Morgan Stanley raised its forecast for China humanoid robot shipments in 2026 from 28,000 units to 50,000 units on Tuesday, which is the second upward revision the bank has made this year. The original January forecast was 14,000. So in six months the curve has gone from “modest commercial uptake” to “second-fastest robotics market doubling in modern history” without anyone running out of words first. The bank also pulled its 2030 number up: $15 billion market size, 446,000 humanoids shipping annually.
The proximate trigger is volume. Most of it is the State Grid Corporation of China procurement order from earlier this year, worth roughly 6.8 billion yuan, or about $1 billion. The unit mix is interesting and not what you would assume from the headline. State Grid bought 500 humanoid robots, 3,000 dual-arm robots, and 5,000 quadruped robots. The humanoids are the prestige line item. The dual-arms and the quadrupeds are the line items that actually get the inspection-and-maintenance work done across the country’s power grid. Read in that light, the order is less “China is buying humanoids at scale” and more “China is buying a complete robotic workforce, and the humanoids are the part that gets photographed.”
Morgan Stanley flags three drivers. One, commercial validation, by which they mean enterprise buyers are signing real procurement contracts instead of running pilot programs that quietly expire. Two, policy support, by which they mean Beijing has issued the kind of provincial-level humanoid-deployment quotas that fund the buyers who sign the contracts. Three, supply chain, by which they mean Unitree, Fourier, UBTech, Agibot, and the rest of the Chinese humanoid OEM cohort can now actually build the units they sell, at unit economics that approach the price points on the brochures.
The comparison nobody wants to print but everyone is doing in their head goes like this. Figure says it is at one Figure 03 per hour at BotQ, on the way to 12,000 units in line one’s first year. Tesla still says Optimus is a million-units-per-year Texas program at some unspecified future date. Boston Dynamics is shipping the first electric Atlas units into Hyundai and DeepMind. Agility Digit is in Toyota Canada at low single digits. Add it up and the entire non-Chinese humanoid cohort is probably under 25,000 units this year, optimistically. China alone is projected at twice that, with most of the volume going to domestic industrial buyers who do not need to win a procurement bake-off against a US incumbent because there is no US incumbent at their door.
The robotics analogue to the 2010s solar panel curve is starting to rhyme. The thing that matters in robotics is not who builds the first impressive prototype. It is who builds the boring tenth-generation unit at a price the second tier of industrial buyers can afford. Morgan Stanley’s note, read carefully, is the first time a major bank has put that thesis in print with a number attached.