UBTech and MetaX's Robot Chip Venture Xixuan Raises RMB 500 Million at a RMB 4 Billion Valuation to Outbuild Nvidia's Thor

UBTech Robotics and GPU designer MetaX are doubling down on the deepest form of chip partnership available: ownership. Xixuan Chuangzhi, the Wuxi-based joint venture the two companies set up in May 2026 alongside manufacturing partner Fenglong, has completed a 500 million yuan angel round at a post-money valuation of 4 billion yuan — roughly $70 million invested at about $560 million valuation — according to announcements from UBTech and the investors on November 12.

GGV Capital and Matrix Partners China co-led the round. A long list of strategic and financial backers joined, including Helicap, Gem Dragon Capital's chip-focused vehicle, SAIC's Hengxu Capital, Touz Capital, Zhengxuan Capital, Heart Capital, Advantage Capital and Boya Ventures. The money is earmarked for core chip and compute platform R&D, building out the software toolchain, expanding the engineering team, and preparing joint validation and mass production with robot manufacturers and industry customers.

The joint venture's first product is a robot main-control SoC that its founders describe as a unified "big brain and cerebellum" processor — one piece of silicon handling multimodal perception, cognition and motion coordination together, instead of splitting those jobs across a cloud model and a separate real-time motor controller. The companies have set an unusually explicit benchmark: the stated research goal is to fully surpass Nvidia's Thor, the chip Nvidia positions for embodied intelligence. That is a target, not delivered silicon — the company says a development and validation platform will reach industry partners in 2027, with the first SoC and a complete embodied-intelligence solution expected in 2028.

The structure matters as much as the number. In the robotics industry, chipmakers and robot makers usually relate through supply agreements or reference designs, and in both cases the "definition rights" for the chip stay with the silicon vendor. The JV exists precisely because neither side could get what it wanted through supply: MetaX needs real robot workloads to define a chip natively for embodied intelligence rather than adapting an autonomous-driving processor, and UBTech wants its understanding of multimodal sensing and motion coordination built into the architecture from the first schematic, not patched in after tape-out. Chairman Chen Weiliang, MetaX's founder, framed the design philosophy as cloud-edge coordination: pairing general-purpose GPUs and a full software stack with robot scenarios so cloud training, simulation and on-device inference evolve together.

UBTech, one of China's largest publicly listed humanoid robot makers, says the chip definition work is based on a survey of embodied AI application scenarios worldwide, and that the first SoC is being designed for semi-structured environments — warehouses, factories and service settings that are constrained but not fully engineered. The near-term focus is a general edge computing platform for perception, understanding and autonomous decision-making in robots.

The round lands in a market that has been pouring capital into exactly this layer. Chinese embodied AI startups raised about 93.5 billion yuan, roughly $13 billion, across 322 deals in the first half of 2026 — five times the year-earlier total — as investors shifted attention from robot bodies to robot brains. Morgan Stanley research published this month found 64 percent of Chinese firms would adopt humanoids within three years. A dedicated robot SoC venture with a four-billion-yuan valuation eight months after incorporation fits the pattern: the bottleneck investors see is not walking machines but the compute that makes them useful.

The bet is also a hedge against Nvidia's grip on the category. Thor has become the default reference point for humanoid computing, and Chinese robot makers — like their autonomous-driving counterparts before them — want a domestic alternative they can shape. Whether a 2028 chip can genuinely outbuild Nvidia's roadmap is unknowable from an angel round. What the round does establish is that China's humanoid and GPU industries are now tightly coupled enough to fund their own silicon, and that the competition for the robot brain is becoming a competition over who defines the chip.