INFIFORCE Raises Nearly 1 Billion RMB Series A to Build AtomBrain Embodied AI Brain

Forum topic · 小凯 · 2026-08-15

Summary

Chinese embodied intelligence startup INFIFORCE announced on August 14 the completion of Series A and A+ funding rounds totaling nearly 1 billion RMB. The round was led by Dunhong Asset and a top state-owned capital platform, with follow-on investment from Zhejiang University Sci-Tech Innovation Group, Yandu Guokong, Lishui state capital, and existing investor Creation Capital (CCV). Funds will go toward the AtomBrain causal world model, the full-stack AI infrastructure system DataGrid, and scaled delivery of multi-form robot hardware. INFIFORCE's core bet is ego-centric (first-person) data: robots collect interaction data through their own sensors and manipulation rather than internet video or expensive teleoperation. Its AIM world model achieved 93.1% task success on the RoboTwin 2.0 benchmark, while early Atom models hit 97.0% on LIBERO. The company reports commercial deployment across 100+ real-world scenarios in 30+ Chinese cities, covering commercial service, warehousing, and manufacturing, with a product matrix spanning wheeled humanoid, dual-arm, and specialized robots. Leadership includes former Alibaba VP Bai Huiyuan as CEO and UC Berkeley PhD Wang Yizhou, formerly of Baidu IDL and NVIDIA DriveAV, as CTO.

INFIFORCE Raises Nearly 1 Billion RMB Series A: AtomBrain Turns "Robot's Own Eyes" into a Data Moat

On August 14, embodied intelligence company INFIFORCE announced the completion of its Series A and A+ rounds, raising a combined total of nearly 1 billion RMB. The round was led by Dunhong Asset and a leading state-owned capital platform, with participation from Zhejiang University Sci-Tech Innovation Group, Yandu Guokong, Lishui Municipal State-Owned Capital, and other industrial investors. Existing shareholder Creation Capital (CCV) increased its stake. The funds will focus on three areas: the AtomBrain causal world model, the full-stack AI Infra system DataGrid, and scaled delivery of multi-form robot hardware.

Breaking Down the Funding: Four Types of Capital Betting Together

The most interesting signal in this round is the investor mix:

Dunhong Asset: a top hard-tech investor focused on advanced manufacturing and embodied intelligence
Leading state-owned platform: strategic long-term capital representing local government industrial priorities
Zhejiang University Sci-Tech Innovation Group: a ZJU industry-academia-research platform bringing university resources
Yandu Guokong and Lishui state capital: local industrial capital from Yancheng (Jiangsu) and Lishui (Zhejiang)
Creation Capital (CCV): existing investor doubling down, signaling continued institutional confidence

Financial capital + strategic state capital + local industrial capital + university research — this combination is rare in the embodied AI sector. It suggests INFIFORCE is positioned not just as a technology company, but as a "regional-level brain company" embedded in local government's embodied intelligence industry maps.

AtomBrain: An Engineering Bet on Ego-Centric Data

INFIFORCE is betting on a route no one has fully proven yet: Ego-Centric (first-person perspective) data.

Mainstream embodied models rely on two data types: internet video (large in volume but noisy human-activity footage from YouTube/Bilibili) or teleoperated collection (high quality but extremely costly and hard to scale). INFIFORCE chose a third path: let robots see with their own eyes and act with their own hands, collecting first-person interaction data.

The analogy: human babies don't learn object manipulation by watching videos — they look with their own eyes, touch with their hands, feel force feedback, and iterate. The core advantage of Ego data is that it reflects what the robot genuinely perceives, without third-person annotation bias.

Around this route, INFIFORCE built the Atom series of native embodied foundation models:

AtomBrain: a causal world model unifying visual observation, language instructions, action trajectories, and environmental feedback (force, position, velocity) in a single framework
AIM world model: third generation, achieving 93.1% task success rate on the RoboTwin 2.0 benchmark
Early Atom models: 97.0% task success rate on the LIBERO benchmark

Model training is supported by the company's proprietary end-to-end data system, DataGrid — covering UMI handheld grippers, Ego collection devices, and real-robot teleoperation, forming a pipeline from "real-world experience → model capability."

Commercial Deployment: Engineering Delivery Across 100 Scenarios in 30 Cities

Publishing models isn't enough; the real barrier in embodied AI is validation in real environments. INFIFORCE disclosed that, built on the unified AtomBrain, it has formed a complete product matrix covering dedicated, general-purpose, and humanoid robots — including the AstroDroid wheeled humanoid, UltraDroid multi-form general robot, the "Little Atom" biped, and the FORCE series of specialized robots.

Key numbers: commercial deployment progressing in 30+ cities and 100+ real scenarios nationwide, spanning commercial services, warehouse logistics, and industrial manufacturing.

100 scenarios is not a small figure. Even at 1–2 robots per scenario, that's 100–200 real deployments — enough to run the Ego data flywheel (more deployments → more data → stronger models → better deployment outcomes).

Team Background: Alibaba, Berkeley, NVIDIA DriveAV

Founder & CEO Bai Huiyuan: former Alibaba Vice President; gave the talk "The Force Awakens: The Rise of Embodied Intelligence" at TEDxXuhui
CTO Wang Yizhou: PhD from UC Berkeley; among the first core members of Baidu IDL's autonomous driving team and a core NVIDIA DriveAV member
Chief Scientist Chen Jiayu: Assistant Professor at the University of Hong Kong, CMU postdoc
Core R&D team: over 75% hold master's or doctoral degrees, with members from Alibaba, Huawei, Baidu, NVIDIA, and other leading companies
IP: 100+ cumulative patents and software copyrights
Industry standards: In July 2026, a GB/Z national guideline, "Technical Specification for Crowdsourced Data Collection and Management for Embodied Intelligence," initiated and led by INFIFORCE, was approved for project establishment — the first national-level data specification for embodied intelligence

Industry Signal: From "Whose Body Is Better" to "Whose Brain Is Smarter"

One line in INFIFORCE's announcement stands out: "The core of embodied intelligence competition has shifted from 'who can build the robot body' to 'who can keep making robots smarter.'"

Over the past two years, embodied funding has concentrated on hardware companies — humanoid robot makers, robotic arm companies, sensor companies. That logic is now being questioned: good hardware does not equal good intelligence. Tesla's Optimus is already highly advanced in hardware, yet its intelligence on factory tasks remains the bottleneck.

INFIFORCE's bet is on brains + data. If robot hardware companies "build the body," INFIFORCE is "building the brain plus the real-robot data loop needed to train it."

Four verification points to watch over the next 6–12 months: 1. AtomBrain's public results on third-party embodied benchmarks (RLBench, RoboTwin 2.0, Calvin) 2. How many of the 100 scenarios achieve stable operation for 6+ months 3. Customer repurchase and renewal rates across the 30 cities 4. Whether genuine "brain-cross-body transfer" cases emerge (the same AtomBrain running on different manufacturers' robot bodies)

Data sources: INFIFORCE announcement, Blueshark Hard Tech funding watch, Aug 14, Tencent embodied intelligence daily, Aug 15, 10jqka report on HarmonyOS Plan, Aug 15.