China Shipped 77.9% of the World's Humanoid Robots in H1, IDC Figures Say

China shipped more than 19,000 humanoid robots in the first half of 2026, or 77.9% of global shipments, according to figures attributed to IDC and reported by TechNode on 29 September. Global shipments reached nearly 25,000 units, up 432.1% year over year. IDC also raised its forecast for global humanoid robot shipments in 2030 to more than 750,000 units, TechNode said, citing IT Home.

That is the shape of the boom: one country, one half-year, and a forecast that multiplies the market roughly thirtyfold in four years. It is also the number that any claim about a humanoid robot factory now has to sit next to.

The counterweight is what is actually installed in factories today. The Robot Report's 29 September manufacturing survey cites the International Federation of Robotics: industrial robot installations doubled in the decade that ended in 2024, and robot density, the number of robots per 10,000 workers, keeps climbing, with South Korea, Singapore, Germany and Japan leading. Humanoids are a rounding error in that stock. The Robot Report's own assessment is blunt. Interest in humanoids is high, but they are just beginning commercial trials, and other form factors are already available.

Inside the factory, the work is still sensing and training

The gap between a shipment number and a working robot shows up in the unglamorous layer: force and torque sensing. ForceN makes force torque sensors for robotic wrists and other joints, and has supplied sensing systems to surgical, logistics and humanoid robots. The company will present a crash course on the subject at RoboBusiness 2026, The Robot Report reported on 29 September. Founder and CEO Robert Brooks will speak at 2:15 p.m. PT on 20 October in Santa Clara, California. The session teaches attendees how to use force and torque sensors in humanoids and how to avoid breaking them, drawing on three years of work with what the company describes as many of the world's top humanoid companies.

Take the framing in that article literally. Without force torque sensors, robots are blind to how much pressure they are applying while working. A humanoid that cannot feel is a humanoid that crushes a part, drops a tool or stalls on a task a two-fingered gripper was never designed to do.

Training is the other half of the problem, and here the measurements are messier. A technical report published by fig.inc on 29 September evaluated frontier models such as Astra and Opus 5.5 across web automation and physical tasks including manipulation, assembly, industrial procedures and driving. Its central finding is what the authors call jagged performance. There is no clear performance leader: in every model pair tested, the lower-scoring model solved tasks the higher-scoring one failed. The same unevenness appeared in the physical domains, where models with higher averages on every domain still did worse on some task categories. The authors released RIDGE, a dataset of item-level results and model outputs across five domains.

That matters for anyone promising a general-purpose factory humanoid. If competence swings that widely from task to task, a single benchmark number tells a plant manager almost nothing about whether the robot will handle the specific job on the line.

Five million robots, thousands of them humanoid

The installed base gives the hype its scale check. Recent headlines, which FLASH24 is not treating as verified facts here, put the world's factory robot population at more than five million, with only thousands of those being humanoids. The Robot Report's survey adds a demand-side number. In the first half of 2026, robot order value rose 6.6% year over year, according to the Association for Advancing Automation, which noted that increasing orders in non-automotive sectors have offset slowdowns among automotive OEMs.

Order value growing at 6.6% while humanoid shipments grow at 432.1% is not a contradiction. It is the difference between a market and a pilot.

Which is why the most interesting humanoid shipping this week is not going to a factory at all. Flourish, a startup split between San Francisco and Paris, is opening sales for Flourish One, a $3,555 wheeled humanoid-style home robot that early adopters can teach new chores in under 30 minutes using only a smartphone, The Robot Report reported on 29 September. The first batch is 50 units, hand-built, and the company says it hopes to ship before Christmas.

"I've automated everything in my computer. ... But when I came home, the most painful stuff in my life was still there," founder Antoine Marcel told The Robot Report.

Marcel's argument for the home over the factory is that every household is idiosyncratic, so Flourish fine-tunes per task and per home rather than waiting for a generalist model. The robot has two arms with a 1.5 kg (4 lb.) payload each, a battery designed for 12 hours of arm operation, a small lidar near the floor on the base, and wrist-mounted cameras on two-fingered grippers. A Raspberry Pi runs the robot, while training and higher-level inference run on cloud GPUs, which the company says keeps the bill of materials down.

The limits are stated plainly. The robot is not waterproof and will not do dishes or wash the dog in its first generation. It has limited dexterity: tidying, trash and basic cleaning such as wiping a table, but not intricate assembly. Teaching uses the phone's inertial measurement units, with the user moving the phone as if performing the task while the robot mirrors the action.

Put the two pictures side by side and the factory question answers itself for now. China's 19,000-plus units and IDC's 750,000-unit forecast for 2030 describe a manufacturing base scaling fast. The Robot Report's survey describes factories where humanoids are still in trials and where integration and return on investment remain hard. ForceN's session is about not breaking a sensor. fig.inc's report is about models that fail unpredictably. Flourish is about a 50-unit batch that cannot get wet.

None of that makes the IDC number wrong. It makes it a shipment figure, not an installation figure, and the difference is where the next two years of humanoid claims will be won or lost.