---
title: "The market is still focused on food delivery, but Meituan has already started building its next moat."
type: "Topics"
locale: "en"
url: "https://longbridge.com/en/topics/43177236.md"
description: "In the late 15th century, after returning from his circumnavigation voyage, Christopher Columbus found that some European nobles considered his achievements to be mere coincidence. Columbus did not refute them. He picked up an egg and asked those present to try standing it upright on the table. No one succeeded. He gently tapped a small dent into the shell of the egg, placed it steadily on the table, and then said: "This matter is not difficult, but only I know how to do it." Many things sound easy. Ask AI to answer a question, ask AI to sort out a travel itinerary that seems logically reasonable, ask AI to recommend a restaurant to users..."
datetime: "2026-08-04T09:26:15.000Z"
locales:
  - [en](https://longbridge.com/en/topics/43177236.md)
  - [zh-CN](https://longbridge.com/zh-CN/topics/43177236.md)
  - [zh-HK](https://longbridge.com/zh-HK/topics/43177236.md)
author: "[锦缎研究院](https://longbridge.com/en/profiles/2576456.md)"
generator: "portal-rs"
---

# The market is still focused on food delivery, but Meituan has already started building its next moat.

**In the late 15th century, after Columbus returned from his circumnavigation voyage, some European nobles believed his achievements were merely accidental. Columbus did not refute them. He picked up an egg and asked those present to try standing it on the table. No one succeeded. He gently tapped a small dent into the eggshell, placed the egg steadily on top, and then said, "This is not difficult, but only I know how to do it."**

Many things sound easy. Asking AI to answer a question, asking AI to sort out a seemingly logical travel plan, or asking AI to recommend a restaurant to users. But getting AI to go from "knowing the answer" to "actually completing a task," from deduction in the mind to implementation in the physical world, this gap is far wider than imagined.

On July 27, Meituan's local life AI-native assistant, "Xiao Tuan," completed a comprehensive upgrade. The core change in the new version of Xiao Tuan goes beyond search, Q&A, and decision support, further upgrading its agent execution capabilities. After users state their needs, Xiao Tuan can combine real-time information to assist in completing local life service operations such as placing orders, hailing rides, and making reservations.

From "able to answer" to "able to act," a difference of one word, yet worlds apart. On the surface, it is a functional upgrade; at its core, it marks the first step for AI to move from an information tool to an action agent.

Many people have not yet realized that **every truly great opportunity in business history is hidden within these six words: "reducing friction costs."**

We start our discussion with this upgrade of Xiao Tuan: where exactly lies the gap between "able to answer" and "able to act"? What subtle changes are occurring in the local life market? And there is a severely underestimated fact: Meituan's AI assets have not yet received their due valuation.

## **01 From Asking "Xiao Tuan" to Letting "Xiao Tuan" Help**

The core change in Meituan Xiao Tuan 2.0 can be summarized in one sentence: Users no longer need to do it themselves.

For example: Book me a locally distinctive restaurant for four people, with a table reservation at 6 PM, accessible by walking.

Common AI responses: Retrieve relevant information from training data and the entire web, recommend 3-4 restaurants, unable to judge the latest status of the restaurants, and may even produce ambiguous answers, such as categorizing "take the subway first, then walk" as a walking plan. Of course, at the execution level, due to many bottlenecks not yet being cleared, many AIs also lack the ability to actually get things done.

In contrast, Xiao Tuan 2.0, through its own ecosystem, always holds the latest information. Once confirmed, it can directly book the table for you and ensure that requirements such as local characteristics and walking distance are met.

The execution process looks quite simple. It is divided into three steps overall: Screening, combining distance, operating status, user reviews, price, and real-time availability to match suitable plans from massive amounts of the latest merchant data; Recommendation, presenting the screening results to the user, along with the rationale—why this restaurant was recommended, what the reviews are like, and how many seats are still available; Execution, after user confirmation, Xiao Tuan directly assists in completing the reservation.

Besides dine-in, Xiao Tuan also has advantages in the delivery model. We used Xiao Tuan to order a cup of milk tea, adjusting only the flavor preferences. Xiao Tuan automatically selected the milk tea shop with the fastest delivery time and, through multi-platform comparison, chose the most suitable coupon to use directly, achieving the optimal cross-platform price.

Xiao Tuan's long-chain decision-making capability and context memory are also quite impressive, and it is embedded in most of Meituan's service items. Querying, planning, recommending, and directly hailing rides via AI are all connected. Xiao Tuan not only completed address queries and ride-hailing services but also considered itinerary time, solo dining, and queue times, ultimately making the decision that best fit the user's needs.

For merchants that have not yet integrated online booking systems, Xiao Tuan has another more concealed but practical feature: "AI Outbound Calls." After users submit a reservation request, Xiao Tuan can simulate human agents to complete phone consultations and confirmations.

Placing such capabilities onto food, clothing, housing, and transportation—the most basic and frequent daily needs of humanity—carries exceptional significance. In the past, every time users went out to eat, run errands, or meet people, they had to repeatedly search, compare prices, confirm, and place orders. The entire process consumed time and energy. Xiao Tuan 2.0, amidst the most comprehensive and accurate information, helps users not only avoid the hassle of switching back and forth between pages but also finds more suitable solutions during the matching and execution processes.

**In our view, the new version of Xiao Tuan might currently be the most practical C-end AI application. It doesn't have explosive benchmark scores or stunning multimodal demonstrations, but it has achieved something no other large model company has done: it has inserted AI into a transaction and review pipeline that has been genuinely operating for over a decade.**

## **02 An Underestimated Gap: The Digital World Speaks, The Physical World Acts**

Earlier this year, "Physical AI" has emerged in a striking manner. How to reach the physical world is an issue that all AI companies must face at this stage. Gartner recently listed it as one of the major technology trends for 2026, while Nvidia has begun to praise it as the next development stage of artificial intelligence.

Figure: Excerpt from Gartner report, mentioning the necessity of reaching the real world, Source: Gartner

Aside from robot products understood in a broad sense, what is actually easier to reach the physical world are links such as autonomous driving and social services. However, for AI to truly reach the physical world, the complexity far exceeds ordinary understanding. The physical world is dynamic, with countless corner cases that cannot be covered in advance through machine learning.

Whether a restaurant is still open, if there are empty seats tonight, how long the current queue is—these pieces of information change every moment. If elevated to the fulfillment level, the requirements for AI systems' real-time response, low latency, and local data quality in physical scenarios are extremely stringent.

AI in the digital world can rely on cloud computing power for repeated deductions, but execution in the physical world does not allow for "trying again." If a reservation fails, it fails; if a ride is missed, it is missed.

**Here lies a rarely discussed truth: The arms race in the AI industry is almost entirely concentrated on "smarter brains." But what the physical world needs, besides smarter brains, is also having truly "been there and seen that" and a pair of hands that can actually work.**

Xiao Tuan's ability to achieve physical world reach first relies not on higher model benchmark scores. Its trump card is Meituan's long-term accumulation of merchant, review, transaction, and fulfillment data, plus its delivery and service fulfillment network. These two things provide the foundation for AI to organize supply information into answers and then connect subsequent actions within the scope of product support.

First, data quality. For local life service AI, the biggest fear is not insufficient computing power, but "untrue information." A restaurant's review might be from half a year ago, its operating hours might have changed, and the number of empty seats changes every minute. Pure large model companies can generate beautiful recommendation copy, but they don't know if this shop has any seats left tonight.

Meituan's data foundation determines the "authenticity" of Xiao Tuan 2.0. To date, Xiao Tuan's information services have covered over 2,800 cities nationwide, cumulatively completing more than 700 million merchant information verifications，收录 ed over 1.3 billion user reviews, and built the dataset closest to the real physical world around the real-time data network of capacity and dine-in businesses.

Second, fulfillment capability. Xiao Tuan can assist in completing reservations, placing orders, and hailing rides, not just because it can understand user needs. More fundamentally, the reason is: Meituan has a delivery and service fulfillment network covering local life services, capable of undertaking every action command initiated by AI. Food delivery capacity, mobility capacity, merchant service networks—pure large model companies without physical world execution capabilities, even if they can generate equally precise recommendations, cannot truly get things done when users say "help me book." 

**Meituan's moat in the physical world was paved inch by inch over more than ten years. This moat was not dug out by technological leadership, but brewed by time. And barriers built by time are precisely the hardest to overturn.**

Going from "able to answer" to "able to act" seems like AI just took one extra step. But this step, like Columbus standing the egg on the table, is one minute on stage, ten years of work off stage.

## **03 The Essence of Business is Reducing Friction Costs**

Looking at it over a long cycle, the history of commercial development is essentially an evolutionary history of constantly balancing friction costs.

The essence of city birth is to reduce human transaction costs: information exchange costs, transportation costs, social interaction costs. Commodity markets can effectively reduce transaction costs in microeconomic operations, mutually reinforcing urbanization in the long term.

The Internet took this baton, further erasing offline spatial decision costs. The essence of the platform economy is to reduce search costs, negotiation costs, payment costs, and trust costs. In the past, users needed to personally walk into business districts and compare prices store by store; later, users only needed to open their phones and switch between several apps to search, compare, and place orders. The spatial cost of decision-making was compressed, but the time and energy costs of decision-making itself still existed.

The upgrade of Xiao Tuan 2.0 takes a step forward along the same logic line. For today's largest and most complex local consumer market, a reshaping brought by AI is taking place. And the core goal of this reshaping is the invisible decision burden in consumers' minds.

The decision cost of service consumption is precisely the friction point most easily ignored in the current consumption chain. The "Insight Report on Chinese Catering Dine-in Decision and Evaluation System 2026" released by the Red Catering Industry Research Institute clearly points out that consumers' dine-in decision paths have become longer and more cautious. Accenture's 2024 Global Consumer Survey showed that as high as 76% of Chinese consumers gave up purchases due to too many choices or decision difficulties.

Figure: Excerpt from Accenture's "Redefining Relevance in the Age of Information Overload" report, Source: Accenture official website

The information carrying capacity of internet platforms is already overloaded. Flavors, per capita consumption, dining environment, coupons—users need to switch repeatedly between large and small screens to complete price comparisons and place orders. As of the end of 2025, China's population aged 60 and above reached 323 million, accounting for 23% of the total population. Among these 300+ million people, how many can skillfully use mobile phones to complete a series of complex price comparison decision actions?

**The true value of Xiao Tuan 2.0 is hidden in this question: It lowers the decision threshold for service consumption. Allowing more groups—not just young people, but also those who are not good at using complex apps—to complete a consumption through a single natural language sentence.**

AI advances from "information matching" to "transaction execution," and users transform from "operators" to "confirmers." This change may rewrite the underlying logic of local life consumption.

Humans are generational. When a more convenient mode appears, people find it hard to return to the past. Today's young people cannot understand a world without the Internet, just as future people may not understand consumption decisions requiring repeated searching, price comparing, and confirming.

**Xiao Tuan 2.0 stands at the starting point of this change. It is not building a better search box; it is digging a canal straight to the destination in the phones of a billion people. Once the canal is built, no one will be willing to climb over mountains and cross ridges again.**

## **04 Meituan's AI Assets Have Not Yet Received Their Due Valuation**

By the time we reach here, the answer to the opening question has surfaced.

Meituan's dedication to letting AI take over more intermediate links is 顺应 ance to the laws of commercial evolution: Every leap in commercial development is another compression of transaction friction.

From a capital perspective, Meituan's accumulation of AI assets is much thicker than what the market sees. Li Auto, Zhipu, Unitree, and Moonshot AI, which shone brightly this year, all have Meituan's shadow behind them. Internally, Meituan's investment is equally significant. At this year's earnings call, Wang Xing revealed that Meituan made large-scale investments in capital expenditure and AI talent, "Apart from enterprises with cloud computing businesses, Meituan's investment scale in AI is likely the largest among domestic enterprises, and it has persisted in layout for over three years."

With "investment" and "research"叠加， Meituan's asset thickness in the AI field has far exceeded what should be configured for a local life company. And Xiao Tuan 2.0 is one of the results of the centralized internalization of these assets.

**But when the market prices Meituan's AI assets, its eyes only stare at investment floating profits. The true value is hidden underwater: Massive real-world physical data resources, a huge fulfillment system operating for over a decade, and organizational determination recognized by everyone from the CEO to business lines that "AI is the first priority."**

These assets have no benchmark scores, no multimodal demonstrations, no shocking benchmark test result sheets. But they have a characteristic that other large model companies can only look on in envy: **They have already been embedded into the real lives of hundreds of millions of users.**

The market has recently begun to re-examine Meituan's value, and its stock price has rebounded. But this is only the beginning of re-pricing. Meituan has done many things in AI, most of which are still hidden underwater.

**History repeatedly verifies one principle: The value of Columbus's circumnavigation voyage will not be extinguished by the shortsightedness of nobles. When the canal is already built and the ship is already launched, the remaining matter is just a question of time. And time always stands on the side of the person who builds the canal.**

### Related Stocks

- [03690.HK](https://longbridge.com/en/quote/03690.HK.md)
- [NVDA.US](https://longbridge.com/en/quote/NVDA.US.md)
- [83690.HK](https://longbridge.com/en/quote/83690.HK.md)
- [MPNGY.US](https://longbridge.com/en/quote/MPNGY.US.md)
- [NVDL.US](https://longbridge.com/en/quote/NVDL.US.md)
- [07788.HK](https://longbridge.com/en/quote/07788.HK.md)
- [07388.HK](https://longbridge.com/en/quote/07388.HK.md)
- [NVDY.US](https://longbridge.com/en/quote/NVDY.US.md)
- [NVDD.US](https://longbridge.com/en/quote/NVDD.US.md)
- [NVDX.US](https://longbridge.com/en/quote/NVDX.US.md)

## Comments (2)

- **一根葱 · 2026-08-05T06:53:12.000Z**: Too flashy. Douyin Group Buying is already eating up your market share. Do you know why? Because the items on Douyin Group Buying are more expensive than Meituan right now.
- **陈鹿鸣 · 2026-08-05T04:11:06.000Z**: Features like Meituan Xiaowen, Meituan can do it, and Douyin can do it too. However, some aspects are already beginning to shake Meituan's ecosystem at a micro level; in fact, Meituan has no real moat. I currently operate many stores, and among them, my impression of Meituan is the worst. They have 


---
> **Disclaimer: This article is for reference only and does not constitute any investment advice.**