share with the whole industry after the model was verified.

This is exactly what Wang Xing said at the earnings call, "AI represents a comprehensive upgrade of Meituan's organization, products and workflows."

The good news is that the food delivery business has started to recover. Over the past year, the food delivery sector was mired in a subsidy war, with almost all players struggling to stay afloat amid losses. Now the industry has returned to rationality, and competition has shifted to the race for service quality, efficiency and fulfillment capability. Meituan's long-accumulated advantages in data algorithms and refined operation capabilities have begun to show their strengths.

In the second quarter, Meituan recorded revenue of RMB 104.6 billion, up 14.4% year-on-year, and the group turned to operating profit from loss quarter-on-quarter. Among them, the core local commercial segment saw a notable profit recovery, posting an operating profit of RMB 5.7 billion in the quarter, a sharp turnaround from the RMB 2 billion loss in the first quarter.

After 9 consecutive months of losses, Meituan has hit an inflection point. However, the current recovery is only at the single-quarter level, and the sustainability of profitability remains to be verified over time. As the competitive pressure in the food delivery sector eases, how Meituan can implement AI technologies in scattered and complex offline scenarios, integrate AI into real business details and truly achieve efficiency improvement has become a clearer proposition for its future development.

A

In the large model track of this year, selling Tokens has become a prominent business. Not only large model vendors, but also mainstream internet companies have opened their APIs one after another, taking model invocation volume as a core indicator.

Does Meituan have the strength to sell Tokens? It does have solid technical foundation. LongCat 2.0 released in June, with a total parameter volume of 1.6 trillion, is the industry's first trillion-parameter model that completes the whole process of training and inference on a 100% domestic computing power card cluster. It performs outstandingly in code writing and Agent invocation scenarios, and has been widely verified by the global developer community.

But Wang Xing made it clear at the earnings call that "we will not become a Token factory".

Wang Xing's choice is not hard to understand. The business of selling Tokens is essentially an arms race of computing power. You deploy more GPUs, I deploy even more; you cut prices, I cut even deeper. For Meituan, whose core business is local life services, this is an endless war of attrition with unclear input-output ratio.

Offline local life scenarios are extremely scattered, with massive long-tail variables in every link including rider scheduling, merchant operation and instant fulfillment, making it difficult for general large models to directly adapt to the complex offline business environment.

Wang Xing has also emphasized on multiple public occasions that "the digitalization of the physical world is the most important foundation of AI". What Meituan aims to do is to act as the connector between the digital world and the physical world, and finally implement AI in the real world. "This is where our strengths lie."

Meituan has the largest and most complete offline interaction scenario network in China. Hundreds of millions of users and tens of millions of daily orders continuously generate data related to fulfillment, consumption and merchant operation, which together form a "training ground" for AI to evolve continuously in the real world.

Therefore, Meituan AI has found its own battlefield, deeply embedding AI into its own ecosystem to improve operational efficiency.

At the Q2 earnings call, Wang Xing introduced the progress of Meituan's AI strategy, which is carried out around three directions: Building LLM, AI at Work and AI in Products.

For the user side, Meituan launched its AI assistant "Xiaotuan" and fully upgraded it. In addition to the functions of search, Q&A and decision assistance, it further enhanced its agent execution capability, evolving from "helping users find answers" to "helping users get things done". During this year's May Day holiday, the services related to "Xiaotuan" recorded a total of over 3.9 billion exposures.

For the merchant side, Meituan launched the full-scenario AI Agent platform CatPaw, which equips merchants with professional AI helpers, helping them implement AI in high-value business scenarios and effectively improve operational efficiency. The platform has been verified in multiple real scenarios including catering, beauty industry and pet hospitals.

For the rider side, Meituan also launched the AI safety assistant "Tuanbao", which is equipped with the traffic safety map function, providing riders with multiple services such as traffic light waiting reminders, accident-prone section alerts, intelligent speed reminders and retrograde recognition.

B

While Meituan AI is making clear investments, its core food delivery business has also hit an inflection point.

The food delivery war was once regarded as the "last battle" of the internet industry. The instant retail battlefield behind it was once highly expected to become a new growth pole for internet platforms.

But after a year of fierce competition, this scuffle that burned a total of RMB 200 billion in subsidies did not improve the financial reports of any player. Instead, it became a "money pit" that dragged down overall performance. The subsidy war pushed up the fulfillment and marketing costs of the whole industry. The scale expanded, but profits vanished.

The reason is that both food delivery and instant retail are businesses with strong scale effect. The more orders there are, the higher the utilization rate of fulfillment infrastructure, the lower the cost per order after amortization, and the profits will come accordingly. However, under the subsidy war, the logic of UE (unit economic model) was disrupted. Platforms kept spending money to grab orders, with subsidies far exceeding the cost savings per order. As a result, the more orders increased, the more losses they suffered.

In other words, the subsidy war pulled the food delivery business from the path of "earning small profits per order" to the path of "achieving success through massive investment". The growth that should have been achieved through efficiency improvement, supply-demand matching and organizational optimization has turned into growth driven by burning money.

That's why Meituan's biggest dilemma in the past year was that as the defender of its market share, it was forced to give up profits, which became the cost of defense. Attackers only needed to spend money on incremental orders, but Meituan, as the defender with a larger market share, had to bear the subsidy cost for a huge number of existing orders simultaneously if it wanted to maintain its market share.

Under the rival's massive money offensive, Meituan's market share would be eroded if it did not follow the subsidy policy, but its profits would decline if it did follow.

However, the food delivery business seems to have no threshold, but in fact its biggest threshold lies in sophisticated operation. With years of in-depth experience in the industry, Meituan still maintained its market share advantage with far less losses than its rivals.

The intervention of regulators and the financial pressure brought by continuous money burning jointly pushed this war to gradually cool down, and the industry finally ushered in a window of returning to rationality. After the logic of exchanging volume with money was stopped, the industry returned to a more essential dimension of competition: whether the UE can be positive.

This just brought the competition back to the track that Meituan is most familiar with. Over the past decade, Meituan has grown through rounds of fierce competition, focusing on squeezing profits from every single order. Optimizing the distribution network, improving algorithm scheduling efficiency, and accurately allocating subsidies are all areas where Meituan excels.

In the second quarter, the UE of Meituan's food delivery business turned positive, which is the concentrated realization of these capabilities after the withdrawal of subsidies.

The financial report mentioned that as the whole industry shifts from subsidy-driven to efficiency-driven, the company's order structure and user structure continue to improve, and core user stickiness is enhanced. Meanwhile, dynamically optimized marketing strategies reduced user incentive expenses, and seasonal factors also lowered the distribution cost per order.

A key signal is that sales and marketing expenditure reached RMB 24.7 billion, up 11.5% year-on-year, but its proportion in revenue decreased by 0.6 percentage points to 23.6%.

While the food delivery business hit an inflection point, all of Meituan's businesses also achieved healthy growth simultaneously.

In the core local commercial segment, the sales revenue of self-operated categories such as pharmaceuticals and alcohol reached RMB 3.6 billion, up 78.9% year-on-year. In the new business segment, commodity sales revenue reached RMB 23.07 billion, up 45.1% year-on-year.

Among them, the number of operating cities for Xiaomai Supermarket has expanded to 68, and the number of offline Happy Monkey stores across the country has reached 40.

At the same time, the losses of new businesses are also narrowing. In the second quarter, Meituan's new business segment revenue increased 25.0% year-on-year to RMB 33.1 billion, with segment operating loss narrowed to RMB 1.7 billion, and the operating loss rate improved year-on-year to 5.3%.

However, it should be noted that the current industry recovery is still in the early stage. At the earnings call, Meituan's management said that the industry's subsidy level is still much higher than that in 2024, and it will take several quarters for the industry to return to normal.

Whether Meituan can convert the single-quarter recovery into long-term and stable growth and profitability advantages still needs continuous verification. At the earnings call, Meituan's management looked forward to the third quarter, saying that the food delivery UE will improve significantly year-on-year, but will decrease quarter-on-quarter due to seasonal factors, and is expected to remain positive.

C

At present, Meituan has achieved phased profitability recovery, and short-term financial pressure has been alleviated. But for the market and the public, what they care more about is: what will support Meituan's next step of growth?

Meituan's answer is the physical AI execution infrastructure.

The capital market prefers grand technical narratives, but Meituan has chosen another path: an AI story focused on improving internal efficiency.

No matter how powerful AI is, the actions implemented in the real world still need to be executed by some entities, which is the opportunity Meituan has spotted.

Wang Xing's judgment is that "even though large models are getting smarter and smarter, we can imagine that even if Einstein worked as a secretary, if you asked him to book a restaurant, he still wouldn't know whether there are available seats in that restaurant. This is not an intelligence problem, but an information problem."

At the Q1 2026 earnings call, Wang Xing also mentioned that in the future, in addition to serving consumers (To C) and merchants (To B), serving AI Agents (To A) is becoming more and more important for Meituan.

In May this year, Meituan launched the "Errand Service Skill", opening its errand order placing capability to the AI assistant ecosystem in the form of encapsulated Skills. After users express their demands to the AI assistant, the system will directly call Meituan's errand service to complete order placing, order receiving and delivery, without the need to open the Meituan App throughout the whole process.

For the merchant side, Meituan launched the full-scenario AI Agent platform CatPaw to equip merchants with professional AI helpers. Relying on Meituan's long-term accumulation in the local life services sector, CatPaw provides enterprise-level Agent development and hosting capabilities, helping merchants implement AI in high-value business scenarios and effectively improve operational efficiency. The platform has been verified in multiple real scenarios including catering, beauty industry and pet hospitals.

This also means that Meituan has redefined AI Agent from a "tool" to Meituan's "customer". What Meituan is doing now is to use AI technology to upgrade its heavy assets into the infrastructure of the AI era.

At present, traffic entrances are shifting to AI. Meituan is advancing on two tracks: on the one hand, it is building "Xiaotuan" and "Xiaomei" as its own AI entrances to help users form the habit of "asking Meituan first for local life services"; on the other hand, it is opening up its capabilities to become an indispensable execution layer for external AI assistants.

The entrance is shifting, but the execution capability is very difficult to shift. Meituan hopes to seize the initiative in the execution link of the physical world.

At the earnings call in the fourth quarter of last year, a Goldman Sachs analyst asked: "As the era of AI super entrance begins, will Meituan lose its position as the traffic entrance in the App era?" Wang Xing replied: "The key to AI super entrance lies in accurately understanding user demands and efficiently executing tasks, which is far more complex than chatbots. We will strive to upgrade the Meituan App into an AI-powered App first, and become the AI entrance for local life demands in the future."

While everyone is competing to be the first entrance in the AI era, Meituan has chosen to become the execution infrastructure that enables AI instructions to be implemented in the real world. This may not be the most eye-catching choice, but it is solid enough.

This article is from the WeChat official account "LetterRank" (ID: wujicaijing), author: Zhang Lin, published with authorization from 36Kr.