10,000 Humanoid Robots Have Shipped. Their Real Work Begins Now.

Shanghai-based Agibot, the world's largest humanoid robot manufacturer by volume, crossed 10,000 cumulative units shipped earlier this year, Forbes reported in May. The company shipped roughly 5,100 units in 2025 alone, capturing a 39 percent share of the global market. That is not a prototype figure. It is not a trade-show demo count or a pre-production trial number. It is robots, built, boxed, and delivered to customers who paid money for them. For a sector that spent the last three years oscillating between breathless keynotes and skeptical eye-rolling, 10,000 units is a number that demands the conversation change register.

The conversation is changing, but not necessarily in the direction the marketing videos suggest. In late June, Agibot livestreamed its G2 humanoid robots working a real production line at Longcheer, a Chinese tablet manufacturer. Over 64 hours, Forbes noted, the robots processed 17,625 tablets with 99 percent accuracy, performing inspection and assembly tasks while cameras rolled without cuts. The livestream was a deliberate counterpoint to the choreographed stage routines that have defined humanoid robotics PR: no dancing, no backflips, just machines doing traceable work on a real line. It was also an admission. If the technology were so obviously world-changing, nobody would need to prove it with a 64-hour broadcast.

The livestream gambit is spreading because the gap between the demo room and the deployment floor remains wide enough to drive a forklift through. Humanoid robot makers spent 2024 and early 2025 filling YouTube with clips of machines folding laundry, sorting colored blocks, and walking across uneven terrain in perfectly lit labs. Those videos built valuation. They did not build evidence. Now the market is asking a harder question: what happens when the robot meets a Tuesday morning? What happens when the lighting is bad, the floor is cluttered, the Wi-Fi is spotty, and nobody cleaned up for the camera crew?

Some of the answer arrived in April, when Stanford's AI Index reported that humanoid robots still fail 88 percent of household tasks. The finding was not about navigation or standing balance. It measured the kind of work that actual humans do in actual kitchens: loading a dishwasher with mixed dishes, folding a pile of laundry that includes a fitted sheet, wiping a counter without knocking over a half-full glass. These are tasks where the variance is infinite and the acceptable error rate is near zero. A robot that sorts tablets on a factory line with 99 percent accuracy is impressive. A robot that sorts a family's breakfast dishes with 99 percent accuracy breaks something every hundredth meal. That math does not work in a home.

The household deployment gap helps explain why the first wave of shipping humanoids has landed almost entirely on factory floors and warehouse aisles. Agility Robotics, the Oregon-based maker of the bipedal Digit robot, announced on June 24 that it had signed a definitive agreement to merge with Churchill Capital Corp XI in a SPAC deal valuing the company at $2.5 billion, Forbes reported. The company disclosed $300 million in pre-orders and noted that Digit had already logged more than 65,000 hours of work at real customer facilities. Those are warehouse hours, not living-room hours. Digit carries totes. It does not carry laundry baskets.

Agility's SPAC, the first U.S. public listing of a dedicated humanoid robotics company, is a milestone for the sector whether or not you think SPACs are a respectable way to reach public markets. The company expects to raise $620 million in proceeds, TechCrunch reported. That is real capital, the kind that funds production lines rather than proof-of-concept videos. It also opens Agility's books to the kind of quarterly scrutiny that separates shipping companies from science projects. The market will soon be able to track revenue per robot, gross margins, and whether $300 million in pre-orders converts into delivered units or lingers as a press-release number.

The Chinese duopoly is not waiting for the SPAC to clear. Agibot's 10,000 cumulative units and Unitree's cleared IPO, TechTimes reported in early June, have created a two-player dynamic that is pressuring every other competitor on speed and price. Agibot's president of embodied business, Dr. Yao Maoqing, told Forbes that the company believes the tipping point for humanoid robots is near, driven by falling component costs and the compounding effect of real-world deployment data. Every hour a G2 spends on a production line is an hour of training data that makes the next unit smarter. That feedback loop, if it holds, is a genuine moat.

The Home Market: 11,000 Pre-Orders and an 88 Percent Fail Rate

While the factory floor is where the volume lives, the consumer market is where the ambition is aimed. On July 1, Chinese robotics company UBTECH unveiled its U1 Series, a full-sized consumer humanoid robot designed for household use, and announced it had already secured 11,000 pre-orders ahead of first deliveries. The announcement followed a quieter milestone: UniX AI disclosed in April that its Panther service robot had begun operating inside private residences, not on a trade-show stage but in actual living rooms, performing tasks that the company described as household chores.

The leap from factory to home is not a simple matter of repackaging the same hardware in a softer shell. A factory robot operates in a controlled environment where lighting is consistent, the floor plan is static, the tasks are narrowly defined, and the people nearby have been trained to work around the machine. A home robot operates in an environment where children leave toys on the stairs, the lighting changes with the weather, the definition of "clean" is negotiated between cohabitants, and the people nearby have not signed a workplace safety waiver. The Stanford 88 percent figure captures this gulf in a single number, but the real challenge is not technical. It is experiential.

Consider what it means to be near a humanoid robot working in your kitchen. The machine is roughly your height, probably heavier, and it moves with a combination of precision and unpredictability that is hard to read. It does not make eye contact. It does not say "behind you" when it reaches past you for a plate. The industrial designers working on these machines are solving for form factors that feel approachable rather than threatening, but approachability is not the same thing as safety, and safety is not the same thing as comfort. A robot that permanently occupies a corner of the living room is a different proposition from a vacuum cleaner that docks itself when it is done.

Norwegian-American firm 1X Technologies is betting that the distinction matters. Its NEO humanoid robot, designed explicitly for home environments, uses soft actuators and fabric-covered joints to reduce injury risk during accidental contact. In May, 1X announced plans to produce 100,000 NEO units annually by late 2027, targeting a consumer price point of around $20,000. That is roughly the price of a compact car, or a high-end kitchen renovation. It is not an impulse purchase. It is a bet that a meaningful number of households will treat a general-purpose robot as a durable good, like a refrigerator or a washing machine, rather than a gadget.

The price-point comparison is useful because it reframes the question of who these robots are for. A $20,000 home robot is not a mass-market consumer electronics product in any normal sense. It is a premium appliance for households with disposable income, accessible floor plans, and the patience to debug a first-generation product. It leaves out renters, people in small apartments, anyone with stairs that the robot cannot navigate, and anyone for whom $20,000 represents more than a considered purchase. The accessibility gap is not an afterthought in this product category. It is built into the bill of materials from the start, and most of the companies shipping humanoids today are not yet talking about how they plan to close it.

What to Watch When the Cameras Are Off

The next twelve months will test whether the sector's momentum is built on substance or spectacle. Agility Robotics will begin reporting quarterly numbers as a public company. Agibot and Unitree will push further into factory deployments while their consumer divisions spin up. UBTECH's U1 pre-orders will either convert into satisfied customers or generate a wave of returns and YouTube teardowns. 1X will either ramp toward its 100,000-unit target or quietly revise the timeline. Each of these checkpoints is measurable, and each will generate data that either narrows or widens the gap between the demo and the reality.

The livestream era is a genuine improvement over the choreographed-demo era, but it is also a trap of its own design. A 64-hour broadcast of robots making tablets is compelling viewing for supply-chain executives. It is not the same thing as a longitudinal study of robot reliability over 64 weeks of continuous operation. The machines that Agibot showed on the Longcheer line were performing tasks they had been trained to do in an environment that had been prepared for them. That is how industrial automation has always worked. The question is whether the current generation of humanoids can handle variance gracefully when the preparation ends, when the task changes, when the person giving instructions is a tired shift supervisor who does not speak robot.

The industrial designers and ergonomics specialists who study this sector from the outside are asking a parallel set of questions. What is the injury rate when a humanoid robot and a human worker share a four-foot aisle? What is the noise level after eight hours of actuator whine? What happens to the social dynamics of a shift when a silent, tireless coworker joins the line? These are not hypotheticals. They are the questions that determine whether humanoid robots become tools or grievances, and they will be answered not by the companies building the robots but by the people working alongside them, whose voices are almost entirely absent from the current marketing cycle.

One figure from the Agibot livestream is worth holding onto. The company reported that its G2 robots completed the Longcheer inspection line with 99 percent accuracy over 64 hours. That means one tablet in every hundred was flagged, misrouted, or mishandled. On a factory line with human inspectors, that error rate would trigger a process review. The fact that a 99 percent accuracy rate is being presented as a triumph tells you where the technology actually stands. It is good enough for a demo that doubles as a proof of concept. It is not yet good enough for a factory manager to go home and stop checking her phone. The next milestone is not a bigger livestream or a higher pre-order figure. It is the moment when a humanoid robot works an ordinary shift and nobody feels the need to broadcast it.