In short
- Morgan Stanley revised its 2026 China humanoid forecast 3.5x in six months (14k → 28k → 50k).
- A single $1B State Grid order and 10,000-unit/year factories are pulling deployment forward faster than models assume.
- VC funding and component makers are being re-priced in weeks, not quarters (e.g. Leaderdrive +73%).
- ~85% of humanoid deployments happen in China, aided by 10–14 day prototype cycles vs 12 weeks in the West.
- Reliability and uptime data are still missing — scale-up isn't proven usefulness.
In January, Morgan Stanley expected 14,000 humanoid robots to ship in China in 2026. By spring, the bank had raised that to 28,000. On June 24, it doubled the figure again, to 50,000 (CNBC, June 24, 2026).
That's one bank revising the same 12-month window upward by 3.5x in six months. Forecasts revise slowly when a technology arrives on schedule. They double when deployment gets pulled forward faster than anyone budgeted for.
The useful question is what's underneath the revisions. Because a bank doesn't double a shipment forecast on vibes.
What actually moved
Morgan Stanley's own report names three drivers behind the June upgrade (BigGo Finance, June 2026): accelerating commercial validation, national policy support, and aggressive supply chain capacity expansion.
The first driver has a hard number attached: a 6.8 billion yuan order (roughly $1 billion) from State Grid, China's electricity utility. That's a single customer placing a billion-dollar humanoid order. Analyst models built around gradual pilot programs and cautious enterprise adoption simply don't have a slot for that kind of purchase. When it lands, the model breaks upward.
The supply side is moving on the same curve. In March, a production line in Guangdong went into operation with a stated annual capacity of over 10,000 humanoid units, reportedly China's first fully automated humanoid production facility. Build time per robot: about 30 minutes. Five more, larger sites are planned. TrendForce projects Chinese vendors will drive annual output growth of up to 94% in 2026, with Unitree and AgiBot together accounting for nearly 80% of shipments (TrendForce, April 9, 2026).
And the deployments are already visible in ordinary places. Beijing-based RobotEra is putting a thousand humanoid sorters into more than ten logistics centers. When I visited a humanoid robotics company in Shanghai this June, the striking thing wasn't the demos. It was the shipping volume: 10,000 robots out the door last year, from a company three years old.
A stable forecast means the model has caught reality. A forecast that keeps doubling means reality is outrunning the model — and every plan built on the last published number is already stale.
The money is running the same pattern
VC funding into humanoids follows the same shape as the forecast: lumpy, accelerating, hard to model. The sector's 44 funded companies have collectively raised $6.89 billion, with Figure alone at $1.75 billion (Tracxn, May 12, 2026). Zoom into the monthly data and the curve looks nothing like a smooth line — four of the last twelve months recorded zero qualifying deals, while September 2025 alone pulled in $1.14 billion (New Market Pitch, May 2026).
Shenzhen-based X Square Robot raised around $100 million in a round led by Alibaba Cloud — its eighth round of financing in under two years of existence (CNBC, September 8, 2025). Eight rounds in two years is a company being re-priced faster than a normal fundraising cadence allows.
The re-pricing reaches down into components too. When Morgan Stanley published its June upgrade, it raised the 12-month price target on Leaderdrive, a Shanghai-listed maker of harmonic drives, from 269 to 464 yuan — a roughly 73% jump on a single supplier, in one report (AI Weekly, June 2026). Component makers get revalued in weeks when the shipment assumption doubles.
Why China's number moves faster
Part of this is structural. Roughly 85% of recent humanoid deployments are happening in China, against about 13% in the US. China dominates the components for rotary actuators, one of the most critical parts in a humanoid (McKinsey, May 2026). And the iteration loop is simply shorter: a prototype that takes 12 weeks to produce in the US or Germany turns around in 10–14 days in Shenzhen, at a fraction of the cost (SVRC Robotics Center, April 17, 2026).
When your hardware iteration cycle is measured in days and your forecasting cycle is measured in quarters, the forecast will always lag reality. Morgan Stanley's revisions are the bank catching up to what already happened on the ground.
The honest caveat
A forecast revised twice in six months signals uncertainty as much as momentum. Rapid scale-up and proven reliability are different things, and the public reporting stays quiet on uptime, failure rates, or how supervised these deployments still are. Fifty thousand robots shipped tells you nothing about how many of them are doing useful, unsupervised work in month six. That data will surface eventually. It hasn't yet.
Watch the revisions
Here's the mental model worth keeping: when a serious institution revises the same forecast repeatedly in the same direction, the revisions matter more than the figure. A stable forecast means the model has caught reality. A forecast that keeps doubling means reality is outrunning the model — and every plan built on the last published number is already stale.
Morgan Stanley now projects 446,000 units and a $15 billion market by 2030. Given how the 2026 number has behaved, treat that as a floor with a wide error bar, and check back in six months.
The 50,000 will be old by the time you've built a strategy around it. The direction and speed of the revisions won't be.
Ruben Horbach
Co-founder · Back From the Future
Ruben researches how organisations adopt AI meaningfully — not as technology, but as a change in work and people. He builds the agent infrastructure behind BFF and speaks about the near future of work.
