On June 2, Unitree Technology's status changed to 'filing for registration,' taking just 73 days from acceptance to approval.

The market is easily drawn to such speed. After all, in the early stages of humanoid robot industrialization, the IPO progress of a leading company is itself seen as an industry signal. In contrast, the shareholder structure in the prospectus hasn't received the same attention: Meituan-affiliated entities hold a combined 9.65%, making them the largest external institutional shareholders.

In other words, behind this high-profile humanoid robot company stands the largest external investor—a company better known for food delivery.

Around the same time, Meituan released its Q1 2026 earnings. During the reporting period, it recorded RMB 91 billion in revenue, up 5.6% YoY; operating losses narrowed by RMB 9.6 billion to RMB 6.5 billion from RMB 16.1 billion in the previous quarter. Core local commerce lost RMB 2 billion, while new businesses lost RMB 2.1 billion—both seeing reduced losses.

These two events together reveal at least two layers of meaning. First, the short-term competitive pressures in local life services haven't continued to spiral out of control, and Meituan's losses are narrowing. Second, even under pressure from its core business, Meituan remains both capable and willing to maintain long-term investments for the future.

Over the past eight years, Meituan has heavily invested in chips, large models, and embodied intelligence, accumulating early stakes in a number of leading tech companies. Meanwhile, it has spent over a decade building a nationwide instant delivery network covering more than 2,800 cities and counties offline. This network is now one of the best testbeds for physical AI deployment.

This isn't just a story of 'Meituan doing AI too.' The AI industry is moving from the cloud to the ground, shifting from model capability competition to real-world scenario delivery. Those who control user demand, real transactions, fulfillment networks, and physical world data are more likely to gain pricing power in the next phase.

Viewed through this lens, Meituan's value may need to be reunderstood—shifting from a food delivery platform and local services platform to a 'digital gateway for the physical world.'

Trend: AI Dividends Are Flowing Downstream

To judge where a industry's money is flowing, looking at history is often more reliable than making predictions. Over the past century, every shift in information infrastructure has followed a roughly three-tiered pattern for monetization.

The first tier is infrastructure—the 'shovel sellers'—who capture early dividends. The second tier drives down unit costs through technology, making information increasingly affordable and profiting in the process. The third tier is the gateway, positioned between infrastructure and users, determining which services users access and how. Among these three, gateways typically capture the largest share.

During the telegraph era, early winners were companies laying undersea cables—without them, information couldn't cross oceans. In the telephone era, lines, switches, and long-distance networks were critical infrastructure. During the mobile internet era, carriers building base stations and selling data plans reaped the first wave of profits. In the AI era, this path has shifted to data centers, GPUs, and computing clusters.

Over the past two years, computing power and large model companies have been the first to be revalued by capital markets. The competition centered on who had more GPUs, could train larger models, or reduce inference costs. This phase resembles the early days of the gold rush, where shovel sellers, road builders, and utility providers profited first.

But the pattern in tech history is clear: while infrastructure is essential, it rarely monopolizes final value. After telegraph networks were established, commercial services were truly packaged by distributors like Reuters. Following telephone network popularize (popularization), the lasting commercial value came from gateway businesses like Yellow Pages and enterprise services, which evolved into today's companies like Dianping and Salesforce. The same happened in the data era: telecom operators once saw high market caps due to data, voice, and SMS, but companies standing at user gateways—Tencent, Alibaba, ByteDance, and Meituan—ultimately captured greater commercial value behind the traffic.

AI will likely follow this path. Tokens represent the smallest billing unit in this generation of information technology, similar to telegraph fees, phone bills, and data charges. Early attention focused on whether tokens could generate sufficiently intelligent answers; the next phase will prioritize two questions: how to make tokens cheaper and how to integrate them into real businesses. The former corresponds to model compression, inference optimization, and edge AI; the latter to gateways, scenarios, data, and execution capabilities.

This trend is especially evident at the application layer. Unlike the mobile internet era, where network effects dominated—where more users meant lower marginal costs and scale itself became a barrier—AI reverses this logic: more users mean more computing power burned, leading to greater losses. Scale no longer naturally forms a barrier.

Meanwhile, foundational models have become powerful enough to directly cover writing, video, and Q&A scenarios. Application-layer advantages are often erased by a single update from base models. Additionally, users face negligible switching costs between AI products, lacking stickiness.

When software no longer constitutes a barrier, truly scarce assets become exclusive data, unavoidable gateways, and implementable real-world scenarios. Overseas signals already reflect this: while OpenAI, a pioneer in this wave, saw revenue grow without profit, Anthropic caught up by offering enterprise services embeddable into workflows, with annualized revenue once surpassing $30 billion.

What's the biggest bottleneck for AI to truly enter the physical world and replace human labor? The lack of authentic data.

Data generated in simulated environments suffers from a 'Simulation-to-Reality' gap, becoming inaccurate in real-world settings. Data learned from videos lacks physical feedback—robots don't know if the ground is soft or hard when stepped on. What's truly irreplaceable is massive amounts of real-world scenario data with physical feedback.

Unitree's Wang Xingxing is clear: the biggest bottleneck for robots is insufficient generalization. The solution is straightforward—deploy tens of thousands of robots to collect ten hours of data daily, solving data scarcity. The funds raised in Unitree's IPO will likely primarily train robots' 'brains,' which require data generated in real physical scenarios.

Looking back, Meituan has spent over a decade building a heavy business: connecting users' instant demands to nearby merchants and completing delivery through a fulfillment network. The more complex this network, the closer it is to the real world. Complexity once meant cost; in the AI era, it means data, scenarios, and gateways.

The real world doesn't conform to model training set formats. Urban traffic, restaurant meal preparation speeds, user preferences, warehouse inventory, and courier routes all vary. For AI to function here, linguistic capabilities alone aren't enough—it must understand transactions, supply, fulfillment, and feedback.

This is precisely where companies like Meituan excel.

Strategy: Securing the Foundation of the Physical World

Many companies frame AI as a new business line, as if adding a model assistant could restart growth narratives. Meituan's AI story stays closer to its roots: local life services, instant retail, merchant operations, and fulfillment networks.

Wang Xing mentioned on the earnings call that Meituan views AI as a strategic opportunity, aiming to upgrade the Meituan App into an AI-powered platform that fully covers local life and instant retail experiences. Wang Puzhong's stance is more specific: building an AI foundation for the physical world to help every merchant adopt their own AI assistant.

Together, these statements outline Meituan's dual AI strategy: one line targets consumers, transforming the App into an interface that truly 'understands demands and completes tasks'; the other targets merchants, gradually handing over operational decisions previously requiring human judgment to AI.

This vision is supported by years of sustained investment. Since early 2023, among Chinese companies without cloud businesses, Meituan likely has the largest AI investment scale, maintaining this commitment for over three years. Investments fall into two main areas: developing proprietary foundational large models and turning models into usable products for consumers and merchants.

For consumers, Meituan focuses on two distinct but aligned assistants. 'Xiaotuan' offers comprehensive services, supporting over 100 million users during the 2026 May Day holiday; 'Xiaomei' emphasizes completing selections to orders in one sentence, minimizing interaction costs; in health scenarios, 'Xiaotuan Health Manager' covers online consultations, medication purchases, and appointments, extending AI assistance from dining to healthcare.

Their commonality lies in embedding directly into real daily needs like eating, entertainment, buying medicine, and seeking care, rather than existing as standalone services.

For merchants, Meituan's logic is to 'help merchants adopt their own AI assistants.' 'Smart Shopkeeper' serves over 700,000 catering (catering) merchants, while 'Digital Staff' supports over 300,000 retail merchants. In the hotel industry, 'Jibai' AI has been validated across different scenarios. These products share a similar logic: transferring operational decisions like scheduling, pricing, and promotions—previously requiring human expertise—to AI, lowering merchants' operational barriers.

Drones, autonomous vehicles, and robots push these capabilities into clearer physical execution layers. To date, Meituan's drones have operated routinely in Beijing, Shanghai, Shenzhen, Hong Kong, Dubai, and other cities, completing over 900,000 commercial orders—ranking second globally. Its self-built 'Urban Low-Altitude Air Network' now operates routinely and is open to industry-wide collaboration, with single landing sites handling up to 423 daily orders at peak and enabling 10-minute delivery for short-distance routes.

Meituan's path reverses that of most AI companies. The typical approach is developing models first, then seeking scenarios for validation; Meituan starts with scenarios, which then generate what models need. This sequential difference is hard to replicate quickly with money. Meituan's Robotics Institute has already established over 40 research collaborations with 20+ global universities, with findings accepted by top conferences like ICRA and IROS, showing this path advances not just commercially but also academically.

In Q1, Meituan reduced losses by nearly RMB 10 billion quarter-over-quarter. The significance of this reduction extends beyond the income statement—when food delivery losses no longer dominate attention, the market gains space to reevaluate the company, seeing how it has allocated its accumulated cash and scenarios into physical AI. From this perspective, the loss reduction inflection point truly opens a window to reunderstand Meituan.

Ecosystem: Supporting AI Tech Companies Across Half of China

To date, Meituan has invested in over 28 unicorn companies, with 7 already listed, covering five fields: AI large models, embodied intelligence, semiconductors, autonomous driving, and AI smart hardware. Key investments include Zhipu, Unitree, Yinhe General, Moonshot AI, Li Auto, Moore Threads, and Muxi Corporation (Muxi shares). The true weight of this list lies in how it nearly sweeps up the frontrunners in this wave of hard tech.

Meituan's tech investments follow a crucial timing difference from major corporations' general pace. After going public in 2018, Meituan shifted its investment focus from consumer goods to hard tech, accelerating further after establishing its 'Retail + Tech' strategy in 2020. While some companies still prioritized gaming and retail investments at the time, Meituan allocated over half its funds to hard tech seemingly unrelated to its core business.

In embodied intelligence alone, Meituan has invested in at least 16 companies, mostly entering early, with 10 now valued at over $1 billion as unicorns. Yinhe General is a typical (typical) case: currently valued at approximately RMB 21 billion, Meituan has been its angel investor since inception. For companies like Xinghaitu and Zibianliang—both valued at over RMB 10 billion—Meituan also participated at Series A.

Unitree is the most recent addition to this portfolio that the outside world has noticed. At the time of approval, Meituan-affiliated entities held a combined 9.65%, making them the largest external institutional shareholders. Meituan led Unitree's Series B2 round at a $1 billion valuation and later participated in Series B3. Other major firms generally invested in Unitree much later, entering only after its humanoid robots had performed on CCTV Spring Festival Gala, its valuation surged past $10 billion, and its IPO path became clear.

When it comes to investing, the difference between an early and late round lies in judgment and resolve.

In AI chips and computing power, Meituan's investments include Moore Threads—a leading domestic GPU company with a market cap exceeding RMB 300 billion on its debut—and Muxi Shares, which debuted with a market cap over RMB 280 billion. Other chip investments include AiXi Zhiyuan, Rongxin Semiconductor, and others valued at over RMB 10 billion. In AI large models, Meituan invested in Zhipu—one of China's earliest large model companies and the 'first global large model IPO'—and Moonshot AI at a very early stage, which now exceeds $20 billion in valuation, with Meituan's Longzhu leading a $2 billion new funding round. In autonomous driving, Meituan is the largest external shareholder of Li Auto and has invested in companies like Hesai Technology, Jiushi Intelligent, and Qingzhou Intelligent Navigation.

Across core tracks like AI chips, large models, and embodied intelligence, Meituan hasn't just covered these areas but has accompanied these companies since their early stages.

AI chips require large-scale testing, and robots need real-world environments for training—both needing testbeds, which Meituan provides as one of the platforms with the most consumer scenarios.

Since 2024, Meituan and Yinhe General have collaborated on robot services in offline retail, smart warehousing, and smart logistics. By 2025, over a dozen Beijing pharmacies used its robots for 24-hour medication sorting, with nationwide expansion underway. Late last year, Hesai's perception and positioning LiDAR secured a mass production nomination from Meituan's drones, tying LiDAR to low-altitude logistics scenarios. Unitree Technology connects robot rentals through Meituan's local services.

The synergy between Meituan's scenarios and invested companies' technologies in physical AI deployment represents a dynamic, long-term variable. The market may not yet grasp its value, but that doesn't mean it doesn't exist.

Epilogue

Most people view Meituan through the lens of food delivery and local hotel/travel services—metrics from an old framework. But the AI era has arrived.

Unitree's IPO progress has brought Meituan's alternative asset sheet into public view: where Meituan has placed its bets, at what stages, and the depth of synergy with invested companies. These details, once scattered, now have a centralized window for presentation.

This also prompts the market to reunderstand Meituan's 'physical AI' logic and reassess its position in the AI era.

*Featured image and illustrations in the text are sourced from the internet.