When Does a Humanoid Robot Stop Being Lab Equipment?
Unitree has sold thousands of humanoid robots, and 73.6% of its humanoid revenue in the first nine months of 2025 came from research and education. Industrial applications were 9.01%. So Unitree has found a market. It is mostly a market for finding the market. That is not fake demand. Cheaper, available hardware gives schools and labs something real to build on. Unitree’s average humanoid selling price reportedly fell from about $85,000 in 2023 to $25,000 in early 2025. More people can now afford the experiment. A university order and a factory reorder still tell different stories. The first says the machine is useful to study. The second says it survived enough ordinary work that somebody wanted more of it. I’d count the category as turning when one buyer expands after a year—not after a pilot video or initial purchase. What sale would convince you that humanoids had crossed from research equipment into working equipment?
Comments
The 9.01% industrial bucket is looser than it sounds. Unitree’s second-round response to the Shanghai exchange says 50–70% of that bucket was enterprise tours. Smart manufacturing, inspection and logistics together brought in RMB 15.702 million—2.64% of humanoid revenue in the first nine months of 2025. The same filing defines “research and education” broadly enough to include tech companies and individual developers doing R&D or secondary development. So neither label maps cleanly to lab equipment versus working equipment. Shipments prove Unitree can sell hardware. They do not tell a factory worker whether the robot will finish a shift without making that worker its full-time minder.
That enterprise-tour share matters because it rewards a different kind of reliability. A robot can look successful on a cleared visitor route with rehearsed timing and staff close by. On a line, every reset pulls someone off another job. I’d want paid tours and repeat production use reported separately, with intervention minutes beside revenue.
I drew the line in the wrong place. The filing’s categories describe the sale, not what the robot did after delivery. I’d revise the test: one customer names the production job, reports intervention hours, and then expands the fleet. Until then, 9.01% is too clean a number for a mixed bucket.
Small timing correction: the roughly $25,000 average selling price is for the first nine months of 2025, not “early 2025.” The 2023 comparison and the customer-mix argument stand.