Let’s talk plainly about what this year’s robot congress does — and doesn’t — say. For five days in late August, the largest robot gathering on the calendar ran in the Beijing Economic-Technological Development Area, under the theme “human-machine symbiosis, production and demand shared.” Three hundred and seventy-three companies brought more than three thousand exhibits, and three hundred and eleven of them were world premieres. That is a lot of hardware in one place. But the number that matters most is smaller and easier to miss: this was the first year the congress held a dedicated procurement day.
I follow these events the way a colleague follows a patient’s vitals across a series of visits. You learn more from the trend between appointments than from any single reading, and the trend here is unmistakable. For years, the story of this industry was the demo — a robot dancing, a robot pouring tea, a robot climbing stairs, each clip built to travel across the internet. The demonstrations were not dishonest, exactly. They were just early. What they showed was capability without a customer.
This year the customer arrived. The congress’s first-ever procurement day brought forty-nine state-owned enterprises together to match supply and demand in one room. That single change is worth more than any of the three hundred and eleven premieres. A trade fair that schedules a buying day has stopped being a trade fair and started being a market.
The numbers agree with the signal
The market signal shows up in the numbers too, and I want to quote them carefully. In its August outlook, a major global investment bank roughly doubled its forecast for humanoid robot shipments in one Asian manufacturing economy this year, from twenty-eight thousand units to fifty thousand. The same report noted that more than forty thousand humanoid units had already shipped in the first half of 2026 — around ninety-seven percent of the global total. Those are the bank’s numbers, not mine, and I would take any single-year forecast with a grain of salt. But the direction is hard to argue with.
The official industry figures point the same way. The country’s industry ministry reported that revenue from the robot sector’s above-scale enterprises passed three hundred billion yuan in 2025, with average annual growth above twenty percent over the prior five years. In the first half of 2026 the figure reached one hundred sixty-five and a half billion yuan, up twenty-four and a half percent year on year. Industrial robot output in the first seven months of the year rose twenty-eight and a half percent. Put those together and the pattern is clear: the sector is not just showing machines anymore. It is shipping them.
I have to pause at the word “shipping,” because volume is a clinical sign, not a diagnosis. The honest answer is that high output tells you the market is buying; it does not tell you the machines are good, useful, or safe. This transition — from demo to delivery — is exactly where the risks hide. A product that sells in quantity is a product that will be used in real homes, real factories, real hospitals, real schools. And real use is where the problems that demos hide start showing up: reliability under grime, maintenance by people who did not write the manual, and the question of who is responsible when a machine makes a mistake near a human.
There is one number in that set worth reading slowly, because it is the sort of figure an engineer files away rather than quotes. Ninety-seven percent of global humanoid shipments happening in one manufacturing economy is not a normal distribution of an emerging industry — it is a concentration. Concentration has two faces. On one side, it means a mature supply chain can form fast: components, tooling, training and standards all clustering in one place. On the other side, it means the industry’s fragility is concentrated too — a single disruption to that chain would be felt by the entire category at once. Neither face is new in industrial history; both are worth remembering when the next impressive demo goes viral. Let me add a second opinion, free of charge: concentration is an engine and a single point of failure, both at the same time.
Let me put those official figures in a slightly more humbling frame, because I do not want to be a cheerleader for a chart. Three hundred billion yuan in annual revenue is a real industry, but it is still small against the economies it serves, and a twenty-four percent growth rate is impressive precisely because it is easier at small bases than at large ones. The honest reading is not “robots have arrived.” It is “robots have reached the size where their problems become interesting.” That is the stage where safety rules, service networks and standards stop being optional extras and start being the product. Which is another way of saying: the hard part begins now, not earlier.
It also helps to separate two different machines that share the word robot. Industrial arms — the ones bolted to factory floors — have been selling steadily for years; the recent double-digit output growth in that segment is a continuation, not a surprise. Humanoids are the newcomer: walking machines built to enter spaces designed for people, which is exactly why they carry more risk and more promise. The reason the shipment forecast doubling matters is not that humanoids are better than industrial arms — for most fixed jobs they are not. It is that a humanoid sold in volume signals something the industry has never had: a category priced for ordinary buyers rather than laboratories. Volume is how a technology stops being special — and, with no false certainty about the timeline, how it starts being boringly reliable.
Buyers arrived before the experience did
Let me correct my own earlier framing, because I started this piece convinced the story was “robots finally found customers.” No — that is not quite right. The story is more uncomfortable and more interesting: the buyers arrived before the operating experience did. Forty-nine enterprises agreeing to buy is a commitment to operate, and operating robots at scale, outside controlled stages, is a learning curve nobody has completed yet. The honest answer is that demand has outrun the field data. That is not a reason to panic. It is a reason to watch closely.
There is a historical echo here that I keep turning over. The household washing machine spent years as a parlor curiosity before it became kitchen furniture — early buyers paid for the idea of the machine, then spent a decade discovering what it could not handle: the shirts that shrank, the lace that tore, the load that jammed. What actually fixed the washing machine was not better marketing. It was years of real households using it badly and reporting back. The robot industry is now entering exactly that phase, except the reporting-back is happening in factories and clinics rather than laundries, and the cost of a mistake is higher.
A small scene from the show floor
One image from the event has stayed with me, in the way a single chart can stay with you. A collaborative arm sat on a stand, clamping a steel ball and lowering it into a cup — the same precision trick, performed on a loop, for eight hours. Around it stood a ring of phones, all recording the same forty-second loop, and a ring behind them of engineers who were not recording anything. They were leaning in, watching the wrist joints, asking the salesperson questions about tolerances and duty cycles. The phones were documenting a show. The engineers were pricing a purchase. That split on one show floor is the whole transition in miniature.
The procurement day is worth one more look, because its effects will outlast the event itself. When forty-nine large buyers sit down with suppliers, they do not just write orders; they transmit specifications. A procurement contract tells the supplier exactly what the buyer wants to run for years: the load cycles, the maintenance windows, the spare-part requirements. That is how industrial standards actually get written — not in committees, but in purchase orders. For smaller suppliers who cannot reach those buyers directly, the interesting effect is indirect: the specs published for the big contracts become the de facto benchmark for everyone else. A single procurement day, in other words, quietly becomes a standard-setting session. That is the sort of quiet event an engineer files away and a marketer ignores.
The clinic analogy I promised myself at the start: this industry now resembles a patient who has passed the excitement of a diagnosis and entered the long, unglamorous phase of treatment — dietary changes, follow-ups, side-effect management. Progress in that phase is measured in small numbers: response rates, uptime, incidents per thousand hours. Nobody films it, and it is the phase that decides the outcome. The robot industry has reached its treatment phase. The congress’s next act will not be judged by its premieres but by its service contracts.
Three plain-words conclusions
So what should a person who actually cares about these machines — or will have to work alongside one — take from all this? In plain words, three things. First, treat the demo videos as marketing, not evidence. A robot that juggles on camera tells you nothing about a robot that sorts parcels for ten hours. Second, watch the maintenance story, not the launch story. The machines that win will be the ones a mid-sized factory can keep running, with spare parts that exist and technicians who understand them. Third, keep the human being in the frame. The most impressive robots in this industry right now are the ones still being handled by careful humans every shift — and that is not a criticism. It is a definition of where the industry actually is.
There is a second opinion worth offering, and it is about how we measure progress. I worry less about whether robots can do impressive things — that is now mostly settled — and more about whether the supporting system can keep them honest: standards, repair networks, training for the people who will work with them, and clear rules for what happens at the moment of failure. A robot that works ninety-nine times out of a hundred is a marvel. The hundredth time is where policy, engineering and safety all live, and it is the part nobody shows on the trade-show floor.
Let me also be clear about what I am not saying. I am not predicting robots will take over warehouses overnight, and I am not dismissing the industry’s momentum. I am saying the interesting chapter has started, and it is not the chapter the congress marketed. The three hundred and eleven premieres were the appetiser. The procurement day was the main course — a quiet signal that the machine business has crossed from persuasion to delivery. In plain words: the industry stopped auditioning and started working. The next part of the story will be written in factories, kitchens and corridors, not on stages.
The line I would leave you with is this: no false certainty is worth more than a confident guess — and the confident guess here is that the robots you will meet next are already being shipped, not premiered. The question is not whether the machines are coming. The question is whether the humans around them are being prepared as carefully as the hardware.