AI use among US manufacturers has risen sevenfold in under three years, faster than the sector adopted robotics. The real barrier left is knowing where to point it.
The US Census Bureau has tracked AI use among American businesses every two weeks since September 2023, asking roughly 200,000 firms whether they use it in any part of their operations. For manufacturing specifically, the share reporting AI use in any business function rose from 1.8% in September 2023 to 13.9% by February 2026, a sevenfold increase in under three years. That’s a genuine, government-collected trend line, not a survey snapshot dressed up as one.
The number that gets repeated instead is the other side of it: 87% of US manufacturers still haven’t integrated AI anywhere in their operations. It’s true, and it’s the more common framing in trade coverage of this same Census data. It’s also the less useful number on its own, because it implies manufacturing is failing to move. Whether that’s true depends entirely on what pace counts as normal for this industry, and manufacturing has a much longer, better-documented adoption history to measure itself against than the three years of AI data allow for.
The International Federation of Robotics has tracked industrial robot density worldwide for over a decade. Global robot density per 10,000 employees went from 74 in 2016 to 162 in 2023, doubling in seven years, a compound rate of roughly 10% a year. That’s manufacturing’s real, long-run tempo for absorbing a new production technology once it’s proven itself: not overnight, but steady, over the better part of a decade. Measured against that baseline, AI’s sevenfold rise in under three years isn’t manufacturing lagging. It’s manufacturing moving at something close to three times its own historical pace. The 87% framing measures the industry against the software sector’s adoption speed, which was never the right yardstick for a sector that also has to worry about safety certification, legacy equipment with a 20-year service life, and production lines that can’t go down for an afternoon, let alone a quarter.
Exhibit 1
AI reached its full measured rise in a third of the time robot density took to double
Sources: US Census Bureau, Business Trends and Outlook Survey, manufacturing sector (Sept 2023–Feb 2026, 1.8%→13.9%). International Federation of Robotics, global robot density (2016–2023, 74→162 per 10,000 employees). Each line connects only the two published measurement points on record, not a continuously observed path.
So if the pace isn’t the real problem, what is? The Manufacturing Extension Partnership network’s national assessment, drawing on responses from 475 manufacturers collected through mid-2026, found only 10% have integrated AI into day-to-day operations and 48% report only basic or limited awareness of it. Asked what’s actually stopping them, cost isn’t the dominant answer. Seventy-five percent cited difficulty identifying which applications would create measurable value as their top barrier, more than half cited a shortage of technical talent to implement or maintain it, and the report’s own conclusion is explicit: the barriers “are less about access to technology and more about organisational readiness and implementation capability.” The same assessment found 17.4% of manufacturers currently plan to adopt AI, against 12.6% who already have. More manufacturers are queuing up to start than have finished starting, and that gap hasn’t closed as adoption climbed, it’s held roughly steady the whole way up.
Exhibit 2
What’s actually stopping adoption isn’t cost or access to technology
Source: Manufacturing Extension Partnership National Network, National AI Strategic Assessment (July 2026), 475 US manufacturers. The talent-shortage figure is reported by MEP only as “more than half”; no more precise published figure was found.
Robots took seven years to double
Robots took seven years to double in density partly because factories had to work out, machine by machine, where a robot arm actually paid for itself. AI is running into the identical problem at a faster clip, not a slower one.
That’s the actual finding worth sitting with. Manufacturing isn’t behind its own history. It’s moving faster than its own history, while running into the same constraint that has always slowed it down: not a reluctance to try new production technology, but a persistent shortage of people who know precisely where to point it.
That has a direct implication for what closes the gap next, and it isn’t more AI vendors selling more capability into a market that already has 87% of its addressable customers sitting on the sidelines. The MEP assessment argues its own network exists to solve exactly this problem, matching manufacturers to specific, provable use cases rather than selling them a platform and leaving them to find one. That’s a self-interested conclusion from an organisation built to deliver that service, worth noting plainly rather than passing on uncritically. But the underlying data supports the direction even if the messenger has a stake in it: when three in four manufacturers say their top barrier is not knowing where to point the technology, the bottleneck to close first is expertise, not technology access, and that will keep being true regardless of who’s positioned to solve it.
Sources
- US Census Bureau, Business Trends and Outlook Survey (BTOS)
- Manufacturing-sector BTOS figures as compiled by Digit Software, “The State of AI Adoption in US Manufacturing”
- International Federation of Robotics, “Global Robot Density in Factories Doubled in Seven Years”
- Manufacturing Extension Partnership National Network, “National AI Strategic Assessment: From Readiness to Value Creation” (July 2026)
- GE Appliances Pressroom, Google Cloud Gemini Enterprise announcement
- AWS Press Center, ArcelorMittal–AWS collaboration
