Two significant developments have recently emerged from Shanghai's prominent GPU company, Metax Integrated Circuits (Shanghai) Co., Ltd. (688802).
On June 12, the company announced its intention to issue H-shares and seek a listing on the Main Board of The Stock Exchange of Hong Kong. The day before, it entered into a strategic cooperation agreement with the humanoid robotics leader UBTECH ROBOTICS (09880). The partnership will establish a joint venture, Xixuan Chuangzhi Technology (Wuxi) Co., Ltd., focused on the research, development, and mass production of embodied AI edge-side chips and complete solutions.
Key Strategic Moves
This collaboration is seen as a powerful union between a domestic computing power leader and a robotics leader. For Metax, which went public on the STAR Market just half a year ago, the pursuit of a Hong Kong listing requires a high-growth narrative like "robotics chips" to support its valuation. After all, NVIDIA (NVDA), often seen as the benchmark in the global AI industry, has repeatedly stated that humanoid robots represent the next wave of AI.
Metax stated that the Hong Kong listing aims to meet the needs of its business development, further enhance corporate governance and core competitiveness, and advance its global strategy. Regarding the joint venture, a company representative said the decision to enter the embodied AI edge-side chip sector is based on a comprehensive assessment of market trends, industrial policy, and internal capabilities.
"Embodied AI is the crucial leap for artificial intelligence from the digital world to the physical world, and chips are the 'brain' of this transition. The establishment of Xixuan Chuangzhi with UBTECH is a deep integration of 'computing power' and 'body'," the representative emphasized.
Navigating a Crowded Field
An AI chief analyst at industry research firm Omdia noted that current high-performance edge-side chips primarily handle large model inference, supporting models for machine vision, recommendation systems, and world models, while also balancing power consumption. GPU-like chip capabilities are well-suited for robotic applications.
A domestic humanoid robotics company indicated they currently use NVIDIA's Jetson Thor and Orin chips for embodied AI edge-side computing. They stated that if domestic chips offer comparable performance, they would consider adopting them in the future.
Beyond Simple GPU Integration
The joint venture, Xixuan Chuangzhi, has a registered capital of 100 million yuan. Metax and UBTECH each hold a 35.01% stake, making them the largest shareholders. The management structure features Metax's founder as chairman and UBTECH's founder as vice chairman, with a professional CEO from H3C overseeing operations, indicating a deep, integrated partnership rather than a simple supplier-customer relationship.
According to the cooperation plan, the joint venture will leverage Metax's heterogeneous computing and intelligent inference chip technology alongside UBTECH's core embodied AI technology and scenario advantages to develop dedicated edge-side chips. The goal is to achieve tape-out in the second half of 2027 and mass production in 2028.
Current embodied intelligent systems, represented by humanoid robots, commonly use a dual-system "big brain, small brain" computing architecture. The "big brain" for AI computing almost entirely relies on overseas general-purpose chips, with NVIDIA's Jetson platform being the de facto standard for many robotics companies.
The analyst pointed out that NVIDIA's deep cultivation in the robotics market has created high dependency among developers, particularly on its CUDA ecosystem and large models. Currently, domestic AI chip companies performing relatively well in robotics include Horizon Robotics and Black Sesame, as autonomous driving chip design principles naturally extend to robotics.
Metax argues that embodied AI edge-side chips are not simply about installing a GPU into a robot; dedicated robotic chips have their own barriers and thresholds.
"GPUs, as a type of general-purpose computing chip, currently mainly handle heavyweight AI tasks like training and inference for foundational models. Dedicated edge-side AI chips perform computations locally on the device or at the edge, handling immediate, lightweight AI tasks but requiring high energy efficiency, low latency, data privacy, and offline operation capabilities," the company explained.
Specifically, differences exist in computing objectives and interface/perception requirements. Embodied AI edge-side chips focus on inference and real-time control, pursuing "sufficient computing power with extremely low power consumption," and need to support low-latency drive interfaces and multi-sensor interfaces to meet real-time control and sensor fusion demands.
It is worth noting that a 100 million yuan registered capital is a modest sum for chip R&D, implying significant future capital pressure. This appears to be a long-term play that will require time to materialize.
Funding Allocation and Financial Context
Other domestic GPU companies have also launched edge-side AI chips for various applications, including AI PCs. Focusing edge-side AI chips specifically on embodied AI, however, appears unique to Metax, achieved through its deep partnership with UBTECH.
Metax has established four GPU product lines covering training/inference, AI inference, graphics rendering, and scientific intelligence. The company stated that combining its high-performance GPU expertise with UBTECH's robotics scenario expertise will build the computing foundation for embodied AI, aiming to give every robot an autonomous, controllable "central system."
The establishment of the joint venture and the initiation of the Hong Kong listing plan reflect a "dual-line financing" strategy, highlighting the capital-intensive nature of AI chip companies and signaling optimism about the embodied AI sector.
According to the announcement, the number of H-shares to be issued will not exceed 5% of the company's total share capital post-issuance (before any over-allotment option exercise), with an over-allotment option of up to 15% of the initial offering size granted to coordinators.
Proceeds from the offering, after deducting related expenses, are intended for next-generation GPU product R&D and commercialization, software ecosystem development, industry chain investment and M&A, marketing and sales system construction, supplementing working capital, and other general corporate purposes.
Metax listed on the STAR Market in December 2025, raising approximately 4.197 billion yuan. Its 2025 annual report showed revenue of 1.644 billion yuan, a 121.26% year-on-year increase. Revenue for Q1 2026 reached 562 million yuan, up 75.37% year-on-year.
Due to high R&D investment in GPUs, the company remains loss-making, with a net loss of 789 million yuan in 2025 and 98.47 million yuan in Q1 2026. Consequently, Metax requires more "financial ammunition" to support its development.
The analyst noted that Hong Kong's ample capital and relatively fast approval process attract AI startups to list there. With the addition of chip and large model companies, Hong Kong is gradually forming a complete AI industry chain from underlying computing power and core models to the application layer, potentially serving as a springboard for Chinese AI companies going global.
Comments