Some investors in embodied intelligence once committed funds in as little as a week, sometimes without conducting preliminary research, according to anecdotal accounts.
Now, those same investors tell founders their projects must go before the board, often with no substantive response afterward.
For entrepreneurs developing artificial intelligence systems that can perceive and act in the physical world, the fundraising climate shifted sharply in fall 2026. From Beijing to Shenzhen, raising capital became markedly more difficult.
Three leading investment firms previously enthusiastic about the sector have tightened their spending, 36Kr learned. At one firm, investors were explicitly told to reduce investment in embodied intelligence and assess projects cautiously. At another, almost no embodied intelligence deals had reached investment committee review in the preceding six weeks.
A Shenzhen founder’s fundraising was progressing normally before September. When he approached financial advisors after that, one told him the firm was no longer accepting embodied intelligence projects. Within two weeks, he had become an unwelcome prospect.
One financial advisor estimated that the number of embodied intelligence projects reaching the firm had fallen by about half. That was not entirely unwelcome, the advisor told 36Kr: deals were no longer closing.
“Projects we could previously place in a month are now stuck. We don’t expect to get them done in less than three to six months,” the advisor said. The firm has had to raise its threshold for accepting clients.
For advisors, embodied intelligence quickly went from a priority to a difficult sell. Some now question founders’ feel for the market, asking why they missed the fundraising window a few months earlier.
One financial advisor privately expressed more regret than criticism. Had one company approached the market at the right moment, he said, it could have closed quickly at a favorable valuation. Its difficulty now reflected how much harder it had become for investors to make decisions.
Several industry participants identified September as the turning point, although none could pinpoint a single event that started the reversal. NE Times reported that funding in embodied intelligence fell by roughly 40% in September from August.
Several developments converged within weeks. Unitree Robotics, regarded as a bellwether, saw its market capitalization fall by half at one point after listing. The CEO of Mech-Mind Robotics then alleged in a WeChat post that some companies were fabricating revenue. Rumors that several robotics IPOs had been paused began circulating privately.
Only months earlier, a founder had proudly shown 36Kr the trajectory of his company’s valuation: from RMB 200 million (USD 29.9 million) in February to more than RMB 3 billion (USD 448.7 million) in July, a roughly 15-fold rise in five months. Other companies fought publicity battles over which would first reach RMB 20 billion (USD 3 billion), while leading players sought to undermine one another in the race to list.
The retreat in capital has made those contests look increasingly disconnected from the underlying businesses. Over three years, the sector produced more than 100 startups and several companies valued at RMB 10 billion (USD 1.5 billion) or more. Most now face an uncertain wait.
One investor said some peers had stopped following embodied intelligence altogether. Others remained optimistic but were holding on to their cash.
In his view, the current investment cycle is nearing its end. He expects a period of volatility before the industry stabilizes, possibly in the second half of 2027 or even the first half of 2028.
How dealmakers built unicorn valuations
Some of the sector’s highest-profile valuations could have been driven as much by dealmaking as by the businesses themselves, 36Kr reported.
Investors described a financing model in which well-connected organizers assemble a startup around a chosen founder. Candidates might be technically gifted scientists or experienced executives from large technology companies. The organizers are often investment firms with little public recognition but substantial local connections and capital networks.
Once a candidate agrees to start a company, the organizers present a plan designed to raise its valuation through successive rounds and bring in the next set of investors.
First comes positioning. A founder’s former employer, technical reputation, or entrepreneurial record is packaged into a story and introduced to the private investment market.
One financial advisor told 36Kr that prospective investors were sometimes given only two days to decide, with a deadline the following evening. Further documentation and interviews with the founder were unavailable.
If an investor hesitated, organizers would invoke their relationship and warn that the allocation could go to someone else. Such pressure could push inexperienced investors into decisions they would otherwise question. “The overall mood was intense FOMO,” one source said, referring to the fear of missing out.
More important than creating urgency was planning the financing sequence. An investor said some organizers brought leading USD-denominated funds into early rounds, then encouraged them to keep investing. Repeated commitments from those firms became a selling point for new investors.
The organizers also advised companies on winning orders. Some interviewees told 36Kr that this extended to “manufacturing” revenue.
In the investor’s account, a typical first-round valuation started at around USD 100 million. The second round would often roughly double it, to about RMB 1.5 billion (USD 224.4 million), followed by RMB 2.5–4.0 billion (USD 373.9–598.3 million) in the third. Projects with an especially appealing story could be oversubscribed and jump to RMB 5 billion (USD 747.9 million) or more in the third round.
Investors did not necessarily have to wait for a robot company to commercialize its technology. Higher valuations in later rounds could generate paper gains. Technical teams, investors, and financial advisors understood the incentives and kept the process moving toward a pre-IPO round.
The mechanism depended on an eventual listing. When the route to an IPO became uncertain and new investors grew reluctant, the sequence of rising valuations began to weaken.
Falling stocks and advancing AI unsettle investors
Some investors had started deliberately slowing their activity in the first half of the year.
“Anyone inside the market knew the investment had become frenzied and behavior had become distorted. Once investment reached a peak, it was bound to fall,” one investor said.
What they could not predict was when the bubble would break, or what would trigger it. Several industry participants told 36Kr that the cooling reflected developments across private markets, public markets, and the research community.
After three or four years of intensive investment, leading firms had largely built out their embodied intelligence portfolios.
By 2026, investment had reached most of the industry’s major technical bottlenecks. In 2024, money poured into robotic bodies. In 2025, it shifted toward robot “brains” and components such as dexterous hands, sensors, tactile systems, and lifelike robotic faces. In 2026, it moved toward embodied data and world models. Almost every segment produced companies valued at RMB 10 billion or more.
For an industry still struggling to commercialize, sustaining so many unicorn valuations indefinitely was always going to be difficult. For a time, participants could find another buyer at the right moment.
Under that approach, higher valuations in the public market justified higher prices for private startups. “The higher companies like Unitree rise, the better it is for unlisted companies raising money,” another investor said.
Unitree’s share price decline suggests public markets are no longer delivering the returns investors expected.
Mech-Mind, which listed in September, traded below its offering price. Geek+ had also fallen below its offering price, while Dobot was well below its record high.
Those companies at least reached the market. Investors were more troubled by the prospect that companies in their portfolios might not obtain approval to list. They could not say how policy would develop.
Technical progress is also challenging earlier assumptions.
Over the past two years, startups using vision-language-action (VLA) models, which connect perception and instructions to actions, and world models have sought to develop general-purpose embodied intelligence. OpenAI’s September release of GPT-6 Astra, although not trained specifically for robots, prompted investors to question those startups’ technological advantage.
Unitree CEO Wang Xingxing said in a recent speech that, in the company’s tests, GPT-6 Astra could direct a robot to paint using prompts alone, without additional training. He said the robot also changed paints and cleaned its brush.
General-purpose models could eventually break through into physical AI, Wang argued.
36Kr also learned that Chinese foundation model developers, including DeepSeek, had recently been in frequent contact with embodied intelligence companies to explore the sector. ByteDance was recruiting engineers and leaders for embodied AI models.
One investor put the challenge plainly: if a startup spends nearly RMB 10 billion training a model whose capabilities a general-purpose foundation model could eventually match, would its technology still justify a valuation above RMB 10 billion?
Raising cash for a longer wait
Bargaining power has shifted.
A founder developing robot intelligence told 36Kr that the company had changed its fundraising strategy. Previously, it sought a large round led by a prominent investor, with several well-known firms participating. Now, it would begin a smaller round whenever an investor was willing to commit.
“We have to move quickly and stock up for the winter,” the founder said.
Some highly valued companies are narrowing their operations. 36Kr learned that one leading embodied intelligence company valued at more than RMB 20 billion was restructuring, bringing previously separated subsidiaries back together and streamlining R&D. Some companies valued above RMB 10 billion had also laid off staff or withdrawn job offers.
Investors are less willing to pay for an ambitious vision alone. One said the firm now required a clear commercial application, genuine market demand, and, preferably, a founding team member with industry connections and the ability to put products to use.
Projects once overlooked are attracting some of the remaining capital. Several investors told 36Kr that money was moving toward companies with identifiable customers and real orders, even if they lacked prominent technical founders or ambitions to build general-purpose robots. These included suppliers of components, robotic joints, and equipment.
“People can finally return to normal investment judgments,” one investor said.
A founder identified a deeper problem: expectations for how quickly robots would develop had become unrealistic.
“It took cars 200 years to become what they are today, but people want robots to get there in five,” the founder said. “Capital believes in the future too early, then loses patience too early when it encounters setbacks.”
KrASIA features translated and adapted content that was originally published by 36Kr. This article was written by Qiu Xiaofen for 36Kr.
Note: RMB figures are converted to USD at rates of RMB 6.69 = USD 1 based on estimates as of November 5, 2026, unless otherwise stated. USD conversions are approximate and, where appropriate, rounded for ease of reference. They may not fully match prevailing exchange rates.