Which Robots Is America Actually Buying Right Now?

Same Order Book, Different Shopping List

A3's first-half 2026 total was 17,995 robots ordered, worth $1.166 billion, up 2.0 percent in units and 6.6 percent in value from the same period in 2025. Broken out by industry, the growth is lopsided – and the segment falling is the one everyone assumes is booming.

Automotive OEM – the historic center of the robotics industry, the segment that put "robot" and "assembly line" in the same sentence for fifty years – is the only line in that table shrinking. Non-automotive orders now make up 56 percent of everything A3 counted in the second quarter, a threshold this market has not crossed before. Collaborative robots cluster heavily in exactly the two fastest-growing sectors: healthcare and electronics together take four out of every five cobot orders placed.

Read plainly, robot buying grew far faster at a chip fab or a pharmaceutical packaging line this spring than at a car plant. It does not say a fab bought more arms than a car plant did – these are growth rates against last year, and A3 publishes no sector volumes to settle that. Read against the headlines, even the growth rates are close to the opposite of the story everyone was telling.

What Seven Percent Growth Was Supposed to Look Like

Going into 2026, the industry's own forecasters expected a real recovery. Interact Analysis projected global industrial robot shipments returning to growth above 7 percent this year, following a 5.1 percent rebound in 2025 and a 2 percent contraction in 2024. That is a volume story – more robots, broadly, across a market coming off two rough years.

What actually arrived in North America looks nothing like that shape. Unit growth for the first half came in at 2.0 percent, well under the forecast pace, even as order value rose more than three times faster. The scopes are not identical – Interact Analysis's figure is global shipments, A3's is North American orders – so this is not a clean beat-or-miss against a single number.

But the two data sets tell a consistent story on direction: the recovery forecasters expected to show up broadly, in volume, instead showed up narrowly, in dollars, concentrated in a handful of sectors that were not the ones carrying the market before.

The Robots Not in This Order Book

It is worth stating plainly what "does not count humanoids" means here, because it is doing real work in this article. A3's tracked categories are traditional articulated, SCARA and Cartesian arms as defined by ISO and IFR, plus the collaborative-robot class it started tracking separately in 2025. A humanoid platform is a different statistical animal, and none of this quarter's growth or contraction numbers include one.

That gap matters this particular week. A Hyundai Korea labor dispute in July drew national coverage – Forbes ran the headline "We Just Had The First Humanoid Robot Strike Ever" – over union demands that no robot enter a plant without a labor agreement, and warnings about an "employment shock" from the 25,000-plus Atlas units Hyundai plans across its US and Korean plants.

This desk covered that same strike in early August, and corrected its own framing after publication: the dispute centered on pay and retirement age, and Atlas deployment "is not part of the current labor-management discussions," with no deployment date set for Korean plants and Georgia not scheduled to start until 2028.

The same week, Washington moved to bar new imports of Chinese-made humanoid and quadruped robots on national-security grounds – a real policy action this desk covered in detail when it happened, and one that reshapes where humanoid supply can legally come from, without adding a single unit to anyone's order book.

The honest counter to all of this: humanoid robots are pre-revenue by design right now, not failing to compete. Figure has said plainly that the units it ships go to its own research and data-collection groups, not to paying customers – a deliberate ramp, not a stalled one. A3's order series was built for an industrial-robot market that existed before humanoids did, so their absence from it is a category-definition fact, not a verdict on whether they eventually scale.

What this quarter's data shows is narrower and more checkable than that: where the money already moved, not where it will move next.

Where the Extra Dollars Are Probably Going

A3 publishes unit counts and dollar totals for each period, plus year-over-year growth rates for both. It does not publish the number those figures imply: the average price paid per robot, period by period. Dividing dollars by units, using A3's own two headline percentages to back out each period's prior-year comparator, turns up something A3 never states outright – the swing shown above was not a steady climb.

Implied average order value per robot fell roughly 6.2 percent year-over-year in the first quarter of 2026 – about $64,100 down to $60,100 – then rose roughly 16.3 percent in the second, from about $59,800 to $69,600. Every 2025 figure there is back-solved from A3's own published growth rates rather than published by A3, so the direction and rough magnitude are what carry weight, not the last digit. Something specific changed in the April-to-June window, not across the whole first half evenly.

The tempting explanation is the sector mix: semiconductor and life-sciences robots, built for cleanroom precision and often paired with vision systems, cost more than the automotive-OEM arms they are partly displacing. That is plausible, and it is not proven. A3 does not publish average order value by sector, and the sectors driving this quarter's growth are also the sectors buying the most collaborative robots – the cheapest category in the data, priced roughly 37 percent below the period's blended average. Healthcare and electronics together account for 80 percent of all cobot orders.

That pull should drag those sectors' average price down, not up. The market-wide average rose anyway, which is consistent with those buyers placing non-cobot systems expensive enough to outweigh their own appetite for cheap collaborative arms. It does not establish it. Cobots were 15.4 percent of first-half units but only 12.7 percent of second-quarter ones – a thinner slice of the cheapest category lifts a blended average by itself, without a single robot costing a dollar more. A3's public data cannot separate the two explanations.

The technology behind that precision demand is real but still maturing. A May 2026 paper on automated optical inspection of wire-bonded semiconductors – the kind of AI vision layer a chip-sector robot cell increasingly runs alongside its arm – describes a persistent friction: segmentation models are "typically device-specific and must be re-trained when new devices or distribution shifts appear," and the paper's own contribution is a method to cut that retraining burden, not eliminate it.

Nobody has published how much of the integration cost behind this quarter's semiconductor orders is arm, and how much is the AI vision stack bolted onto it.