Weifan Intelligent Secures Over 100M Yuan Seed+ Funding for Robot Edge Chips
Beijing Weifan Intelligent Technology Co., Ltd. recently announced the completion of a Seed+ round exceeding 100 million yuan. According to the company, existing shareholders Yanchen Group and Haiyi Investment continued to increase their stakes, with participation from funds under CICC Capital, Jiukun Venture Capital, Zeyu Capital, Shanhan Chaoxin, Daohe Yuanqi, and Jiuzhao Capital. Humanoid robot maker Songyan Dynamics also joined as an investor. The proceeds will be used for mass-production chip R&D, productization and commercial delivery of the BrainOS operating system and OmniRT deployment toolkit.
This is the company's latest disclosed funding round. In May this year, Weifan had completed a seed round of several hundred million yuan, co-led by Zhongguancun Capital and its Qihang Investment, with participation from Shanghai Future Industry Fund, Shixi Capital, Biwin Storage, Yanchen Group, Haiyi Investment, and Tanyuan Venture Capital. The two rounds are about three months apart.
Founded in May 2025 and incubated from Peking University's PAICORE Lab, Weifan focuses on developing integrated 'big brain and small brain' chips for embodied intelligence, combining perception and reasoning with motion control on a single chip. The company targets two industry pain points: insufficient edge computing power for robots and over-reliance on overseas toolchains for deployment. Its technology stack consists of three parts: the BiGPU dual-mode isomorphic brain-inspired chip, which fuses spiking neural networks with general-purpose computing; the BrainOS operating system for robot edge devices, with framework development completed; and the OmniRT deployment tool suite, which will open to initial strategic partners in September, offering edge deployment, low-power optimization, SOTA algorithm adaptation, and compute tuning services.
According to the company, the PAICORE 2.5 test chip, co-developed with Peking University's PAICORE Lab, has completed tape-out and return, with prototype board design and initial performance testing done. In an internal demo video, the GR00T N1.6 model with OmniRT optimization runs stably above 10 FPS on Jetson Orin AGX and above 20 FPS on Jetson Thor. For reference, public benchmarks (FP16, batch size 1) show 7B OpenVLA at 1.5-3Hz, 3.3B π0 at 5-8Hz, and 3B OpenPI at 3-5Hz on Jetson AGX Orin; NVIDIA's official demo of GR00T N1's VLM module at 10Hz used a 300W L40 server GPU, while Jetson AGX Orin is a 60W edge module. Weifan claims an order-of-magnitude energy efficiency improvement, running 3B-parameter VLA models above 10 FPS at about one-fifth the power of a server card.
On commercialization, the company says it has engaged with UBTech Robotics, Beijing Humanoid Robot Innovation Center, Haier Group, and Songyan Dynamics, completing joint debugging and scenario tests on multiple prototypes. It will offer low-power optimization, algorithm adaptation, and customized deployment services. The team has grown to about 100 people in the past six months, covering chip architecture, design verification, embodied algorithms, and software engineering, with core members from leading semiconductor and robot companies. In May, a Beijing key laboratory for 'brain-inspired intelligent computing chips and systems' jointly established with Peking University and Beijing University of Posts and Telecommunications was approved.
Investors commented positively. Yanchen Group said the upper limit of robot intelligence depends on edge computing power, and edge AI inference chips are the highest-value component in the system, with huge market prospects. They believe Weifan is expected to become a leader in domestic chips for core robot computing. A fund under CICC Capital noted that the embodied intelligence track lacks not single-point chips but complete 'chip + system + toolchain' delivery capability. Weifan uses software ecosystem and customer co-development to drive commercial adoption, while building a universal compiler for general and brain-inspired computing, positioning itself as the key carrier of the embodied intelligence brain.