For half a century the humanoid robot was a creature of science fiction and trade-show theatre, the thing that always arrived with cables visible just out of frame. The quiet news of 2026 is that it has become a product. Manufacturers are no longer releasing choreographed stage demos; they are releasing footage of identical units rolling off assembly lines in batches of hundreds, and, more tellingly, they are signing purchase orders.
The signals pile up. Unitree, fresh from a Shanghai listing where its shares jumped more than 450 percent on day one, has launched the GD01 as a production unit rather than a research kit. UBTech has confirmed orders for hundreds of robots bound for Geely, BYD, and Foxconn lines. Agility Robotics' Digit now works in GXO Logistics warehouses under a robotics-as-a-service contract, a financial structure that signals commercial maturity more convincingly than any backflip. At a Shanghai industry fair this month, China's JAKA showed two wheeled humanoids cooperating to assemble an engine, one carrying it into place while the other tightens the screws.
The capital has noticed. Startups in the "humanoid" category pulled in roughly 11 billion dollars in venture funding in the first nine months of this year, more than double the 4.8 billion recorded for all of 2025. What changed is less the hardware than the software wrapped around it. As one researcher told the trade press, building a stable walking robot used to be a doctoral thesis; now it is a bill of materials. The contest has moved upstream, to the vision-language-action models that let a supervisor redirect a machine in plain speech rather than reprogram it for every new task.
This is where the story turns geopolitical, and where a sharp argument in Foreign Affairs deserves attention. While Silicon Valley pours its effort into scaling ever-larger language models, Beijing is fusing its software and hardware strengths into what it calls embodied AI, and has named it an industry of the future alongside 6G and quantum. The bet is simple and unsettling: if the most consequential advances in AI turn out to hinge on mastery of the physical world, then America's lead in frontier models matters far less than the capacity to build robots, deploy them, and harvest the real-world data they generate.
On that scorecard China holds awkward advantages. It already makes around 80 percent of the world's LiDAR sensors and controls large stretches of the components supply chain. Its vast manufacturing base doubles as a training ground, feeding operational data back into the models. Washington has started to react, banning some advanced foreign robots in July to carve out space for domestic firms, but American companies remain fewer, costlier, and increasingly reliant on Chinese parts.
None of this is settled. Chinese regulators are tightening the path to public listings because the technology is not yet commercially viable at scale, and the headline market forecasts, from Goldman's cautious 38 billion dollars by 2035 to Morgan Stanley's giddier trillions, span an almost comic range. The honest reading is that the demos are over and the deployments have begun, which is exactly when the hard questions start. The open one is whether the country that writes the smartest models also ends up being the country that builds the bodies to put them in.