Huixi closes Pre-B+ and Series B, lifting 2026 funding for robot chips past 2.5 billion yuan

Huixi Intelligence (辉羲智能), a Beijing company that designs compute chips for embodied AI, said on Oct. 22 it has closed Pre-B+ and Series B rounds that take its 2026 equity financing past 2.5 billion yuan. The Beijing Artificial Intelligence Industry Investment Fund led, with the Beijing Robot Industry Development Investment Fund and the National Artificial Intelligence Industry Investment Fund among the new investors; existing backers including Shunwei Capital, Cathay Capital and SenseTime's Guoxiang Capital put in more.

The money is a bet on robot demand, yet the evidence of that demand Huixi puts forward still comes mainly from one customer that also owns part of it. Its Guangzhi R1-based compute platform went into AgiBot's Genie G2 robot in 2025, and its headline demo is an AgiBot one too: in June 2026, several R1-equipped G2 robots completed 17,625 operations over six days of livestreamed production-line work with a 99.99% task success rate, Huixi says. AgiBot took a strategic stake in Huixi's Pre-B round in February, which raised more than 800 million yuan.

A Beijing state AI fund now leads where Xiaomi once did

Huixi was founded in April 2022 and is based in Beijing's Yizhuang district. Xiaomi led its angel round in December 2022 with more than US$50 million, and SenseTime joined a Series A in September 2024, according to business records cited by Xindongxi. This year the lead is the Beijing Artificial Intelligence Industry Investment Fund, with Beijing's robot industry fund and the national AI fund among the new investors. Founder and CEO Xu Ningyi ran the hardware computing group at Microsoft Research Asia and led design of Baidu's Kunlun AI chip; co-founder Zhang Jianyong was assistant vice president of autonomous driving at NIO.

One chip family sold to robots, delivery vehicles and cars

Huixi's case against being read as AgiBot's supplier is breadth. It says R1 MAX, rated at 500 TOPS, is in volume delivery to multiple embodied-AI customers, while R1 PRO, at 375 TOPS, puts CPU, NPU, GPU and real-time interfaces on one chip designed for robots. Above the silicon sit the RISE compute platform and RhinoVLA, a pre-trained VLA model. Its volume products also go into L4 driverless logistics vehicles and passenger-car driver assistance, and the company calls humanoid robots a "narrow entry point."

That entry point is getting crowded. Black Sesame launched its SesameX embodied computing platform at this year's World Robot Conference, Axera brought an embodied "brain" controller and its AX8910 vision chip to the World Artificial Intelligence Conference, and Qualcomm aimed its IQ10 series at full-size humanoids at CES in January. Robot chip volumes, Xindongxi notes, are still small next to driver-assist chips.