The AI data centre boom is remaking US factory supply chains, and Trump's tariffs are the wildcard

This comes with an order backlog already standing at $1.6 billion and roughly 1,000 new jobs, a 10% headcount increase, in the pipeline.

The build-out is lifting a much wider slice of US manufacturing

Generac is one of several manufacturers finding a second growth engine in the data centre boom, a boom that the International Energy Agency has found is pushing data centre electricity consumption on course to triple by 2030, with capital expenditure from five major technology companies already exceeding $400 billion in 2025 alone. Cooling system makers, electrical transformer manufacturers and construction machinery suppliers are also seeing demand pull through, and that demand cascades further still, down to producers of wire cables, pipes, cement and the prefabricated metal wall panels used on data centre buildings.

Timken, the Ohio-based maker of engineered steel bearings, is a case in point. Chief executive Lucian Boldea said data centres now add another leg of growth alongside the company’s traditional defence and aerospace customers, since those data centres need massive buildings, roads and gas turbines, all of which require the kind of components Timken makes.

Generac’s own chief executive, Aaron Jagdfeld, put the scale of the moment plainly: “The question on everybody’s mind is how long this build-out will go.”

A split economy, even within the same companies

The picture isn’t uniformly strong. The latest ISM manufacturing survey points to a gloomier mood across much of the sector, underlining a split between AI-linked niches booming on data centre orders and the wider industrial base still struggling. That split shows up inside individual companies too: Generac’s core home-generator business remains soft, weighed down by a struggling housing market and consumers facing high food and petrol prices that discourage big-ticket purchases like backup power.

Jagdfeld frames the dynamic as a virtuous circle rather than a contradiction: as AI adoption spreads, including inside his own company, demand for the data centres that run it keeps growing in turn.

The politics complicate the picture

The build-out’s manufacturing spillover sits awkwardly against the Trump administration’s own framing of a broader American factory revival. The administration has repeatedly credited its policies with driving a manufacturing boom that extends well beyond AI-linked sectors, but its shifting tariff regime has also been blamed for holding back some factory expansions, particularly those that depend on imported machinery.

The tension is visible in the investment data itself. Spending on data centre construction is up close to 18% this year, while spending on factory construction more broadly is down around 2.5%, a gap that undercuts the idea of a uniform industrial revival even as AI-linked manufacturing genuinely booms. Foxconn, OpenAI and SoftBank’s plan to repurpose the former General Motors plant in Lordstown, Ohio, into a manufacturing hub for AI equipment, complete with its own demonstration data centre, captures that tension in a single site: an emblem of America’s older industrial base being rebuilt specifically to serve the AI economy, rather than to return to what it made before.

Why it matters for the wider build-out

For an industry accustomed to tracking AI capex through hyperscaler earnings calls and GPU shipment numbers, this is a different kind of leading indicator. Order backlogs at generator, bearing and transformer manufacturers offer a read on how durable the data centre boom looks to the industrial suppliers actually building it, once removed from the hyperscalers’ own capital spending announcements.

Right now, those suppliers are telling Reuters they can’t see an end to the build-out, though Capacity has previously heard warnings from within the industry itself against taking headline specs and timelines at face value, a reminder that confidence expressed by suppliers is not the same as confirmation that every announced project gets built on schedule.

Whether tariff policy and a soft consumer economy blunt that confidence elsewhere in their businesses is the more complicated part of the story, as is the parallel rethink already under way among data centre developers themselves, who are grappling with the same power and land constraints that are driving hyperscalers toward supplier-heavy, infrastructure-first solutions in the first place.

Turning risk into reward in the AI data centre era

AI data centres could triple electricity consumption by 2030, IEA warns