--- title: "Humanoid rentals in China fell from 15,000 to 3,000 yuan a day — and orders still climbed" date: 2026-11-02 category: Humanoids & Robotics site: NeuroAI canonical: https://neuroai.site/a/na-robot-rental-raas language: en --- # Humanoid rentals in China fell from 15,000 to 3,000 yuan a day — and orders still climbed > A humanoid that rented for 15,000 yuan a day in early 2025 now goes for about 3,000. The price collapse did not kill the business — it forced it to become a service industry. A shopping mall in Hangzhou books a humanoid (人形机器人) for a Saturday. It arrives in a van with one engineer, stands at the entrance for six hours, waves, performs a short routine, and leaves. The invoice is around 3,000 yuan (≈ US$420). Eighteen months earlier, the same six hours would have cost about 15,000 yuan (≈ US$2,100). By the arithmetic of most industries, that is a collapse. What is strange is that the number of people willing to book one went up, not down. ## Key takeaways - Mainstream humanoid (人形机器人) day rates in China have fallen from five figures in early 2025 to roughly 2,000–3,500 yuan (≈ US$280–490), according to checks by *Securities Times* and *China Economic Net*. - Robot dogs rent far lower still — quoted at a few hundred yuan a day (≈ US$30–70), and in some cases priced per two-hour appearance rather than per day. - Rental platforms consolidated fast. QingTianZu (擎天租), founded in December 2025, reported more than 3,000 dispatchable robots and coverage of over 100 cities, integrating more than 600 service providers. - Thin hardware margins pushed the business toward content and services: one platform cut motion-development time from three days to about one hour. - The unresolved question is whether rentals graduate from "hiring a spectacle" to "hiring labour". ## How the price fell this far The mechanism is not mysterious. In early 2025 demand was concentrated and supply was tiny, so rates were set by scarcity rather than by cost. Day rates of 10,000 to 30,000 yuan (≈ US$1,400–4,200) pulled in capital, and capital pulled in inventory. Once inventory arrived, the price had nowhere to go. *China Economic Net* reported Unitree's G1 at 6,000–8,000 yuan (≈ US$845–1,130) a day in the second half of 2025, 4,000–5,000 yuan (≈ US$565–705) in the first half of 2026, and about 3,000 yuan (≈ US$420) at the time of writing. *Securities Times* put the same model at 2,500–3,500 yuan (≈ US$350–490) a day, the dancing U2 at around 4,000 yuan (≈ US$565), and robot dogs at a few hundred yuan. The drop was not a discount. It was a repricing of what customers were actually buying — attention, not work. ## Why the volume went up anyway Three things happened at once. **The buyer changed.** Early bookings were single units for novelty. Operators report orders now arriving in tens of machines for shopping centres, scenic areas and brand roadshows, where the robot's job is to generate footfall and social-media reposts. **The unit economics moved.** "Bare rental" margins are now thin enough that several operators describe hardware hire as customer acquisition rather than profit. Money is made on what surrounds the machine: custom choreography, scripted interaction, on-site engineering, logistics and insurance. **Platforms appeared.** QingTianZu (擎天租) was set up in December 2025 with backing from AgiBot (智元机器人) and others, and by early 2026 said it could dispatch more than 3,000 robots across over 100 Chinese cities through more than 600 service providers, with models from seven embodied-intelligence makers on the platform. It describes a "shared rental + platform dispatch" model, with regional pre-positioned warehouses so a partner who wins a large order does not have to own the inventory. It also reported cutting motion-development time from three days to roughly an hour, and ran a 999-yuan (≈ US$140) experience package that accounted for about 15% of platform orders. State-owned entrants followed. China Mobile's embodied-intelligence innovation centre launched its own rental platform, arguing that the main obstacle for customers is not price but the question "if it breaks, who comes?" — a service-network problem more than a hardware one. ## The three stages, and where it stalls The most useful framing of the market comes from Zheng Gang of Zhejiang University's management school, who describes rentals moving through three phases: renting presence, renting labour, and renting outcomes. Most of the market is still in the first phase. Renting presence means a robot at a mall entrance, an annual gala, or a store opening. It is real revenue, but it is seasonal: operators describe full order books around the Spring Festival and National Day, and idle warehouses in March, April, July and August. Renting labour — welcoming, guiding, patrolling, carrying, delivering plates — is where the constraint stops being intelligence and becomes reliability. A robot that fails once in ten runs is fine at a mall and unacceptable on a shift. Renting outcomes, billed per item, per patrol area or per visitor served, is the version that would make this a real industry. Nothing in the current price list suggests it has arrived at scale. The industry's own leaders are cautious about how far prices can fall. QingTianZu's chief executive Li Yiyan has argued that a healthy range sits around 5,000 yuan (≈ US$705) a day once development, fulfilment and asset-scheduling costs are accounted for, and that a race to the bottom is limited by those costs. According to Li An, chief scientist at BrainNet (脑机网), the rental boom is best read as a product-market-fit test running ahead of the technology: it tells you which jobs people want done long before a robot is good enough to do them reliably. ## Honest limitations Price figures in this piece come from reporters' own checks of platforms at specific moments and vary by source, city, season and whether an operator is bundled with the machine. They are indicative, not a published index. Platform statistics — robots dispatchable, cities covered, service partners, order counts — are self-reported by the companies and were not independently audited. The attrition number is the weakest data point. One report said more than 90% of rental counters in Shenzhen's Huaqiangbei electronics market had withdrawn in the second half of 2025; that claim comes from a single outlet and no independent count was found, so treat it as a signal of churn rather than a measurement. Not covered: actual utilisation rates per machine, maintenance cost per rental day, depreciation assumptions, and whether any major platform is profitable. Those are the numbers that decide whether this is a durable channel or a pre-purchase marketing subsidy. ## What readers can do now - **If you are booking a robot for an event**, price the whole day, not the machine. Ask separately for the operator, transport, insurance and any custom choreography, and confirm in writing what happens if the unit fails on site. - **If you run a venue and want more than a photo opportunity**, pilot against a measurable job — a fixed patrol route, a defined delivery task — for at least two weeks. Reliability over two weeks tells you more than any demo. - **If you are considering entering the business**, do the arithmetic on the quiet months. The failure mode reported by small operators is not low prices; it is a warehouse full of idle machines in April. --- Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-robot-rental-raas Free to quote with attribution and a link to the original.