A Chinese joint venture set up to design processors for humanoid robots has raised 500 million yuan, about $75 million, in one of the semiconductor industry’s largest-ever angel rounds, as domestic chipmakers try to build a Chinese alternative to Nvidia’s dominance in robot brains.
The company, formally registered as Xixuan Chuangzhi Technology in the eastern city of Wuxi under the brand Xilect, pairs two partners with complementary assets. MetaX is one of China’s leading designers of graphics processors for data centers, while UBTech Robotics is among the country’s best-known humanoid robot makers, selling 1,079 full-size robots in 2025 for 820 million yuan in revenue and deploying them in car factories run by companies such as BYD and Geely. Each parent holds a 35.01 percent stake.
The funding round was co-led by GGV Capital and Matrix Partners China, two of China’s most established technology investors. The money will pay for core chip and computing-platform research, the software that lets robot makers program the silicon, a larger engineering team across research centers in Shanghai, Shenzhen and Beijing, and joint testing with robot manufacturers.
The benchmark the venture is aiming at is no secret: Nvidia’s Thor, the chip the American company has designed specifically for autonomous machines and advanced driving. Reports from the semiconductor supply chain say Xilect is targeting Thor-class performance, with its first chip scheduled for tape-out — the point at which the design is sent to a fabrication plant — in the second half of 2027 and mass production in 2028. The engineering problem is different from the data-center chip race that has defined U.S.-China technology rivalry. A humanoid robot runs on batteries, and after accounting for motors, sensors and cooling, the onboard computer often gets a power budget of just a dozen to a few dozen watts. A flagship data-center accelerator such as an Nvidia H100 draws roughly 700 watts. Simply shrinking a desktop graphics chip, analysts caution, cannot produce a processor that perceives its surroundings, plans motion and controls dozens of joints within the energy budget of a walking machine.
That is the gap Xilect says it will close with an edge-computing architecture designed from scratch. The structure gives the startup both silicon expertise and a ready test bed: UBTech’s factory deployments generate the real-world manipulation and navigation data needed to train robots, and offer an immediate customer for whatever the chip becomes. The round is the clearest sign yet of how China’s robot and semiconductor industries are converging. Venture money is flooding into embodied AI — software that lets machines understand and act in the physical world — while Washington’s export controls have pushed Chinese firms to build domestic computing at every layer, from training chips in server rooms to processors inside moving machines. Chinese rivals including Huawei’s Ascend line and startups backed by local governments are pursuing the same ground.
Plenty can go wrong. Tape-out is nearly two years away, and by 2028 Nvidia and other competitors will have shipped at least one newer generation, while robot makers remain few and cost-conscious. MetaX reported a third-quarter profit this year on strong demand for domestic AI chips, which helps fund the bet. Whether a 20-watt Chinese processor can match Thor-class intelligence will do much to decide whether the country’s humanoid robots carry foreign or domestic brains.

Dgcity