The story phase of Chinese embodied AI ended some time in the second half of 2026. What replaced it is a much narrower examination, and the questions are delivery counts, cash collected and whether a unit can pay for itself.
Too much money, too fast
IT Juzi counts more than 600 financing deals in the sector between the start of 2026 and 20 October, worth over RMB 160 billion. That produced the largest club of ten-billion-yuan valuations the industry has ever had: by October, at least 32 Chinese embodied AI companies carried a valuation or market capitalisation above RMB 10 billion.
The crack appeared in second-hand shares before it appeared in headline numbers. Securities Times reported on 29 October that old-share transfers in the previous two months were priced at 65 to 80 per cent of the last round, including at the largest companies. Early shareholders are leaving at a discount and primary and secondary valuations have inverted.
The two largest names set the tone
Unitree listed on the STAR Market in August. Its market capitalisation touched RMB 444.9 billion on the opening session and nearly halved within days, erasing more than RMB 200 billion. AgiBot began preparing a Hong Kong listing in July, then restarted a funding round in October at a locked valuation of RMB 50 billion. Under Hong Kong rules that round pushes its listing into next year and closes the near-term window.
When the leaders slow their listing timetable and stop ratcheting valuations upward, confidence in the tail follows. The sector runs on ten- to twenty-year cycles, its technical routes are not settled, and it burns cash faster than almost any other hardware category. If listing and follow-on funding both narrow, the only remaining test is whether a company can fund itself.
Most currently cannot. The overwhelming majority of players are still at the prototype and small-batch stage, without stable orders, repeat revenue or a closed commercial loop.
Three thresholds, in order
The first is the move from demonstration to industrial product. Tolerance in a lab and tolerance on a line are two different worlds. The second is the move from showcase projects to repeatable volume orders in factories, logistics, retail and entertainment. The third is unit economics: cutting hardware cost until gross margin turns positive, instead of losing money on every machine shipped.
What buyers now ask for has changed with them. Investors once read technical routes, founder pedigrees and patents. Industrial capital, which now dominates the cap table alongside internet platforms and state-guided funds, reads three things: the delivery ledger, cash received and the number of live sites.
Business models are widening in response. Beyond selling whole machines, companies are testing leasing, software subscription and data services. That diversification is itself a sign of maturity, because it forces an answer to who pays and why.
The ten thousand unit line
Competition has shifted from algorithm precision and motion smoothness to supply chain control, hardware cost, production efficiency and repeat purchase. Once robots enter volume manufacturing, cost and delivery decide survival, and that is manufacturing logic rather than laboratory logic.
The valuation yardstick is being rewritten with it. Shipment counts alone no longer suffice. Hardware revenue, manufacturing cost, research spending and gross margin have to be read together. AgiBot’s shift from a deployment model to Worken is presented as a commercial upgrade, but it is really a change of valuation paradigm: how much labour a machine replaces, how many hours it stays on shift, and how quickly a customer recovers the purchase price.
The industry has settled on a single blunt marker. Ten thousand units on duty is the only proof that a ten-billion-yuan valuation sits on something solid. Everything below that line is presentation.
Editor’s note: This is an adapted translation of the original OFweek 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/2026-11/ART-8321205-12001-30710685.html.