Product1 distinct publisher2 min readPublished
A Shanghai R&D park with more than 100 robotics firms is the part of this story Western buyers will negotiate with, not the drunken-boxing routine on the main stage.
The Product Desk · Product desk
Compiled by The Product DeskSomething wrong?How this is made
The residual is where the procurement problem lives. If nearly 90% of the 13,000-plus two-armed, two-legged robots delivered worldwide last year were built in China [1], then every other manufacturer on the planet delivered something in the region of 1,300 machines between them [9], against roughly 11,700 out of China [10]. That is about nine to one [11]. A buyer who wants humanoids with a non-Chinese bill of materials is bidding into a pool an order of magnitude thinner than the one his competitor is using, and thin pools set their own prices and lead times.
Which is why the interesting thing at that park on Shanghai's outskirts is the tenant list, not the stage. The same site's firms research, develop, manufacture and market robotic equipment [4], and much of that equipment is not humanoid at all: machines that haul heavy loads, machines that inspect sewage pipes [5]. Proximity of that kind is a lead-time asset. A team iterating on a coffee-making humanoid, as DexForce was exhibiting [6], sits down the corridor from people who already ship industrial hardware in volume.
MIT Technology Review's correspondent is blunt that the day itself was mostly for show [7] and describes the wider effort as a PR campaign [8], and the same piece catalogues why humanoids have stalled elsewhere: heft and balance make them hard to make safe, prices are high, batteries are short [2]. Plenty of roboticists doubt the human shape is the right one at all [12]. Those objections are about the product. The delivery share is about the factory. You can be entirely right that bipedal is the wrong form and still discover that the assembly capacity for whatever the right form turns out to be is concentrated somewhere you have no leverage.
What the source does not tell us is who took delivery of the 13,000, at what price, or what fraction went into productive work rather than a lobby. Given that humanoids are performing in shopping malls and at tourist attractions in several Chinese cities, and running half-marathons in Beijing [3], some of that volume is entertainment. Worth pricing in rather than dismissing: novelty demand still keeps a line running, and a line that is already running is the thing a Western buyer will eventually be negotiating over.
Ranked by verification strength, evidence, and original report placement.
Nearly 90% of the more than 13,000 two-armed, two-legged robots delivered globally last year were made in China.
Outside China, humanoid progress has been slow, marred by safety challenges due to the robots' heft and their need to balance on two legs; high prices and short battery life have also limited their appeal.
Shopping malls and tourist attractions in several Chinese cities host humanoids that perform for crowds, and in Beijing bipedal robots are running half-marathons.
The R&D hub on the outskirts of Shanghai houses more than 100 companies that research, develop, manufacture and market robotic technologies and equipment.
Many firms at the hub build specialized robots for practical tasks such as transporting heavy loads or inspecting sewage pipes.
DexForce exhibited a humanoid model that can make coffee at the carnival; its product manager Yang Duan said such machines "can free people from physical work and mundane chores".
Follow any of these and your For You feed starts watching them — no settings page required.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Single first-hand dispatch, one unsourced market figure
All claims rest on one reporter's day at one event from one publisher. The load-bearing quantitative claim — nearly 90% of 13,000+ global deliveries — carries no named data provider, methodology or definition of a delivery, and the constraint and skepticism claims are unattributed generalizations. Direct observation of the carnival, the hub's scale, and a named vendor quote are solid, which keeps evidence mid-range rather than low.
Real unit volume, but observed use is demonstration
There is a concrete global delivery number in the thousands and a dense supplier cluster, which is more than vaporware. But every use the source actually observes is exhibition or entertainment — carnival demos, mall and tourist-site performances, half-marathons — and no paying industrial customer, workload, utilization or duration is disclosed. Practical task robots at the hub are described generically without deployment counts.
Leadership framing runs ahead of demonstrated productive use
The framing — world leadership, embodied AI in the five-year plan, machines that will 'free people from physical work' — outpaces what is evidenced: demo tasks, performances, and an unsourced unit count with no revenue or workload data. The gap is moderate rather than severe because the same article discloses the deflating facts itself (mostly for show, PR campaign, safety, price and battery limits, roboticist skepticism about the form factor).
Vendor, event and state promotion all in frame
The observed material is generated by parties with clear promotional interest: a vendor product manager pitching labour-freeing benefits at his own booth, an R&D park staging a public carnival, and a national five-year plan that makes embodied AI a policy objective. The reporter explicitly names the showcase dynamic as a PR campaign, so the incentive structure is visible rather than hidden, but it heavily shapes what was available to observe.
Low-to-moderate: one publisher, one day, one unsourced number
Confidence is limited by single-source, single-occasion reporting with no corroboration and no provenance for the pivotal delivery statistic. What raises it above low is that the observational claims are first-hand and internally consistent, the publisher is established, and the piece self-discloses the promotional context rather than concealing it.
invest
XPeng's $900m robot spin-out comes with a seven-year clock and a $1bn put2 distinct publishers
invest
LG's wheels ship in 2026, its legs get a launch event in 20271 distinct publisher
build
Unitree's 8,000x book is now the comparable every humanoid banker will quote2 distinct publishers
invest
Unitree prices at 35 times sales, and humanoid robotics finally has a public comparable1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 25, 2026