Chinese Embodied AI Startups Raised 13 Billion Dollars Across 322 Deals in Six Months, Five Times a Year Earlier

After years in which Chinese companies wowed the world with humanoid robots that run, dance and flip, the money is now chasing a different prize: the software that makes those machines actually useful. Embodied AI companies in China raised 93.5 billion yuan, about 13 billion dollars, across 322 deals in the first half of 2026, five times the amount a year earlier, according to research firm ITjuzi, in figures cited by Reuters.

The shift is deliberate. As competition and manufacturing scale push down hardware prices, robot makers have concluded that the margin will not come from the hardware but from the brain, as Dermot McGrath, founder of Shanghai consultancy ZenGen Labs, put it to Reuters. Funding is increasingly flowing to specialist developers of robot intelligence rather than to chassis and actuator manufacturers.

A first tier of specialists has already emerged. X Square, Galaxea and Spirit AI have each reached valuations of roughly 3 billion dollars, according to company disclosures. Established manufacturers are responding: Hong Kong-listed UBTech advertised a chief scientist position to lead AI model development with total compensation of up to 124 million yuan, about 18 million dollars, while Unitree, one of the world's largest humanoid robot makers by shipments, has assembled a team of about 30 researchers working on AI that predicts how physical objects behave, according to a person briefed on the matter.

The training bottleneck is data, and Chinese firms are attacking it with unusual intensity. Spirit AI has hired around 1,000 people to record everyday and industrial tasks for training. Ant Group's robotics unit Robbyant says it trained a model on 160,000 hours of material, including 50,000 hours of human-guided robot movements, across 20 robot configurations from 17 brands, aiming for one AI brain that works across different machines. DexForce takes a simulation-first approach, with chief operating officer Nicole Yang saying the company can prepare some industrial tasks for deployment within six to eight hours and reporting success rates above 99.9 percent in specified precision operations, a company-reported figure that has not been independently verified.

Deployment is following the data. Battery maker CATL says Spirit AI robots already participate in battery pack testing in its factories, inserting flexible cables with a success rate above 99 percent, and Robbyant's robots are handling orders in two Shanghai pharmacies. X Square has published research showing robots acquiring manipulation skills from a single human demonstration video without retraining.

The race has a geopolitical edge. American startups including Physical Intelligence, Skild AI and Generalist lead in some advanced robot AI systems, and Generalist says its GEN-1.5 model completes around 60 percent of simple tasks after viewing a demonstration lasting between 3 and 12 seconds. Chinese developers are betting that proximity to factories, logistics networks and industrial customers will let them close the gap faster than labs without physical-world access, even as US export controls squeeze their supply of advanced Nvidia chips and regulators scrutinize crowded IPO pipelines.

Unitree founder Wang Xingxing has argued the industry may be approaching its ChatGPT moment, defined as robots completing 80 percent of tasks in 80 percent of unfamiliar environments from verbal instructions alone. Whether that threshold arrives in two years or five, the 13 billion dollar first half makes one thing clear: in Chinese robotics, the contest has moved from the body to the brain, and the fundraising cycle is just getting started.