95% of Humanoids Are Made in China. Who’s Actually Buying Them?
Shipment headlines are easy. The harder question is whether factories are buying these robots to do work, or whether research, exhibitions, and data collection are still carrying the volume.
TL;DR: Chinese vendors shipped the overwhelming majority of humanoid robots in the first half of 2026. That is not the same thing as factories buying them to replace labor. Two trackers, IDC and Counterpoint, both show research, exhibitions, and data-collection demand still doing a lot of the work. Manufacturing and warehousing are in the numbers. They are not yet most of the numbers.
When I look at a market, I want to know who the customer is. “Units shipped” is a starting point. It doesn’t tell you whether the buyer is a plant manager with a payback model, a lab, a show floor, or a government program collecting training data.
I also want to be honest about the data. I’m looking at two trackers, not at a customs database I audited myself. They don’t perfectly agree. I’ll name each one rather than flatten them into a single story.
What IDC is counting
In a September 30 piece, John Koetsier reported IDC’s figures for the first half of 2026. Almost 25,000 humanoid robots shipped, up 432.1% from a year earlier. Revenue was more than $740 million, up 322.7% (Koetsier / Yahoo).
Chinese vendors accounted for more than 95% of those shipments. Chinese customers accounted for about 78%, or more than 19,000 units, up 426.3% (Koetsier / Yahoo).
On the vendor side, as Koetsier reports IDC, Agibot shipped more than 8,600 units, or about 35% of the half. Unitree was around 5,000 (Koetsier / Yahoo).
The more interesting split is the buyer. Research and education, performances and exhibitions, and government data centers together made up 69% of first-half shipments, down from 83.8% for all of 2025 (Koetsier / Yahoo). That is a real shift toward other uses. It is also still most of the volume.
Implied average selling prices, from the same figures, came down to about $29,600 from around $37,000 a year earlier. IDC’s 2030 forecast, as reported there, is more than 750,000 units a year (Koetsier / Yahoo).
What Counterpoint is counting
Counterpoint, in an August 20 note, put first-half shipments at more than 22,000, up nearly 300% year on year. It had AGIBOT at about 9,700 units, more than 43% of the total, and Unitree at more than 7,000 units, or 31% (Counterpoint).
Those Unitree numbers are not the same as IDC’s. IDC, as reported by Koetsier, is around 5,000. Counterpoint is more than 7,000. AGIBOT’s share is also higher in Counterpoint than in the IDC write-up. I’m not going to pretend I can reconcile two trackers from the outside. The direction is the same: a small number of Chinese vendors have most of the volume.
On use cases, Counterpoint says entertainment and performance, plus data production and research, still accounted for more than 60% of shipments. Manufacturing was 13%. Warehousing was 5%. Its full-year 2026 forecast is more than 50,000 units (Counterpoint).
That’s the line I care about. Thirteen percent in manufacturing and 5% in warehousing is not nothing. It is also not “the factory of the future has arrived.” A lot of what is shipping is still being used to learn, to demonstrate, or to collect data.
So what do I take from that?
My read: the “China is shipping almost all of them” headline is true, and it is incomplete. China is also buying most of them. The open question is what those buyers are doing with the robots.
If 69% (IDC) or more than 60% (Counterpoint) of first-half volume is still in research, education, exhibitions, performances, or data collection, then unit growth can run ahead of unit economics for a long time. Prices can fall, as the implied ASP drop suggests, without a warehouse manager having a payback model that works.
The mix is moving. IDC’s 69% versus 83.8% a year earlier is the evidence for that. I would watch the manufacturing and warehousing shares in the next couple of reports more closely than I would watch the headline shipment number.
So what for operators and investors
My thought is this:
- If you’re buying, don’t let a shipment headline close the case. Ask what job the robot is already doing at a comparable site, for how many hours, and at what intervention rate. Research volume and show-floor volume don’t answer that.
- If you’re investing, look at the customer mix, not just the units. A company selling into government data-collection programs or universities is in a different business from one selling into a plant with a labor-substitution budget. Both can be real. They are not the same multiple.
- If you’re trying to read China from the US, the 95% vendor share and the 78% customer share are the same story: this market is still mostly inside China. Export and Western-factory demand will show up, if they show up, as a change in that mix.
What would change my mind: a later tracker report where manufacturing and warehousing are clearly the majority of shipments, with named commercial deployments behind them. Or, the other way, if the 69% research-and-display bucket stops shrinking.
Open question for readers: if you already bought a humanoid, what job is it actually doing — and would you buy the second one?
Disclosure: I hold no positions in any company mentioned. No company reviewed this piece. My own consumer brands are manufactured in China. This is a reading of published tracker figures, not investment advice.
Sources & notes
- John Koetsier, “Humanoid Robot Shipments Jump 432%, China Has 95% Market Share,” originally published on Forbes.com, Sep 30, 2026, via Yahoo Finance, reporting IDC figures: https://finance.yahoo.com/technology/ai/articles/humanoid-robot-shipments-jump-432-025835160.html
- Counterpoint Research, “Global Humanoid Robot Shipments Soar Nearly 300% YoY in H1 2026,” Aug 20, 2026: https://counterpointresearch.com/en/insights/global-humanoid-robot-shipments-soar-nearly-300-percent-yoy-in-h1-2026
- The two trackers disagree on Unitree’s first-half volume (IDC, as reported by Koetsier, around 5,000; Counterpoint, more than 7,000) and on AGIBOT’s share. This piece names each source and does not reconcile them.