The party that repriced embodied intelligence
Recent public reports say the Hong Kong Stock Exchange will consult next year on revising its Chapter 18C regime for specialist technology companies, including a review of the market-cap threshold. That will affect the dozens of embodied-intelligence companies already in the queue. The backdrop is August 2026, when Unitree, China’s first humanoid-robot maker on the A-share market, listed on the STAR Market. Its opening price shot toward 100 yuan and its market value briefly passed 400 billion yuan, then within a handful of trading sessions the shares slid and the value nearly halved, with more than 200 billion yuan wiped out. Around the same time, a few embodied-intelligence component suppliers that listed in Hong Kong also broke their issue price on debut.
Overnight, the sector that capital had placed on a pedestal was pushed under the spotlight. The old assumption was that, just as the chip sector enjoyed a wave of listings, humanoid robots would follow the same script: a leader reaches the market, then a batch of whole-robot makers queue up, win approval and keep banking the valuation premium. Reality did not follow that script.
After Unitree, patient capital instead of feverish capital
What surprised the market was that no swarm of humanoid OEMs rushed to list after Unitree. The ones that actually rang the bell were not whole-robot makers but supply-chain firms. RoboTrek, SEER and LiDAR maker Leishen all listed in Hong Kong this year; most recently, Mech-Mind landed on the HKEX on 1 September, focused on robot 3D vision and dexterous-hand components, while Benmo makes robot power modules. Both can be called pick-and-shovel sellers for the embodied era. Wind data shows more than 50 embodied-intelligence companies now queuing at the HKEX, but given the regulatory mood, few whole-robot bodies will clear the bar soon, while upstream component suppliers have the better chance.
The reversal means the sector’s capital dividend is being redistributed. Whole-robot makers are still at the prototype and small-batch test stage, while component firms have already shipped commercially, selling to the very OEMs that buy the bodies. The anchor for pricing is no longer the humanoid-robot concept alone. Capital is no longer willing to pay sky-high premiums for distant promises, and whether a company has real orders and stable customers is becoming a heavier weight in its valuation.
Doubts about real revenue
Unitree’s sharp pullback was not pure sentiment. Comments by Mech-Mind’s founder questioning whether some leading humanoid firms inflate revenue stirred market speculation. One is a whole-robot maker, the other a component supplier, and both share a pattern of a high open followed by a low close. Unitree retraced more than 50 per cent from its peak; Mech-Mind’s public offer was about 3,835 times oversubscribed, a record for the Hong Kong robot track, yet it broke issue price on day one and kept sliding.
The two cases also expose a shared problem hidden behind the glamorous fundraising. Prospectuses and financial reports will make the market ask where embodied-intelligence revenue really comes from: genuine commercial customers, or research procurement and related-party orders, or even fabrication. According to the Humanoid Robot Scenario Application Alliance, in the first half of 2026 domestic public winning bids for humanoid and embodied robots totalled 218 projects worth 1.723 billion yuan. Of that, education and research institutions took 55.5 per cent, government and state-owned platforms 20.6 per cent, and industrial orders just 21.1 per cent. Most industrial orders were prototype tests, showroom demos and training pilots, not normal production-line deployments, so real commercialisation remains weak.
If much revenue comes from universities and research institutes, those orders carry policy and project attributes and rarely repeat. Once a research cycle ends, the sustainability of revenue growth is in question. Secondary-market investors are waking up to the fact that hundreds of billions of yuan in market value correspond not to a mass-industrialised business but largely to prototype and research sales.
The regulators tighten
Several financial-media outlets report that HKEX scrutiny of humanoid-robot IPOs has hardened. Beyond Hong Kong, when the Shanghai exchange revised its STAR Market listing rules in June it named robotics and embodied intelligence as key supported directions while also demanding stricter gatekeeping on admissions and a firm bottom line on financial authenticity, to prevent sick filings and a herd rush. In this environment, a company’s own recurring revenue and technical moat may become the key pass.
That is hard, though. Among the embodied firms in the review pipeline, Yunji Deepak turned a profit in 2025 while Dobot and Leju remained in the red, suggesting the others may not look cheerful either. A technically solid but unprofitable firm can consider Hong Kong; a firm with no real business logic will not fix it by changing market.
Humanoids are not short of stories, but short of market moats
At least a hundred domestic whole-robot makers exist. As supply-chain technology matures, many can build a walking, demoing prototype, yet few can build a moat for scaled commercialisation. The core components, reducers, servo motors and controllers, come from a highly converged supply base. Many prototypes source their key parts entirely externally, with limited self-developed moat; differences show up mostly in appearance and demo moves, not in a real generational gap on the factory floor. At volume, procurement cost and yield become everyone’s shared test.
Unitree’s case is now the reference for those behind it. Riding sector sentiment to repeat a first-day spike is hard; investors have been educated and their patience for concept stories has clearly dropped. For the queue, listing at a more reasonable valuation and proving the business through steady operation is the realistic path this round.
An oligopoly has already formed
SAG data shows the oligopoly of the humanoid track in the first half of 2026: AgiBot shipped 8,400 units, taking 44 per cent of the global market, while Unitree shipped 5,900 units for a 31 per cent share. The two leaders together hold more than 70 per cent of the world, and concentration has risen sharply. The tail players are far smaller, each shipping under 1,000 units in the half, and the gap between leaders and the middle is deep and hard to cross.
The era of valuing humanoids by story and breaking out through premium may be over. But the capital’s recalibration is only one slice of the sector’s development. Unitree’s halved value does not announce the end of the humanoid industry; it pushes the track from the story-telling stage to the commercial-validation stage. At the Inclusion Outer Bund conference on 11 September, several experts argued that over the next year or two the fastest-growing deployment scenes will be industrial, logistics and rehabilitation, where environments are enclosed, task chains short and exceptions few, before robots move into the broader consumer market.
Secondary markets remind builders that between a tech demo and a sustainable business lies a long road of industrialisation. More than 50 firms queue for HKEX review, many of them whole-robot bodies with teams, prototypes and patents but a market not yet big enough for all of them. Capital will keep selecting. Firms with recurring commercial orders and shrinking losses can use the listing to fund R&D; those reliant on project orders with no real scene will find the road harder.
Editor’s note: This is an adapted translation of the original OFweek Robotics report. It has been trimmed and restructured for readability for an international business audience. The full original (in Chinese) is at https://robot.ofweek.com/2026-09/ART-898890-12003-30705235.html.