Lin Fangzhou reporting from ROBO-Qbit | Edited by Li Huanhuan | QbitAI ROBO | WeChat Official Account: AI4ROBO
One investor has not reviewed an embodied AI project in nearly six months.
Another investor, who started tracking the embodied AI sector in 2023, scanned through projects in the second half of this year and made only one investment.
A third investor revealed that several leading embodied AI companies are currently raising new rounds. However, among many existing shareholders, there is significant internal debate over whether to participate.
It is clear that embodied AI has entered a period of capital cooling.
Data released by the industrial service platform "NE Era" confirms this trend: embodied AI financing volume in September shrank by 40% compared to August.
The fundraising journey that other industries previously took years to complete was compressed into two or three years for embodied AI. Valuations soared relentlessly, but the maturity of technology, products, and commercialization failed to keep pace.
Facing persistently high corporate valuations on one side and uncertain expectations on the other, the embodied AI industry has encountered its first capital winter.
Not Just Embodied AI Cooling Down
Financial Advisors (FAs) are most sensitive to market temperature—the investment wind direction in the embodied AI sector shifted around September.
In August, an embodied AI project handled by a certain FA was progressing as planned; by September, the investment institutions reviewing the project gave consistent feedback.
The project is good, but it is priced too high.
The intended valuation of CNY 1 billion for the project dropped to a psychological price range of CNY 700-800 million at institutions just one month later.
Now, this FA no longer accepts embodied AI projects for a simple reason: they are hard to sell.
Embodied AI companies themselves have also felt the chill.
A company specializing in embodied data used to receive inquiries about financing intent from investment institutions almost every week; after September, however, calls dried up.
"Financing across the entire industry has cooled down by 100% now," lamented an industry practitioner, noting that even state-backed funds, which were previously enthusiastic, have reversed their stance by 180 degrees.
This is not an isolated phenomenon.
An investor observed that not only embodied AI, but the entire primary market for high-valuation tech sectors is cooling down.
For instance, World Models, which were hotly pursued in the first half of the year, are experiencing a more pronounced retreat than embodied AI.
A star World Model company, once fiercely competed for in terms of share allocation, aimed to double its valuation in a new round but received a lackluster response, resulting in only a standard 30% increase.
No one can clearly say how many "World Model companies" exist in China today. Media reports have estimated that hundreds of companies claim to be deploying "World Models."
However, investors have gradually reached a consensus: the financing window has essentially closed for "prodigy" or "star professor" projects lacking practical application scenarios.
From FOMO to Losing Interest: What Happened?
In the first half of the year, capital was still trapped in the FOMO (Fear Of Missing Out) sentiment regarding embodied AI. Why did interest suddenly wane within just a few months?
The turning point occurred in September. Two events happened in the industry during this month, both related to the secondary market.
First, after Unitree Robotics went public, its stock price experienced a significant pullback.
As a bellwether for the sector, Unitree's performance in the secondary market directly cooled the imaginative space in the primary market.
Second, the CEO of a listed embodied AI company publicly accused some peers of inflating revenue on social media.
The incident quickly escalated and attracted regulatory attention, followed by rumors that the CSRC issued window guidance to certain institutions, resulting in a substantial tightening of IPO review standards for embodied AI companies.
Embodied AI companies queuing for listing on A-shares and HKEX were suddenly forced to hit the brakes.
The wealth creation myth of the secondary market had once fueled the ambitions of the primary market; now, facing headwinds, it has equally amplified capital's fear.
At the peak in the first half of the year, HK-listed companies were telling stories of valuations in the tens of billions; after lock-up periods ended in the second half, reality landed on the cold ground of market caps in the single-digit billions.
Expectations have completely changed.
An investor noted that when promoting embodied AI projects internally in the first half of the year, everyone believed the technology held promise; in the second half, promotions were met with a barrage of questions: Does the technology actually work?
The company remains the same, and the technology remains the same. In just a few months, the essence hasn't changed, but investors' emotions and mindset have shifted rapidly along with secondary market performance.
When Angel Round Valuations Exceed CNY 10 Billion, VCs Should Step Back
Beyond the emotional disturbances caused by the secondary market, looking at cyclical patterns, early-stage venture capital in embodied AI has naturally reached a phased conclusion.
Lü Xing, an investor at early-stage VC Zeran Venture Capital, told QbitAI ROBO:
Even without these events, our phased tasks are basically complete, and we will not invest as frequently as we did in the previous two years.
On one hand, high-quality targets are scarce, and investment win rates have declined. Most institutions completed their early-stage layouts by late 2022 or early 2023.
Lü Xing stated that it is not that there are fewer new projects in the market, but rather that the probability of new projects growing into industry leaders is increasingly low.
There are still opportunities in the market, but the time and effort required for validation far exceed those in other sectors, making it unattractive for early-stage VCs.
On the other hand, the current overall high valuation levels in the industry have also amplified investment risks.
An investor who initially entered a project at a CNY 300 million valuation saw it pushed up to CNY 5 billion.
Although he remains bullish on the company, continuing to chase investments at this stage has severely imbalanced the risk-reward ratio.
In Lü Xing's view, when "angel round valuations exceeding CNY 10 billion" appear in the market, it often signals the end of the mission for early-stage VCs.
Since the second quarter of this year, such embodied AI projects have emerged one after another.
As early-stage VCs gradually exit, the baton is being passed to industrial capital and Private Equity (PE).
Next, giants possessing production lines, implementation scenarios, and customer resources will drive the industry toward mass production and scaled delivery.
Seeking New Certainties in Segments and Upstream Sectors
Although the market is cooling, the investment and financing environment has not frozen over.
Leading embodied AI enterprises generally do not lack funds; investors judge that their previous financings are sufficient to support operations for at least three years.
The truly urgent test for embodied AI companies now is no longer "whether they can raise money," but "whether they can turn money into deployable products."
Lü Xing identified a notable shift: grand narratives of "General Intelligence" have temporarily fallen silent, while pragmatic "Scenario Implementation" has become the hard currency.
For example, specific tasks such as cleaning, logistics, and inspection have become the primary scenarios for companies to prove their self-sustaining capabilities.
This resembles the path Autonomous Driving once traveled.
After experiencing setbacks in pursuing top-down full-scenario L4 autonomy, the industry returned to a bottom-up, pragmatic path of accumulating data from specific scenarios.
Lü Xing believes the embodied AI ecosystem should resemble a pyramid.
The base layer consists of functional robots, such as robotic arms, logistics robots, floor-sweeping robots, and cleaning robots.
Above them are node-type robots, like wheeled dual-arm robots, applicable in semi-open scenarios such as coffee shops and inspections.
The apex comprises general-purpose humanoid robots.
A healthy development path for embodied AI should first solidify the base, accumulate real-world scenario data, and then evolve step-by-step upwards.
Currently, while paying more attention to commercial deployment, capital is also penetrating upstream and more segmented areas of the supply chain.
For instance, Lü Xing's firm recently invested in a dexterous hand company.
Although valuations of several leading dexterous hand companies are already high, upon examining shipment volumes, they believe a stable market structure has not yet formed.
The technical routes for dexterous hands have not fully converged. They judge that requirements vary across different scenarios, and various driving methods will coexist in the future.
Investors still tracking embodied AI projects are focusing not only on upstream components of the "Embodied Brain" such as robot chips, operating systems, and data, but hardware such as sensors, small joints, and reducers has also returned to their radar.
Their judgment is: before large corporations fully commit to these niche markets, small startups still have opportunities.
Liu Xin, an investor at Zeran Venture Capital, remains optimistic about the prospects of embodied AI. He told QbitAI ROBO:
The decline in investment growth rate and actual industry growth should be viewed separately.
Currently, the number of bodies, finished product orders, and demand for supporting supply chains continue to grow rapidly.
According to IDC data, global shipments of humanoid robots in the first half of 2026 approached 25,000 units, a year-on-year increase of 432.1%; Bank of America Research predicts that humanoid robot shipments will reach 1.2 million units by 2030.
Any emerging industry undergoes this classic curve: "Policy incentives and capital speculation—cooling off and bubble bursting—supply chain maturation and cost reduction—volume expansion," much like New Energy Vehicles (NEVs) in their early days.
Current fluctuations and occasional chaos are merely normal interludes in cyclical evolution, appearing in the development process of every industry, and do not affect the long-term underlying logic of industrial development.
The industry will not experience just one cycle. In the future, when valuations return to reasonable levels, technological breakthroughs are achieved, and business loops are validated, embodied AI may welcome its next spring.