UBTech and MetaX Form Joint Venture to Build Dedicated Physical-AI Chips, Citing 33% BOM Cost Pressure优必选与摩尔线程成立合资公司，专攻物理AI芯片，直指机器人BOM成本痛点
Reported on Wednesday, UBTech — the Hong Kong-listed humanoid maker — has established a joint venture with MetaX, a Chinese GPU design firm, to develop and mass-produce chips dedicated to physical AI. UBTech stated that overseas chips currently account for up to one-third of a robot's bill of materials, framing domestic chip development as a structural cost problem, not merely a sanctions-response.
The move is part of a broader wave: Tars-Z-Hang, a startup founded by former Huawei and Baidu autonomous-driving engineers, plans to equip its next-generation industrial robot 'A3' with a 560 TOPS Chinese-made edge AI chip that runs inference on-device rather than in the cloud. SenseTime separately announced its 'Galaxy Plan' — a domestic-chip compute cluster co-developed with Huawei, Cambricon, and MetaX.
The pattern across these moves is consistent: Chinese robotics and AI companies are pulling chip localization upstream into their own product stacks, treating it as a competitive and cost lever rather than waiting for a domestic chip ecosystem to mature independently.
MetaX, the Shanghai-based GPU designer, has confidentially filed for a Hong Kong listing, targeting an IPO by the end of 2026, according to sources cited by the South China Morning Post on Friday. The company is working with Huatai International Financial Holdings on the share sale.
The filing follows MetaX's STAR Market debut in December 2025, when it raised roughly 4.2 billion yuan and its stock surged close to 700% on the first trading day. In June 2026, MetaX shareholders approved the issuance of H-shares — capped at 5% of enlarged capital — with proceeds earmarked for next-generation GPU development, software ecosystem build-out, supply-chain investments, and potential acquisitions.
The move lands as Beijing intensifies pressure on domestic AI firms to replace Nvidia silicon. MetaX is one of a cluster of Chinese GPU designers — alongside Biren and Moore Threads — racing to fund the compute stack that would make domestic substitution credible at scale.