Product1 distinct publisher3 min readPublished
Counterpoint puts five Chinese makers at 86% of the 22,000 humanoids shipped in the first half of 2026. That is the cost curve the Covered List and the new tariffs do not touch while they reprice US bills of materials.
The Product Desk · Product desk

Compiled by The Product DeskSomething wrong?How this is made
The person who answers for this on Friday is whoever owns the bill of materials on a drone or a mobile robot sold into the US. Two dates go on that person's wall: September for the drone tariffs, and 2027 for the additional component tariffs [2]. What is missing is scope text. TechCrunch calls the drone tariffs steep without publishing rates, and describes the Covered List's newest category as advanced robotic devices without naming which devices or vendors are in it [20]. A dated rule with an undefined parts list does not produce a supplier swap; it produces a hold on purchase orders while legal reads the entry.
That is the pattern the list has followed since 2021, when it covered telecommunications and surveillance equipment from Huawei, ZTE and Hikvision [3] before extending to foreign-made drones and then to robotics [4]. Each extension moves a line item out of sourcing and into documentation, and documentation work is the part no roadmap slide budgets for.
The arithmetic on the other side of the wall looks different. Counterpoint counted 22,000 humanoid shipments worldwide in the first half of 2026 [5], with AgiBot, Unitree, Galbot, UBTECH and Leju Robotics taking 86% of them [6]. That leaves about 3,080 units for every other manufacturer on the planet [7]. Volume at that ratio is what sets component pricing, and Counterpoint's Soumen Mandal says Chinese humanoid makers are cutting costs further by pulling more of the stack in-house, with Unitree building more components itself [14]. Ankur Saxena of TDK Ventures describes the loop that follows: cheaper robots get deployed in larger numbers, which generates real-world data that improves the product, and the volume pushes cost down again [15]. "You cannot sanction your way around a cost curve. You can only out-build it, and America has yet to begin making the decade-long investment that will require," he told TechCrunch [9].
The next BOM review needs a simple test applied line by line: does a non-covered source exist at any price before your ship date, and will your buyer pay the difference? Yes and yes means you reprice and keep going, and you are selling into the segment where security requirements are part of the spec, which is where the analysts and executives who spoke to TechCrunch expect US and allied manufacturers to compete [11]. Yes and no means you are a cost-plus vendor facing a competitor whose unit costs keep falling [8]. No and yes means you are designing a part instead of buying one, so the date that slips is your schedule. No and no means the SKU is finished, and saying so in September is cheaper than finding out when component tariffs widen in 2027 [2].
The export line gets no cover from any of this. Mandal says Chinese robotics firms are already working price-sensitive markets with severe labor shortages in Europe, Southeast Asia, Latin America and the Middle East [12], and he expects humanoid makers to repeat the Chinese EV route of building scale at home, expanding overseas, then producing locally [13]. A tariff only works at the border it was written for. Saxena's own division of labour, with the US ahead in frontier AI and software and China ahead in manufacturing scale and supply-chain depth [10], is a description of which of those two things a purchase order can buy this quarter.
Ranked by verification strength, evidence, and original report placement.
In July and August, Washington tightened restrictions on foreign-made advanced robotic systems and imposed steep tariffs on imported drones and their components, both moves citing national-security concerns.
The drone tariffs take effect in September, with additional component tariffs following in 2027.
The FCC's Covered List, established in 2021, initially targeted telecommunications and surveillance equipment from companies including Huawei, ZTE and Hikvision.
The Covered List expanded to foreign-made drones and, most recently, to advanced robotic devices.
Global humanoid robot shipments hit 22,000 units in the first half of 2026, the vast majority from Chinese manufacturers, according to a Counterpoint report.
The world's five largest humanoid robot makers by shipments (AgiBot, Unitree, Galbot, UBTECH and Leju Robotics) were all Chinese and together accounted for 86% of global shipments in the first half of 2026, according to Counterpoint.
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 30, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
build
Unitree's 8,000x book is now the comparable every humanoid banker will quote2 distinct publishers
leadership
A 2.85x bill of materials: what the FCC's robot ban does to an automation budget1 distinct publisher
build
Agility wants public money for the jump from nine sites to volume, and Unitree just set the price1 distinct publisher
invest
A $904m robot IPO is outdrawing a $3bn Shein listing, and the multiples explain why1 distinct publisher
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.
One newsroom, one dataset, no tariff rates
The quantitative spine — 22,000 units, 86% from five Chinese makers — is a single Counterpoint reading passed through a single newsroom, with no link to the report and no second count anywhere in the coverage. The policy half is weaker still: TechCrunch calls the duties steep and never prints a percentage, and describes the Covered List addition without naming a device or a vendor. What holds up well is the attribution discipline; every interpretive claim carries a named speaker.
Real units shipped, concentrated in five hands
Unlike most robotics coverage, this story has a hard denominator: 22,000 humanoids in six months, with roughly 18,900 of them from AgiBot, Unitree, Galbot, UBTECH and Leju. That is genuine deployment, not pilots. The gap is on the American side — no U.S. shipment figure, no fleet count, nothing beyond Agility saying Digit is built domestically, so the scale asymmetry the story asserts is measured in only one direction.
Deflationary thesis, slightly overconfident inputs
The argument leans against hype rather than into it — its whole point is that policy will not do what its authors hope. The overshoot is narrower: a single analyst figure is treated as settled fact, and the forward half of the piece, from the EV-style localization path to the two-ecosystem drone market, is interview consensus carried in the same register as the shipment data. Confident framing, thinner footing than the framing implies.
The compliant side does most of the talking
Follow who benefits. Heven AeroTech sells NDAA-compliant drones and describes a market splitting in favor of NDAA-compliant drones. Agility Robotics welcomed the FCC decision and then pointed at its U.S.-assembled humanoid. Counterpoint's business is the shipment data the story is built on, and TDK Ventures has portfolio exposure to the 'out-build it' prescription. None of that makes the claims wrong; it does mean no quoted party loses if U.S. barriers stay up, and no Chinese manufacturer is heard from at all.
Trust the shipment gap, not the policy detail
Two different reliability tiers sit in one story. The scale picture — Chinese concentration, in-house component work, price-led expansion into labor-short regions — is coherent, named and internally consistent, and I would act on its direction. The regulatory particulars are close to unusable for planning: two dates, no rates, no covered-vendor list, and no second source on the calendar. Confidence lands mid-range because the load the story asks its one dataset to carry is heavier than one dataset should carry.