Gasgoo Munich- On November 12, Xixuan Chuangzhi Technology—a chip venture launched by UBTECH and MetaX—announced it has closed a 500 million yuan angel round, giving it a post-money valuation of 4 billion yuan. The company has set an ambitious target for its first self-developed chip: to "fully surpass NVIDIA Thor."
The rhetoric is bold, but the product remains on paper. Under the current roadmap, Xixuan Chuangzhi Technology's chip won't officially launch until 2028.
Why is UBTECH entering the chipmaking fray now, and what makes it believe it can take on NVIDIA?
Image Source: UBTECH
Computing Power Held by Others, OEMs Constrained by Generic Chips
To understand this push for proprietary chips, we first need to look at what UBTECH has relied on in the past.
Based on public specifications for the Walker X and TianGong Walker, UBTECH has long relied on external solutions from Intel and NVIDIA for the high-level computing platforms in some of its robots. The Walker X is equipped with two Intel i7-8665U processors and an NVIDIA GT1030 graphics card. Meanwhile, the TianGong Walker, designed for research and education, uses NVIDIA's Orin AGX 64 as its development computing unit in certain versions—delivering up to 275 TOPS of compute per chip, with configurations supporting either one or two units.
It is clear that from early Intel processors to later NVIDIA platforms, UBTECH's high-end AI computing infrastructure has depended heavily on external suppliers.
Because these off-the-shelf platforms aren't defined around the specific tasks of a single robot maker, UBTECH has had to navigate trade-offs between model deployment, sensor interfaces, power consumption, and system cost within the constraints of generic hardware.
This inevitably drives up deployment costs and adaptation complexity. A practical challenge is that handling more complex perception and decision-making tasks often requires larger models. Yet with limited space, battery life, and thermal management in a robot chassis, on-device inference hits a "power wall," while cloud-based inference runs into a "latency wall."
Xie Yipeng, head of industrial solutions at Unitree, has previously noted that power consumption constrains the large-scale deployment of computing power in embodied robots. Limited by onboard batteries and cooling, high-compute chips must first overcome power and thermal hurdles to sustain high-load operations on a robot. Consequently, automotive high-performance solutions cannot be simply transplanted to robotics.
An even more pressing challenge is the deep entanglement of supply chains and ecosystems.
Reliance on external suppliers for advanced AI computing platforms means a robot maker's product cadence, costs, and software adaptation are vulnerable to shifts in the partner's product roadmap and ecosystem.
In other words, buying chips solves the question of "can it work," but not "is it the right fit."
Partnering with a GPU Maker to Embed Real-World Tasks into Chip Design
Xixuan Chuangzhi Technology's solution is to bring chip experts and robotics specialists under one roof.
Established in May 2026, the company is a joint venture funded by UBTECH, MetaX, and Fenglong.
MetaX, founded in 2020 and listed on the STAR Market in 2025, specializes in full-stack, self-developed GPUs. With capabilities spanning chip architecture and instruction sets to software ecosystems, it fills the gap in UBTECH's chip engineering expertise.
UBTECH, for its part, contributes the robot hardware, system integration, and application scenarios.
Image Source: UBTECH
Xixuan Chuangzhi Technology's first chip is defined as a "big and small brain integrated" SoC designed for robotics. Using a general-purpose GPU as the core computing unit, it plans to build a toolchain around a CUDA-like ecosystem. The goal is to reuse developers' existing models and applications, minimizing adaptation costs when switching underlying computing hardware. Whether a truly smooth migration is achievable remains to be seen once the platform launches.
Xixuan Chuangzhi Technology repeatedly emphasizes that this is not a simple transplant of a generic chip. Instead, it is using real-world robotics tasks to drive chip design, incorporating requirements for multimodal perception, cognitive decision-making, and motion coordination into the product definition phase.
To support this, Xixuan Chuangzhi Technology's team is structured around "chip expertise plus commercialization." CEO Yang Xianbo is a former senior vice president at New H3D; CTO Li Zhaoshi holds a Ph.D. in integrated circuits from Tsinghua University; and Hardware VP Wei Li has extensive experience in tape-out and mass production of multiple general-purpose GPU chips. MetaX founder Chen Weiliang serves as chairman, while UBTECH founder Zhou Jian acts as vice chairman.
Zhou Jian made the intent clear: "We hope to bring the requirements accumulated through robotics R&D and application into chip design earlier, and validate the product through real-world tasks."
Notably, while Xixuan Chuangzhi Technology plans to use a single "big and small brain integrated" SoC to handle multimodal perception, cognitive decision-making, and motion coordination, CASIC Zhongke Huiling has mapped out an X100+P100 dual-processor solution for next-generation humanoid robots, separating high-level decision-making from real-time motion control.
However, CASIC Zhongke Huiling has also verified the feasibility of the X100 single chip handling both types of tasks, suggesting the two approaches are not mutually exclusive.
Ultimately, whether a route is suitable depends not just on the number of chips, but on real-time performance, power consumption, cost, and system-level validation.
Ambitious Targets, But First Product Arrives in 2028
Evaluating this funding round requires separating ambition from reality: "Surpassing Thor" is a goal written into Xixuan Chuangzhi Technology's R&D roadmap, not an accomplished fact.
Image Source: UBTECH
According to the timeline, Xixuan Chuangzhi Technology plans to provide a development and verification platform to industry partners in 2027 for adaptation and testing, with the first SoC product and accompanying embodied intelligence solutions slated for launch in 2028.
In terms of actual progress, Xixuan Chuangzhi Technology is still working on the architecture design of its proprietary main controller and has not yet completed tape-out.
A 4 billion yuan valuation at the angel stage for a company with no product or tape-out reflects capital market expectations for the edge computing sector, rather than a valuation based on existing performance.
But the higher the expectations, the greater the pressure to deliver in two years. The company must clear engineering hurdles like tape-out and mass production, while also solving software ecosystem and customer adaptation challenges. Any delay in these steps will push the timeline back.
Of course, achieving technological autonomy is an unavoidable backdrop to this investment, but domestic substitution alone cannot substitute for product competitiveness.
Ultimately, whether a robot chip succeeds will depend on specific factors: energy efficiency and real-time performance under real-world conditions, stability of motion control, whether the total cost can beat existing solutions, and whether a CUDA-like ecosystem can convince OEMs and model developers to make the switch.
