A Chinese joint venture backed by GPU designer MetaX and humanoid-robot maker UBTech Robotics has closed a 500 million yuan (roughly $75 million) angel round with a single, explicit target: building a domestic computing platform that benchmarks against Nvidia's Thor1,2. The move is less a fully funded chip program than a down payment and a declaration of intent: the round's existence, and the specificity of the Thor benchmark, signals that investors and strategic partners judge the geopolitical constraints tight enough, and the robotics-silicon market promising enough, to begin placing capital against Nvidia's roadmap even while that roadmap is itself still pre-production.
The venture, Xilect (formally Xixuan Chuangzhi Technology), was registered in Wuxi with 100 million yuan of registered capital. MetaX and UBTech each hold 35.01 percent of the company. GGV Capital and Matrix Partners China co-led the round, according to a report by 21st Century Business Herald carried by IThome. Xilect said the funds will go to core chip and computing-platform R&D, its software toolchain, a larger R&D team, and joint validation with robot makers and industry customers. Other investors and the valuation were not disclosed.
The Thor comparison is the sharpest detail in the filing. DigiTimes reported that Xilect benchmarks its domestic computing platform against Nvidia's Thor. When the venture was announced in June, the partners said only that the product would benchmark against top international brands. The shift from vague aspiration to a named competitor could reflect genuine progress on a technical specification, or it could reflect investor-narrative management, sharpening the pitch as the angel round closed. No spec sheet, process node, or architecture detail has been disclosed. DigiTimes reported the Thor benchmark, and the June announcement referenced only generic international brands; the gap between invoking a benchmark and matching it remains enormous.
Xilect's first embodied-AI chip is slated to tape out in the second half of 2027, with mass production in 2028. MetaX chairman Chen Weiliang chairs the venture; UBTech founder and CEO Zhou Jian is vice chairman; and former New H3C senior vice president Yang Xianbo serves as chief executive. The company has R&D centers in Shanghai, Shenzhen, and Beijing.
MetaX and UBTech each hold 35.01 percent of Xilect. That means neither parent holds outright control. The equal-stake structure gives Xilect room to bring in additional strategic partners without either founder being diluted below a blocking position. It also means the venture's technical direction will be shaped by negotiation between a chip company and a robot company, not dictated by one side.
MetaX brings GPU design experience and a freshly public balance sheet: the company completed an IPO on July 24, 2026. But MetaX's product history is in discrete GPUs for graphics and general-purpose compute, not in the integrated, real-time control domain that an embodied-AI SoC demands: sensor fusion, low-latency control loops, and functional-safety certification are distinct disciplines from training-cluster GPU design. How much of MetaX's existing IP transfers to Xilect's problem set is an open question the company has not addressed.
UBTech brings a customer base: the company reported 2025 revenue of 820 million yuan and sold 1,079 full-size humanoid robots in 2025. A robot maker that simultaneously holds 35.01 percent of its own chip supplier, as the sourced ownership data shows, has an unusual incentive alignment: Xilect's first customer is, in effect, a co-parent. That captive demand could accelerate validation cycles. But the arrangement also raises the question of whether the chip will be designed too narrowly around UBTech's own architecture to attract the broader ecosystem it will need for unit-cost scale.
The timing intersects with turbulence at Nvidia itself. Jensen Huang was ousted as CEO on November 12, with CFO Colette Kress promoted to interim chief executive the same day and Harvey C. Jones elevated to board chair a day earlier. Nvidia's leadership transition does not, on available evidence, change the Thor roadmap. No sourced reporting connects the C-suite change to any shift in Thor's timeline or priority. The transition is worth noting as context: Xilect is raising capital against a roadmap whose organizational sponsor just changed. But it would be overreach to treat an unrelated executive departure as a competitive variable for a Chinese startup's chip bet.
Vector Wire has also tracked the broader cost pressures facing Chinese AI chip buyers. In September, we reported that HBM costs were driving 20 to 60 percent price hikes across Chinese AI accelerators. Those price increases could strengthen the pitch for a domestically designed robotics chip optimized around different memory or packaging choices, but Xilect has disclosed neither its memory strategy nor its packaging approach, so the competitive advantage remains hypothetical until the company reveals architectural details.
Xilect is starting from a tape-out target in late 2027 and mass production in 2028. The challenge for any fast-follower in silicon is that the target is not stationary: matching Thor's published specs at tape-out does not mean matching whatever Nvidia ships in the same window. Xilect's real competitive lever may not be raw performance but supply-chain sovereignty, the ability to guarantee Chinese robot makers access to advanced compute regardless of export-control shifts. Vector Wire reported in November that the fallout from Nvidia chip-export restrictions was widening, and every tightening of controls makes the domestic-alternative pitch more compelling to Chinese OEMs even if the domestic chip trails on paper. But "available" requires fabricable at scale on an accessible node, and Xilect has not disclosed a foundry partner or process node. Without that disclosure, the sovereignty pitch remains a thesis rather than a plan.
ANALYSIS What to watch: whether Xilect discloses a process node or foundry partner before the tape-out window in the second half of 2027, which would signal how close the design is to fabrication. Whether additional robot OEMs beyond UBTech sign validation agreements, testing whether the chip can attract demand outside its captive parent. And whether Nvidia's new leadership signals any change to the Thor timeline or positioning, which would reset the benchmark Xilect has chosen to chase.