The investor is easy to identify by reputation rather than name. He appears on the early cap table of one listed robot company and sits behind another of the sector’s leaders. For years he kept his distance from publicity while operating in the primary market with a style that other investors describe as aggressive and fast. One of them put it plainly: no lengthy diligence, just a decision, and the money arrives within a week.
That man has gone quiet. Founders who approach him now hear that the project needs to be shown to the board. Then nothing comes back.
His retreat is not idiosyncratic. Three of the sector’s most committed institutions have tightened at the same time. One told its investment team in a meeting to reduce embodied AI exposure and to judge such projects cautiously. Another has taken almost no embodied project to an investment committee in six weeks.
Half a month to become unwelcome
A founder in Shenzhen was raising normally until September. When he went looking for a financial adviser afterwards, he was told the firm does not take embodied projects anymore. In roughly two weeks he went from a company with momentum to a company nobody wanted to represent.
The adviser is not being cruel. A senior adviser at a leading firm said the embodied projects crossing his desk have fallen by about half. That sounds like less work. In practice it means the backlog is stuck: projects that used to sell in a month now sit, and he expects three to six months before any of them clear. The response has been to raise the bar for taking a mandate at all.
Founders are also being graded on timing. More than one has been asked why they missed the hot window months ago, and whether that says something about their sense of the market. One adviser kept returning to the same word, saying that if the company had caught the timing it could have closed quickly and at a good price, and that the failure was really about how expensive it had become for anyone to make a decision.
The data point that people cite is September. Industry data from NE Times shows embodied AI funding that month fell 40 per cent from August.
What broke in September
Several things happened within weeks of each other. Unitree, the sector’s reference stock, lost half its market value at the low. A chief executive of a listed embodied company used his personal social feed to accuse peers of fabricating revenue. Word began circulating privately that several robotics listings had been paused. Companies that had been expansive became, in the words of one observer, visibly tense.
The contrast with the months before is stark. One founder proudly showed off his valuation curve: RMB 200 million in February, more than RMB 3 billion by July. A fifteen-fold increase in five months. Companies ran public relations campaigns over who would be first through the RMB 20 billion mark. Others briefed against rivals competing for the same listing slot.
Once capital withdrew, that competition looked absurd. Three years of enthusiasm produced more than a hundred startups and several companies valued in the tens of billions of yuan. Most of them are now suspended, neither funded nor dead.
The assembly line that made the unicorns
The reporting describes a financing model that participants call assembled entrepreneurship. A resourceful intermediary, usually a firm with deep local networks and capital contacts but limited brand recognition, selects a founder profile: a brilliant scientist, or an executive with a large platform on the CV. The intermediary then persuades that person to leave and supplies a complete plan whose single objective is to raise the valuation and pass the parcel.
The packaging is the easy part. Big-platform background, technical genius, prior exits, all arranged into a story that can be shown to the primary market. Then the pressure begins. Advisers describe investment decisions compressed into two days, with a deadline of tomorrow evening and no time for founder interviews, backed by the implicit threat that the next buyer is already waiting. If hesitation persists, the intermediary confides that several named institutions are interested and that the allocation will go elsewhere.
Inexperienced investors lose their discipline under that pressure. The emotion has a name in the industry and everyone uses it: fear of missing out.
The more important technique is sequencing. Assemblers bring a top-tier dollar fund into the early rounds and then work to keep that fund following on. Each subsequent cheque from a recognised name becomes the argument for the next investor. Assemblers also coach founders on winning orders and, according to the same accounts, on how to manufacture revenue.
The valuation path is close to standardised. A first round around USD 14 million equivalent. A second that roughly doubles to RMB 1.5 billion. A third between RMB 2.5 and 4 billion. Projects with a strong story have oversubscribed and jumped straight past RMB 5 billion.
Nobody in that chain needs the robot to work. The fund earns a paper return from the next round’s markup, the founder is wealthy on paper, and the adviser is paid. The machine only requires one thing: that the company eventually lists.
Where the bottom of the ladder went
Some investors were already slowing in the first half. They knew the market was behaving irrationally and that a peak would be followed by a decline. What they could not predict was when and why the break would come.
It came from the exit. When the listing route narrows, new money stops arriving, and the valuation chain loses its foundation. The paper gains that were meant to be converted at IPO are no longer convertible, and the incentive to keep funding the next round disappears with them.
The cycle is unlikely to end quickly. One investor’s estimate, offered without much confidence, is that the industry needs to shake out until the second half of 2027, possibly the first half of 2028, before it settles. In the meantime the useful work continues in the companies that were built to sell something. The rest are waiting to be repriced, and the market has begun to do exactly that.
Editor’s note: This is an adapted translation of the original Sohu Tech report. It has been trimmed and restructured for readability for an international business audience. The full original (in Chinese) is at https://www.sohu.com/a/1094202597_114778.