The headlines about AI transforming manufacturing rarely mention what is actually happening on most shop floors. Walk into a typical mid-sized facility today and the tools running production, dispatch, and inventory tracking often look remarkably similar to a decade ago: paper logs, whiteboards, and spreadsheets nobody fully trusts. Understanding both sides of this picture, the genuine acceleration and the stubborn gap behind it, matters for anyone trying to figure out where manufacturing actually stands right now, rather than where the more optimistic headlines suggest it stands.
The Growth Numbers Are Real
AI usage in US manufacturing has genuinely accelerated. According to a report from Digit Software analysing US Census Bureau data, the share of manufacturers reporting AI use in any business function rose from just 1.8 percent in September 2023 to 13.9 percent by February 2026, a sevenfold increase in under three years. Growth nearly doubled year-over-year between 2024 and 2025 alone.
That trajectory looks dramatic on a chart, and it is a real shift worth paying attention to. It also obscures a much larger, less discussed fact sitting right beneath it.
Most Manufacturers Still Haven’t Made the Switch
The same research found that 87 percent of US manufacturers have yet to integrate AI into their operations in any meaningful way. As automation.com’s coverage of the report put it, the growth is impressive as a multiple but considerably less dramatic once you remember that nine out of ten manufacturers are still standing on the sidelines.
Business size turned out to be the strongest predictor of adoption. Larger manufacturers, with more capital and dedicated IT resources, are moving first. Smaller and mid-sized operations, which make up the bulk of the sector, are largely still running the same processes they were running years ago.
Why the Infrastructure Gap Persists
For a lot of manufacturers, the barrier isn’t scepticism about AI itself, it’s that the underlying data infrastructure was never built to support it. Systems that track inventory, production schedules, and bills of materials often live in disconnected spreadsheets or, in a surprising number of cases, still rely on paper records passed between departments by hand. A shift change or a supplier substitution logged on a clipboard rather than in a shared system simply doesn’t exist as usable data for any AI tool downstream.
AI cannot meaningfully analyse or act on data that doesn’t exist in a structured, accessible format, a point covered in more depth in this article on how AI actually works in manufacturing settings. A predictive maintenance model is only as good as the sensor and machine data feeding it, and a demand forecasting tool is only as useful as the sales and inventory records behind it. This is one reason connected ERP platforms are increasingly framed as a prerequisite for AI adoption rather than a separate initiative running in parallel. Manufacturers trying to bolt AI onto an already fragmented data environment tend to see far weaker results than those who fix the underlying data flow first.
Where the Real Constraints Sit
Cost is part of the story, but not the whole story. Legacy systems that would need replacing, a shortage of staff who understand both manufacturing operations and data tooling, and simple organisational inertia all play a role. Even manufacturers that want to modernise often discover that the compute, storage, and data governance requirements involved are more demanding than expected. Air-gapped networks, edge hardware constraints, and regulatory data residency rules all add complexity that a typical cloud-native AI rollout in another industry simply doesn’t have to account for.
For manufacturers without that infrastructure already in place, adopting AI often means solving a data problem first and an AI problem second. Skipping straight to the AI layer without addressing the data underneath it tends to produce disappointing pilots rather than lasting change.
What This Means for the Rest of the Decade
The 1.8 to 13.9 percent trajectory suggests manufacturing AI adoption is still in its early, steep part of the curve rather than anywhere near saturation. That is a meaningfully different story than the one implied by headlines suggesting AI has already reshaped the factory floor, and it matters for how manufacturers should actually be thinking about timing and investment.
For manufacturers weighing when to invest, the practical takeaway is straightforward: the gap between AI-ready and AI-capable operations comes down almost entirely to whether the underlying data is connected and structured in the first place. Manufacturers that solve that problem now will be positioned to move quickly once the technology matures further, capturing early advantages while competitors are still working through pilots. Those that don’t will likely find themselves further behind with each passing year, watching the adoption curve steepen from the sidelines rather than riding it, and facing a steeper, more expensive catch-up later on.
