After years of building humanoid robots that can run, dance and do backflips, Chinese companies are now pouring money into a different problem: the software that would make those machines actually useful. Reuters reports that Chinese embodied AI companies raised 93.5 billion yuan, about $13 billion, across 322 deals in the first half of 2026, five times the amount raised a year earlier, according to research firm ITjuzi. The hardware works. The brains do not, at least not yet. The companies that figure out the brains first could own the next big market in tech.
Embodied AI is the name for software that lets a robot understand its surroundings, learn unfamiliar tasks and adapt without being programmed for every single job. Today's humanoid robots are impressive in a demo and mostly useless in a warehouse, because each new task has to be coded in by hand. The goal is a general-purpose robot brain that can watch a person do something once and then do it itself. Generalist, an American startup, says its GEN-1.5 model completed around 60 percent of simple tasks after watching a demonstration lasting between 3 and 12 seconds. China's X Square has published research showing robots picking up handling skills from a single human video without being retrained for each task. These are early results, but they show where the field is heading.
The money is flowing because the hardware business is getting squeezed. As competition drives down the price of robot bodies, the profit is moving to the software. Dermot McGrath, founder of Shanghai consultancy ZenGen Labs, told Reuters that robot makers have concluded the margin will not come from the hardware but from the brain. Specialist Chinese startups including X Square, Galaxea and Spirit AI have each reached valuations of roughly $3 billion, according to company disclosures. Hong Kong-listed UBTech advertised a chief scientist job overseeing AI model development with pay of up to 124 million yuan, about $18 million, including equity and benefits. Hangzhou-based Unitree, a leading humanoid maker, has about 30 researchers working on AI that predicts how objects behave.
The race to train these brains is taking some strange forms. Unlike chatbots, which learn from text on the internet, robots need examples of physical interactions. Spirit AI has hired around 1,000 people to record everyday and industrial activities. Ant Group's robotics unit Robbyant says it trained a model using 160,000 hours of material, including 50,000 hours of human-guided robot movements, across 20 robot configurations from 17 brands. DexForce takes a different route, using computer-generated environments, and says it can prepare some industrial tasks for deployment within 6 to 8 hours. Simulations have limits, though, since they cannot perfectly reproduce how things feel and move in the real world.
The American side of the race is not standing still. Startups including Physical Intelligence, Skild AI and Generalist are building software meant to teach robots different skills without extensive retraining. Skild AI raised $1.4 billion in January 2026 in a round led by SoftBank, valuing the company at more than $14 billion. The American labs lead in advanced robot AI and have better access to powerful computing chips, while the Chinese companies can work closely with robot makers and industrial customers to test and improve their systems. The country that best combines AI with machines could gain an edge in automating factories and logistics, and that is why this is becoming a new front in the US-China tech rivalry.
There are real deployments already, just not the ones in the demo videos. Battery maker CATL says Spirit AI robots connect flexible wires during battery-pack testing at its Zhongzhou plant, with insertion success above 99 percent. Robbyant says its robots handle orders in two Shanghai pharmacies. But software that can learn new jobs without extensive adaptation is still some way off. Unitree founder Wang Xingxing said last month the company had tested OpenAI's GPT-6 on robots with encouraging results, and he said GPT-7, if released next year, could bring robotics its own ChatGPT moment. His test is robots completing 80 percent of tasks in 80 percent of unfamiliar environments from spoken instructions. Consultant Georg Stieler offered the sober version: a demo breakthrough would not mean robots were ready for broad commercial use, and if the question is when a customer can count on it, add years, not months.
Chinese robot firms also face headwinds, including shortages of advanced Nvidia chips because of US export controls and a squeeze on IPO funding. The robots learned to move. Now $13 billion is riding on teaching them to think. The company that cracks that problem will not just sell robots. It will sell the brains inside every robot that follows, and that is a much bigger business than the bodies.
---------------
Author: Seta Tsuruki, Asia Newsroom