Meituan is hard to explain in one line if you have not lived inside China's internet economy.
In the US, you would have to combine DoorDash, Uber Eats, Yelp, Instacart, Groupon, Booking.com, and parts of Amazon's local retail ambition to get close. In India, imagine Swiggy and Zomato mixed with Blinkit, Zepto, MakeMyTrip, local restaurant discovery, coupons, grocery delivery, and everyday city services.
Even that comparison understates it. Meituan is not just an app for ordering food. It is infrastructure for local consumption.
People use it to order meals, buy groceries, discover restaurants, book hotels, find services, buy vouchers, and get things delivered fast. Merchants use it to find demand. Delivery workers use it to turn city movement into income. The company sits between restaurants, riders, stores, consumers, maps, payments, reviews, inventory, and time.
That makes Meituan worth studying now, because the next phase of AI will not be decided only inside chat windows. It will be tested inside companies that already run messy, physical, time-sensitive systems.
Meituan's recent numbers look bad at first glance. According to the South China Morning Post, the company posted an adjusted net loss of 4.97 billion yuan for the three months ended March 31, 2026. That was its third consecutive losing quarter. Competition from Alibaba and JD.com has made China's food delivery and instant commerce market brutal again. When rivals subsidize milk tea, lunches, groceries, and one-hour delivery, margins vanish quickly.
Meituan's own Q1 release said revenue reached 91 billion yuan, while R&D investment rose 22 percent year over year to 7 billion yuan. So the company was fighting an operating war and still funding the technology layer.
Delivery is a difficult business. Consumers want speed. Merchants want demand. Riders want fair pay. Platforms want scale. Competitors want market share. Everyone wants someone else to absorb the cost.
The more interesting detail sat beside the loss.
Meituan disclosed a 7.6 billion yuan gain from investments in companies such as Zhipu AI, also known internationally as Z.ai. The same SCMP report said Meituan held a 3.86 percent stake in Zhipu, and that the stake translated into 24.3 billion yuan in financial gains based on Zhipu's market capitalization. Meituan also owns a 7.61 percent stake in Unitree Robotics, one of China's best-known robotics companies.
That accounting gain does not fix Meituan's core business. It was recorded through other comprehensive income, not normal operating profit. A financial windfall is not the same thing as better unit economics.
But it reveals what Meituan has been buying: exposure to AI models, robotics, automation, and the systems that could make physical services cheaper, faster, and more intelligent.
That is the bigger story.
For the last two years, most AI debate has been trapped inside software work. Will AI replace developers? Will designers use AI? Will support agents become bots? Will every employee have a copilot? Those questions matter, but they are too narrow.
The more valuable question is what happens when AI moves from generating content to controlling work.
Not just: write this email.
Route this order. Forecast this demand. Price this subsidy. Match this rider. Detect this fraud. Recommend this restaurant. Update this merchant profile. Plan this warehouse movement. Coordinate this robot. Reduce idle time. Predict which store will miss the delivery promise before it happens.
That is a different kind of AI.
ChatGPT made people feel AI because it talked back. The companies that may create the most value from AI are the ones with ugly operating problems: logistics, mobility, healthcare, manufacturing, retail, agriculture, insurance, construction, energy, and local commerce.
In those sectors, intelligence is not a nice interface. It is a cost structure.
If AI can reduce failed deliveries, improve route density, automate merchant support, predict stockouts, optimize delivery batching, or make robots useful in constrained environments, the value is not abstract. It shows up in fewer wasted minutes, fewer refunds, fewer manual checks, higher utilization, and better margins.
That is why robotics matters.
A newer South China Morning Post report made the shift even clearer. Two days after Nvidia launched its Cosmos 3 model for physical AI, Hangzhou-based Spirit AI said its embodied intelligence model, Spirit v1.6, became the first Chinese model to top the RoboArena global leaderboard.
Spirit v1.6 scored 1,924. Nvidia's Cosmos3-Nano-Policy scored 1,881. RoboArena is built to test how well general robot policies translate into real-world actions.
That is not just another leaderboard flex. It shows where the AI race is moving. The competition no longer stops at who has the strongest chatbot or the biggest language model. It is moving toward embodied intelligence: software that helps machines act in physical environments.
This is why Meituan's Unitree stake matters. Robotics is not just about humanoids walking on stage. The real question is whether intelligence can move from screens into physical work.
A delivery platform like Meituan already has demand, map data, merchant relationships, rider networks, timing constraints, and customers trained to expect speed. If robots become practical in even small parts of that system, the impact could be large.
Maybe it starts with warehouse movement. Maybe it starts with last-mile experiments in controlled areas. Maybe it starts with restaurant automation, inventory handling, inspection, packaging, or repetitive tasks that do not need the full humanoid dream.
The point is not that robots will replace every rider next year. That is lazy futurism.
The point is that AI becomes more valuable when it has a real operating surface.
Meituan has that surface. So does Amazon. So does Tesla in a different way. So do Indian companies working in food delivery, quick commerce, mobility, logistics, and manufacturing. The companies that already understand demand, routing, supply, pricing, and customer behavior have something pure AI startups often lack: a live system where intelligence can be tested against reality.
This also explains why some AI investments look irrational from the outside.
If you only read today's profit and loss statement, spending more on AI during a loss-making period can look irresponsible. But a company like Meituan cannot win forever by subsidizing lunch. At some point, it needs a structural advantage.
Cheaper capital is not a moat.
Discounts are not a moat.
More riders are not always a moat.
The moat is knowing how to run a city-scale service network better than competitors.
AI and robotics are one possible path to that moat.
For builders, the lesson is simple: the next wave is not "add AI to your app."
That was the first wave. Some useful products came out of it, but a lot of it became wrappers, demos, and pitch decks. The AI had no deep connection to the business.
The better question is: what expensive decision happens here again and again?
Good AI opportunities usually have four traits:
- The decision repeats often.
- A mistake costs real money.
- The workflow already produces useful data.
- A better prediction or action changes the economics.
A restaurant recommendation is nice. A system that knows which restaurant will prepare late, which rider should be assigned, which customer is likely to cancel, which merchant needs operational help, and which route protects the delivery promise is much more valuable.
That is not content generation. That is operational intelligence.
If the decision is frequent, messy, data-rich, and tied to money, AI has room to matter. If the decision touches the physical world, the upside may be even larger, because the physical world is full of waste that software has not fully absorbed yet.
Meituan's AI and robotics bets are interesting precisely because the company is under pressure. They show a pattern more companies will follow: use the cash, data, and distribution from a current operating business to buy or build exposure to the next operating layer.
Some bets will fail. Many will be early. A few will look silly for years.
But the direction makes sense.
Software made coordination cheap. Mobile made demand instant. Marketplaces made supply visible. AI may make decisions cheap. Robotics may make some physical actions programmable.
Put those together, and the next big companies may not look like pure software companies. They may look like operating companies with AI inside the core loop.
That is the mistake in treating AI only as a job replacement story.
AI is not just coming for tasks. It is coming for bottlenecks.
The bottleneck in a delivery network is not one employee writing one email. It is millions of small decisions made under time pressure. The bottleneck in quick commerce is not one product manager writing a roadmap. It is inventory, forecasting, picking, packing, routing, pricing, and customer trust. The bottleneck in robotics is not a viral demo. It is making machines useful enough, cheap enough, and reliable enough to earn a place in daily operations.
That is harder than making a chatbot.
It is also more valuable.
The lesson from Meituan is not "ignore losses because AI will save everything." Losses still matter. Unit economics still matter. Competition still matters. A financial gain from an AI stake does not fix a weak operating business.
The better lesson is that the next AI winners may be companies that understand both intelligence and operations.
They will know where AI should reason, where software should decide, where humans should stay in the loop, and where robots can slowly take over repeatable physical work. They will not treat AI as a feature. They will treat it as an operating advantage.
That is a more interesting wave than chatbots.
And it will be harder to copy.
For anyone trying to understand where public markets may pay attention next, that is the quiet signal. Not a list of tickers. Not a prediction about tomorrow's prices. More a direction of travel: capital tends to notice where technology can change margins, speed, and defensibility.
The AI story after chatbots may gather around robotics, logistics, manufacturing, energy, healthcare, and local commerce because those sectors are full of expensive bottlenecks.
The interesting signal is not one company. It is the bottleneck. Follow the bottlenecks, and you start to see where the market's next AI obsession could form.
Source notes
- South China Morning Post: US$1b windfall: how Meituan's bets on AI, robotics outshine quarterly loss - Meituan's Q1 2026 adjusted loss, Zhipu AI investment gain, and Unitree Robotics stake.
- South China Morning Post: A Chinese robotics start-up beat Nvidia on a global AI ranking. Is a new tech war brewing? - Spirit AI's RoboArena ranking, Nvidia comparison, and the embodied intelligence race.
- Meituan: 2026 Q1 earnings release - revenue of 91 billion yuan and 22 percent year-over-year increase in R&D investment to 7 billion yuan.
- Forbes: Meituan company overview - high-level company description and business segments.