On August 10, Smart Analytics Global posted its first-half humanoid robot shipment count: roughly 19,100 units worldwide, up 272% from 5,100 a year earlier. AgiBot shipped about 8,400 of them, took 44% global share, and moved past Unitree into first place.
The number that traveled was 97%. That is the Chinese vendor share of global shipments, and every wire desk led with it.
It is the wrong number to lead with. It describes where the robots are built. The sentence that describes what they are for sits three sections down: industrial and commercial applications accounted for more than 70% of first-half shipments, up from roughly 50% a year earlier.
The count
AgiBot’s 8,400 units are up 562% year over year, lifting its share from 25% in H1 2025 to 44%. SAG credits the spread of the portfolio: full-size bipedal A-series, half-size bipedal X-series, wheeled G-series, with industrial and commercial deployment as the stated focus. The company has product on the ground in the UK and Germany.
For scale: AgiBot shipped more units in six months than Unitree shipped in all of 2025, when Unitree led the market with 5,500.
Unitree delivered about 5,900 units, up 170%, for 31% share and second place. Nothing about that is a stumble. A 170% growth rate would lead most industries. Unitree got out-run. SAG adds the detail that explains the ranking: half-size bipeds carry the portfolio, cumulative G1 volume reached roughly 11,000 units by mid-2026, and research, education, and performance deployments still make up a meaningful slice of where those machines end up.
Below the top two the market thins fast. Galbot shipped about 900 units for third, working wheeled designs into structured industrial and commercial settings. UBTECH took fourth at about 700, carried by the full-size bipedal Walker line. Leju rounded out the top five at about 600. AgiBot and Unitree together account for roughly 75% of global shipments; everyone else splits the remaining quarter.
SAG’s full-year call is close to 60,000 units in 2026 on about $1.6B of industry revenue, rising toward $3B in 2027, and 500,000 units by 2030.
The ranking changed because the buyers changed
Put the two leaders side by side and the interesting gap is not capability. It is customer mix.
We covered Morgan Stanley’s third upgrade to its China humanoid forecast on June 27, when the bank took 2026 China shipments to 50,000 units. That piece made one call: the number tracks production, not proven demand. SAG’s 60,000 global figure points the same direction on a different basis. What SAG adds is the destination.
A year ago, about half of global humanoid shipments went to labs, universities, and stage events. Those orders close quickly. They do not require a payback case. A machine that stands in a lobby and dances has satisfied the purchase order. In the first half of 2026 that half fell to under 30%. The other 70%-plus went into manufacturing, logistics and warehousing, and structured commercial floors, where acceptance is measured in cycle time, yield, and downtime.
AgiBot’s share came out of that shift. Unitree’s share was diluted by it.
It also explains a divergence worth naming. Unitree priced its Shanghai listing at 150.8 yuan on August 6 into record retail subscription demand, while its shipment share was heading the other way. Reading the prospectus on August 1, we flagged the risk disclosure: units can move fast, but if research and demonstration revenue does not shrink as a proportion, the valuation has to rest on something else.
Small volume, but no longer toy orders
19,100 units is nothing against any labor pool. Global manufacturing employment runs to the hundreds of millions. Two thousand robots per month does not register. That is not the exposure.
Exposure is two things: what the machines are asked to do, and which country they land in.
SAG answered the first with 70%. It answered the second too: China represents more than 85% of global demand. So the first measurable humanoid displacement will show up in Chinese 3C assembly lines, warehousing and logistics, and structured commercial service work — greeting, wayfinding, in-store product walkthroughs, the jobs with a fixed floor plan, a repeated motion, and a controlled environment. Western white-collar work is a long way down the list.
The second signal to track is spillover. AgiBot has delivered into the UK and Germany, and targets overseas sales above 30% of its total in 2026, with 50% named as the follow-on. Its European representative described the buyer plainly: automotive and warehousing manufacturers plus universities, pulled in by rising labor costs. That sentence is worth more than most industry forecasts.
Note who else is already there. Unitree took the G1 into Europe in July, carrying a Pentagon advisory with it. Both leaders are now pushing the same continent, and they are pushing it for the same stated reason. European warehouse and logistics handling roles are the next place this gets tested with numbers.
The US route is blocked at customs. The FCC barred new Chinese humanoids and robot dogs on July 28, and SAG lists geopolitics as a standalone risk to its own forecast. On August 10 we covered RoboStore opening a Long Island plant thirteen days after the ban. The restriction stops imports. It does not stop the same machines arriving by another route. US job exposure will not show up in trade data. It will show up in domestic assembly and systems-integration capacity.
The ceiling holds, and the ceiling is moving
SAG writes the counterargument into its own report, which is more than most vendor-adjacent research does. Hardware is scaling faster than the technology under it. Real-world training data, model capability, reliability, manipulation accuracy, safety, and cost each still cap how far deployment can go. Moving from machines that can run a demo to machines that can work a full shift is not finished work.
That lines up with our July 24 piece on what China’s factory humanoids can actually complete: 97% success separating fabric, and still no shirt finished without a person closing it out. This morning’s piece on three automakers assigning their humanoids the same first job is the same fact from the Western side. BMW, Hyundai, and Mercedes each put their most expensive machine on kitting, because kitting is where the machine can still deliver.
So the read is not that robots are about to take the jobs. The read is that in twelve months this industry moved its center of gravity off the demo stage and onto the production line, while the set of tasks it can do on that line stayed narrow. Narrow, and widening. Small, and pointed hard at industry.
For an individual, the test compresses to one question: does your work happen in a fixed location, on a fixed process, with predictable materials? If it does, SAG’s 70% is aimed at your column. If it does not, the 500,000-unit 2030 forecast does not reach you yet.
