IFA Filled Berlin With Humanoids That Cannot Work Yet

IFA Berlin 2026 hosted the largest humanoid robot exhibition in the show's history. Robots danced, performed tai chi, did backflips and the splits, and walked a literal runway in the show's Robots on the Runway program. Chinese exhibitors participated at their largest scale ever.

Also true, from the same show floor: over 70% of humanoid revenue is concentrated in research and education, and more than 60% of shipments go to performance and data-generation purposes.

Those two facts describe the same industry, and holding them together is the only honest way to read this week.

Mobility is solved. Work is not.

The backflips are not fake and they are not trivial. Dynamic balance, whole-body control, and recovery from perturbation were genuinely hard problems five years ago, and the Chinese manufacturers filling IFA's halls have solved them at a price point that would have been implausible in 2023.

What the demos do not show is autonomy. As observers at the show put it, the demonstrations displayed mobility and presentation — not verified autonomy, endurance, or safety for real home and workplace tasks. The "brains" needed to replace human labour in an industrial setting or a house remain underdeveloped.

That gap has a specific technical shape. A choreographed routine is an open-loop trajectory executed in a known environment with a flat floor and no obstacles. Folding laundry in an unfamiliar room is a closed-loop manipulation problem with deformable objects, occlusion, and no reset button. The first is a controls achievement. The second requires a perception-and-manipulation stack that nobody has shipped at consumer reliability.

Hardware is ahead of software, and the demo format is optimised to show hardware.

What the revenue mix actually says

The 70%-research-and-education figure is the number to carry out of IFA.

It means the humanoid industry's current customers are buying platforms, not workers. A university lab buying a bipedal platform to run manipulation research is a real, healthy customer — but it is a capital-equipment purchase with a small addressable market, not the labour-replacement market the valuations imply.

The 60%-performance-and-data-generation figure is more interesting, because part of it is strategic rather than decorative. Teleoperated and demonstration deployments generate training data, and manipulation data is the binding constraint on the whole field. There is no ImageNet for bimanual manipulation; every serious player is buying data by putting robots in front of tasks and recording.

So a meaningful share of "robots doing spectacle" is actually robots doing data collection with an audience attached. That is a defensible strategy. It is also, financially, R&D expenditure dressed as revenue.

PrimeBOT and the consumer bet

PrimeBOT showed the T1 and Q1 at IFA, with a ShowStoppers preview on September 3. The T1 won an official IFA Innovation Award for content-creation technology; the Q1 was an IFA NEXT honoree.

The Q1 is the more instructive product: an 88 cm platform with 22 degrees of freedom, supporting 10-plus martial arts movements, positioned for deep customisation, programmability, multi-modal interaction, and — explicitly — safer close-range interaction. It goes on sale in China later this month at a price not yet announced.

Read the spec sheet for what it concedes. Eighty-eight centimetres is deliberately not human scale. Twenty-two DOF is a fraction of what dexterous manipulation requires. The stated markets are academic research labs and interactive home companionship.

This is an honest product. PrimeBOT is not claiming the Q1 will do your dishes; it is selling a research platform and a companion device, and it is sized so that a fall does not injure anyone. Compare that against the market's dominant narrative — humanoids replacing warehouse labour within two years — and the gap between what is being sold and what is being promised becomes visible.

The competitive reality

Chinese firms dominated the floor, which matches the market: China holds roughly 85% of global humanoid share. That dominance is manufacturing-driven — actuators, harmonic drives, batteries, and assembly at a cost structure Western entrants cannot match — and it is durable in exactly the way hardware advantages usually are.

It is also now running into policy. The FCC moved in July to block new imports of foreign-made humanoid and quadruped robots on national-security grounds, following a White House interagency determination that foreign-built robots could pose cybersecurity risk to critical infrastructure. The restriction applies to new device models; previously purchased and already-authorised models are unaffected, and federal government use is exempt.

For a Chinese manufacturer showing a consumer humanoid at IFA, that removes the US consumer market from the plan. Europe becomes correspondingly more important — which is a reasonable explanation for why IFA, a Berlin consumer-electronics show, hosted the largest humanoid exhibition in its history.

The read

The correct posture on humanoids right now is neither dismissal nor extrapolation.

The hardware progress is real, fast, and cheap, and it is concentrated in China for structural reasons that will not reverse quickly. The autonomy required to convert that hardware into labour is not close, and the revenue mix proves it: this is a platform-and-data business wearing a labour-replacement costume.

The signal that would change this assessment is not a better backflip. It is a manufacturer publishing sustained task-completion rates on unstructured work — hours of autonomous operation, intervention frequency, failure modes — instead of a runway show.

Until someone publishes that, IFA 2026 is the most impressive demonstration to date of a capability that does not yet have a job.