AI’s Next Bottleneck Is Physical

Built environment firms have spent years hearing that AI will change their world. But how are they changing AI’s world in return?

When Jacobs raised its 2026 profit forecast this week, the company pointed straight at demand for AI-related infrastructure as one reason. A recent Bridgewater analysis estimates big tech will pour roughly 650 billion dollars into AI this year, much of it into data centres, chips and the power and cooling they require (Reuters, 2026). That money lands first with people who secure land, reinforce grids and design buildings, long before it shows up in the next chatbot demo.

The geography of AI is starting to look a lot like the geography of permitting. Nokia’s chief executive has already warned that Europe risks falling behind the United States and China in the build up of AI data centre capacity (Reuters, 2026). Behind that warning sit very specific bottlenecks: sites close enough to transmission lines, substations that can actually take the load, industrial parks that can handle heavy cooling, and local authorities willing to live with the noise and traffic. The regions that sort those details out move up the AI map; the ones that do not become spectators.

Inside projects, AI has stopped being a distant buzzword and started turning into everyday tooling. Autodesk’s 2026 construction trends roundup describes contractors using AI for schedule optimisation, clash resolution and site coordination (Autodesk, 2026). BuildCheck AI talks about “physical AI” on sites: systems that can track progress, flag deviations and help supervisors move labour and equipment more intelligently (BuildCheck AI, 2026). For project teams, this is about fewer reworks, tighter risk management and slightly saner weeks before handover.

Jobs are changing alongside the software. A recent piece on AI and digitalisation in the built environment argues that roles across construction, operations and facilities are already tilting towards hybrid skill sets: people who can read both drawings and dashboards (The Sustainable Times, 2026). Site managers who know how to question an AI-generated risk score, check it against what they saw on their walk, and then adjust the plan are starting to look like core talent rather than a niche. The practical “AI skill” here is not learning to code; it is learning how to interrogate the tools that increasingly sit between you and the work.

Policy conversations are lagging behind this shift. Most governance debates still orbit model behaviour and online harms, while the hard limits on AI growth are increasingly physical: power, water, land and labour. If grid upgrades and zoning approvals for data centre clusters take years, the next generation of models will wait on planning hearings and transformer deliveries as much as on new chips. That is squarely a built-world problem, but it rarely shows up in the way AI risk is framed.

For firms in construction, engineering, real estate and facilities, the question is no longer whether AI is “coming for” their industry. The question is how deliberately they plan to sit at the table where decisions about siting, resilience, cooling strategies, skills and community impacts are being made. The organisations that treat those choices as central strategy, not back-office support, will end up deciding what the AI boom physically looks like and who it serves.