A quiet corner of Beijing's Yizhuang (亦庄) district is becoming a pressure point in China's robotics supply chain. The companies building the humanoid bodies get the headlines, but the chips that decide how fast those bodies think are where the strategic money is now concentrating.
Huixi Intelligent (辉羲智能), a fabless designer of large-compute chips for embodied intelligence (具身智能), said on October 22 it has closed its Pre-B+ and B rounds. With those completed, the company's cumulative equity financing for 2026 has crossed ¥2.5 billion (≈ US$350 million). The round was led by the Beijing AI Industry Investment Fund (北京市人工智能产业投资基金), with the National AI Industry Investment Fund (国家人工智能产业投资基金) and the Beijing Robotics Industry Investment Fund (北京机器人产业发展投资基金) among the participants.
Why a chip company, not another robot builder
The raise tells a story about where strategic capital thinks the real bottleneck sits. Embodied intelligence needs on-robot compute that is cheap, low-latency and power-efficient enough to run large models in real time. That is a harder problem than bolting a GPU into a metal frame, and it is the layer Huixi was built to own.
- Founded in 2022 and headquartered in Beijing Yizhuang, Huixi organized itself around a "data closed-loop defines the chip" (数据闭环定义芯片) philosophy.
- Its flagship product is the Guangzhi R1 (光至 R1), a high-performance AI system-on-chip built on a general programmable architecture.
- Around the silicon it has assembled a software stack spanning model frameworks, compilers, operator libraries, runtime, middleware, operating system, simulation and real-time control.
- The roadmap also names an RPU compute core, a RISE compute platform and a RhinoVLA (RhinoVLA 模型) model as the pillars of its technology system.
Founder and CEO Xu Ningyi (徐宁仪) is a Tsinghua University electronic-engineering graduate who previously led the hardware computing group at Microsoft Research Asia and served as chief architect of Baidu's smart-chip department, where he designed the Kunlun (昆仑) cloud AI chip. That pedigree is part of why the company could pitch itself as a "compute platform" rather than a point-solution vendor.
What the money is actually for
Huixi said the proceeds will go toward three things:
- Research and development of the next generation of large-compute chips for embodied intelligence.
- Faster scaled delivery of those chips to robot makers.
- Building out the industrial ecosystem, including the "Huixi Embodied Computing Ecosystem Alliance" (辉羲具身智算生态联盟) announced alongside the raise.
The company also showed its Huixi Embodied (Huixi Embodied 算力产品系列) product line at the 2026 World Robot Conference (世界机器人大会), signaling it wants the chip to sit inside many robot forms rather than one flagship product.
The proof point investors cite
The reason the state funds showed up is a deployment milestone rather than a slide deck. A computing platform built on the Guangzhi R1 was integrated into AgiBot's Lingxi G2 (智元精灵 G2) robot. In June 2026, several of those robots ran a continuous six-day live-streamed production-line trial and completed 17,625 tasks at a 99.99% success rate.
That is the kind of number robot makers care about, because it moves a chip from "validated in a lab" to "trustworthy on a factory floor." Huixi says its products now reach industrial manufacturing, smart retail, commercial catering and unmanned logistics.
A state fund that keeps showing up
The National AI Industry Investment Fund behind this round is worth understanding on its own. Established on January 17, 2025 and overseen by the Ministry of Industry and Information Technology and the Ministry of Finance, it carries a total scale of ¥60.06 billion (≈ US$8.5 billion). Its sole limited partner is Phase III of the national integrated-circuit fund (国家大基金三期), which committed ¥60 billion, with Guozhitou (国智投) acting as manager.
The fund is explicitly built to "invest early, small and at the frontier" (投早、投小、投前沿) across the whole AI chain — compute, algorithms, data and application. Before Huixi, its disclosed bets included China Telecom's AI unit, the smart-driving chip firm Xinchip Hangtu (新芯航途), the AI-security company Chaitin (长亭科技) and the embodied-robotics maker Galaxy General (银河通用). Huixi is therefore a continuation of a pattern: the state is underwriting the plumbing of embodied AI, not just the finished machines.
Who else is at the table
Beyond the lead and the national fund, the round drew a long roster of government and market investors, including Hidden Hill Capital (隐山资本), Loyal Valley Capital (正心谷资本), GP Capital (金浦投资) and a set of municipal vehicles, while existing backers Shunwei Capital (顺为资本), Genesis Capital (元生资本), Cathay Capital (凯辉基金) and SenseTime's Guoxiang Capital (商汤国香资本) followed on. The mix of "national fund plus municipal fund plus market institutions" is becoming the standard template for Chinese deep-tech rounds of this size.
Deployment takeaways
According to Li An, chief scientist at BrainNet (脑机网), the strategic signal here is less about one company's valuation and more about state capital explicitly treating robot brains as infrastructure rather than a component to be imported.
Honest limitations
- The ¥2.5 billion figure is a cumulative 2026 equity total, not a single-round size; the split between the Pre-B+ and B tranches was not disclosed.
- Participant names come from Chinese financial press (Securities Times / 证券时报 and Star Market Daily / 科创板日报); Huixi has not published a cap-table breakdown, so exact stakes are unknown.
- A 99.99% success rate from a vendor-disclosed live demo is not the same as independent, audited field reliability. Treat it as a strong signal, not a guarantee.
- China's "embodied intelligence" chip field is crowded, and Huixi's moat depends on software stickiness, which is slower to prove than hardware specifications.
- The National AI Industry Investment Fund's ¥60.06 billion scale and January 2025 establishment are documented facts, but its pace of deployment and ultimate returns remain unproven.
What readers can do now
- Track the Beijing AI Industry Investment Fund's subsequent deals — it is becoming a leading indicator for which embodied-AI layers the state prioritizes.
- If you evaluate robotics supply-chain equities, separate "body makers" from "brain makers"; the chip layer carries different margin and consolidation dynamics.
- Watch AgiBot's procurement signals: continued use of Guangzhi R1 across multiple models would be the cleanest third-party validation of Huixi's deployment claims.