The AI boom is a construction boom

A very good quarter

Skanska builds hospitals, highways, offices and metro stations across the Nordics, Europe and the United States, and has seen its share of booms. Even by that standard, the fourth quarter of 2025 stood out: roughly SEK 9.5 billion, about $1.05 billion, of new data-center orders. One quarter, one client category. When analysts asked about the AI boom on the February earnings call, CEO Anders Danielsson did not reach for a vision statement. He reached for the order intake:

He sees no AI bubble threatening data-center development and, in his words, no signs that the market will slow down; the demand comes from repeat clients who keep investing. Skanska closed 2025 with an order backlog of SEK 257.9 billion and about 23 months of booked work in the United States. This is worth taking at face value. An order backlog is one of the industry's most honest documents: it is money clients have committed, not a forecast. And for scale, that one quarter of data-center orders is roughly the price of four kilometres of new Stockholm metro.

The queue behind the orders

Skanska's US building arm publishes a market trends report every season, and the autumn 2025 edition reads like a status report from a sector running close to its ceiling. Data-center demand is "fueled by AI and cloud computing expansion". Lead times on wide-flange structural steel have stretched to 14–18 weeks. The industry's labour shortfall stands at roughly 450,000 workers, with wages rising to hold on to the people who exist. Steve Stouthamer, who leads project planning at Skanska USA Building, describes a market "shaped by both volatility and opportunity", and expects data centers, semiconductor plants and pharmaceutical manufacturing to carry the momentum into 2026.

Steel and electricians are the visible half of the constraint. The other half sits earlier in the timeline and gets less press: someone has to design all these buildings, and someone else has to review the designs.

Engineering capacity follows the same arithmetic as site labour, only slower. A steel mill can add a shift. A reviewing structural engineer takes five years of education and several more of project experience before anyone lets them sign off a load calculation.

The paperwork does not know it is for AI

From a permitting point of view, a data center is an industrial building. Before ground breaks it needs the same pre-construction chain as any factory: site studies, an environmental impact assessment, permits, a grid connection application (an unusually heavy one, in this case), structural design, and review round after review round. Every document in that chain is produced by one engineer and checked by another. The site study waits for the survey, the permit application waits for the assessment, the drawings wait for the review comments, and the review comments wait for an engineer with time. The chain runs at human speed, and it does not care that the future tenant is a machine that writes documents for a living.

Which is where the irony comes in, gently. AI is now generating demand for buildings faster than the sector can grow the engineering capacity to design and review them. The models scale by the data center; the reviewers scale by the graduating class. Skanska can book ten billion kronor of data centers in a quarter. Nobody can book ten billion kronor of review capacity, at any price, because it is not sold in quarters. It comes in careers.

When the loop closes

None of this is a complaint about Skanska, which is doing exactly what a well-run builder should: winning the work and saying so plainly. Nor is the answer to lower the bar for data centers; the studies, permits and reviews exist because industrial buildings drawing hundreds of megawatts next to towns deserve scrutiny. The constraint is narrower than either. The document chain in front of every project still moves at the speed of the humans producing and checking it, while the demand behind the projects has started compounding at the speed of the technology ordering them.

So the loop is still open. It closes when AI also does the pre-construction work its own buildings require: drafting the studies, checking the design against the requirements, answering the review comments, keeping the document chain moving between the human decisions that actually matter. The technology filling the order books is the same technology that can shorten the queue in front of them. At that point the AI boom stops being merely a construction boom and becomes a faster one. Until then, the machines wait for their permits like everyone else.