---
title: "Three Rounds in Three Months: Hangzhou's Bet on a Robot Brain"
date: 2026-11-20
category: Capital & Markets
site: NeuroAI
canonical: https://neuroai.site/a/na-capital-infiforce-state-capital
language: en
---

# Three Rounds in Three Months: Hangzhou's Bet on a Robot Brain

> An embodied-intelligence startup in Hangzhou has drawn market, provincial and municipal money in quick succession — a map of how China now funds the layer between robots and models.

A robot in a car-parts plant reaches into a bin of brass fittings, misses, and adjusts its grip. The failed attempt is not thrown away. It is logged, tagged and fed back into the model.

That habit — treating mistakes as training material — is what a cluster of investors in eastern China is now paying to scale.

The company involved has just added a backer that does not usually write venture cheques.

The startup is INFIFORCE (原力无限), a Hangzhou firm working on what it calls an embodied brain (具身大脑) — the software layer that lets a machine understand a physical scene, decide what to do and keep learning afterwards. Its newest round, a Pre-B, was taken in full by Hangzhou Capital, the city's state-owned investment and operating platform. The size of that cheque was not disclosed as a precise figure. It came after a round of close to RMB 1 billion (≈ US$140M / HK$1.07B) in August and a nearly RMB 100 million (≈ US$14M / HK$107M) investment from Zhejiang Province's venture capital arm in October, according to reporting by the Beijing News and Hangzhou Daily.

## A stack of investors, not a single whale

What makes this sequence unusual is not the total. It is the layering.

The August financing was led by Dunhong Asset Management and a state-owned platform, with participation from Zhejiang University's science-and-investment group, Yandu state assets and the Lishui municipal state-owned company, plus returning backer CCV. October brought in Zhejiang Venture Capital Group, a provincial-level fund. The latest cheque comes from the city itself.

In practice that means three different tiers of Chinese capital — market-oriented funds, provincial platforms and a municipal platform — have each written a separate ticket within roughly a quarter. Chinese robotics coverage often compresses this into "a funding round." The reality is closer to a relay, in which each new investor validates the last one's thesis before the next hand-off.

For founders, that structure is both a blessing and a constraint. Money arrives in stages, tied to visible milestones. It also means the company's roadmap is being read by institutions whose mandates are not purely financial: provincial and municipal platforms are usually also buying local industrial capacity, jobs and a reason for a region to be named in a press release.

## What INFIFORCE says it will build

The company's public descriptions centre on three pieces.

- **AtomBrain**, the embodied brain itself, built around what the team calls implicit world models and continual learning. The pitch is that a robot should form an internal model of how objects behave, then update it from real interactions rather than being retrained from scratch for each new task.

- **DataGrid**, described as a full-stack AI infrastructure layer for collecting, organising and reusing physical interaction data.

- **Multi-form robot bodies** deployed in factories and commercial sites, which generate the feedback the first two layers need.

The research claim is that the company has produced more than twenty academic papers or results accepted at venues including ICLR, ICML, ECCV, IROS, AAAI and IEEE TPAMI, and has iterated through two generations of its base model. It says its systems now touch more than thirty cities and close to a hundred application scenarios, including car manufacturing, home appliances, warehousing, retail and infrastructure. Its planning and design institute has signed a strategic agreement with China Railway Hi-tech Industry (中铁高新工业), and the company is taking part in a national-level embodied-intelligence pilot base in Hangzhou.

These are company and local-media statements, not independently audited figures. That distinction matters, and it is the reason this story is about financing behaviour rather than about proven performance.

## Why causal world models became a funding category

Two years ago, a robotics pitch in China was mostly about the body: how many degrees of freedom, how fast it walked, whether it could do a backflip. In 2026, the money has moved up the stack.

Investors are now funding the argument that the scarce resource is not actuators but the ability to predict consequences. A robot that has seen thousands of videos of a cup being pushed may learn that cups fall. It may not learn why, so the prediction breaks when the cup changes shape or the table changes surface. "Causal world models" are the proposed fix: represent the relationship between action, state and outcome, so that experience transfers to a new setting.

That idea is not unique to this company. It has become a shared vocabulary across Chinese embodied-AI funding, which is precisely why it is attractive to state investors. It is a layer that, in theory, sits underneath any robot body and any downstream industry — useful to a car plant, a pharmacy or a logistics hub alike. A regional fund buying into that layer is not betting on one machine; it is buying an option on many.

Hangzhou Capital's own scale is part of the story. Reports put its consolidated assets at more than RMB 100 billion (≈ US$14B / HK$107B) at the end of 2025, and it operates two funds each described as being at the "hundred-billion-yuan" scale. Its stated remit covers artificial intelligence, robotics and brain-computer interface (脑机接口). A small Pre-B from such a platform is less a bet on a single startup than a signal of where a city wants its industrial base to sit.

## Key takeaways

- INFIFORCE (原力无限) received a Pre-B investment from Hangzhou Capital, a municipal state-owned platform, after a roughly RMB 1 billion (≈ US$140M / HK$1.07B) round in August and a nearly RMB 100 million (≈ US$14M / HK$107M) provincial investment in October.

- The three rounds came from different tiers of capital — market funds, a provincial platform and a municipal platform — rather than one large lead investor.

- The company is spending on an embodied brain (具身大脑), a data infrastructure layer and real-world robot deployments, not on a new humanoid (人形机器人) reveal.

- The underlying thesis is that prediction and continual learning, not hardware, are the scarce inputs in Chinese robotics in 2026.

## Honest limitations

The Pre-B amount was not disclosed as a precise figure, so the total raised cannot be added up reliably. Terms, valuations and equity stakes are unknown. Most operational claims — paper counts, city coverage, deployment numbers, the significance of the research results — come from the company or from local and business media repeating company material; none has been independently verified here. The word "brain" is a marketing category as much as a technical one, and different firms use it to mean different things. This article does not assess whether the approach works, only how the capital is being assembled. Currency conversions are approximate and use the rates implied by the cited reporting.

## What readers can do now

If you are tracking Chinese robotics, watch the investors rather than the robots: a provincial fund followed by a municipal platform usually means a local deployment pipeline is being prepared, and that is where the next announcements will appear. If you work in manufacturing, ask any vendor pitching an "embodied brain" to demonstrate the same task on an object it has never seen — transfer, not rehearsal, is the claim being funded. And treat the phrase "world model" as a question to put to a supplier, not a feature to accept on a slide.

---

Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-capital-infiforce-state-capital
Free to quote with attribution and a link to the original.
