Start with an AI chip itself, something the size of a stamp, pulsing with billions of transistors, each one sitting on a circuit board roughly the size of a paperback book.

Pull back a little, and you will see it is one of a handful locked inside a liquid‑cooled blade, a high-tech “drawer”, stacked by the dozen in a towering steel cabinet – a single rack that stands as one pillar in a cavernous hall that runs the length of a stadium.

Zoom out further, to a bird’s eye view, and add another building, and another, anything from three to fifteen more, which, taken together, add up to millions of square feet in aluminium, concrete, and wire. 

This is a hyperscale data centre campus, the physical backbone of the digital economy – vast, industrial-scale assets that demand the same level of planning, investment and coordination as any piece of national infrastructure.

Right now, more than a thousand are scattered around the world, hundreds across the US, UK, and Europe, but many, many more are being planned, with recent research suggesting hyperscale capacity will likely triple by 2030.

But can it really happen? Can all of these data centre projects, all those blades, racks, and cabinets, all that concrete, copper, steel, and wire, realistically be built in the next four years?

Big Money, Brutal Deadlines

Competition has added to the pressure, with major AI operators now expecting hyperscale builds to wrap in just 14 to 15 months, nearly double the pace of typical construction projects, and with even more aggressive schedules looming.

It’s a very tight timetable, but one that Aviad Almagor, VP of technology innovation at construction tech firm Trimble Inc, believes is in reach.

“It’s not about whether it’s possible to do it; it is possible. It’s not easy, but it’s possible.”

Meeting those expectations is a big ask, but Big Tech, venture capitalists and AI firms have enough cash to brute‑force their way through almost anything. Forecasts estimate that as much as $7 trillion could be spent globally on data centres by 2030, with average-sized facilities costing between $500 million and $2 billion to complete.

For contractors, even with final margins often being squeezed down to just 2-6%, this means a mountain of money is waiting for those who can survive the pressure and build fast.

“It’s a very tough position for a general contractor looking at the potential revenue stream from a data centre to say no, I’m not taking it,” said Almagor. “The better strategy for those contractors is to figure out how technology can help them cope with this challenge.”

AI Becomes the Master Builder

It’s not enough for data centre contractors to rely on old tools of the trade, like spreadsheets, paper plans, and manual workflows, argues Almagor. Delivering at hyperscale demands a tech‑first approach that fuses proven tools like 3D modelling and BIM with next-gen tech, like agentic AI and digital twins.

“A data centre is not a standard building,” says Almagor, “it’s a very delicate machine that needs to operate in a very reliable way.”

“In an ideal scenario, data centres would be built digitally first and only then built physically, while keeping the physical and digital environments in sync.”

Risking a metaphor of the snake eating its own tail, AI is the anchor of this tech-first execution, argues Almagor, central to the planning, modelling, optimising, and operation of fast hyperscale builds.

Agentic AI, in particular, can automate design, test logistical scenarios, spot errors long before a human would, and parse dense environmental data – all in parallel and with a level of coordination that is difficult for distributed human teams to achieve.

“What we can do with emerging tech and AI is ensure that we are utilising a connected data environment where everything is basically under one roof,” says Almagor.

“Everything is connected, from 2D drawings to 3D models, to scans of the site, to machine control information, everything is under one single environment, which means that when you deploy AI, the data that you will get is reflecting reality.

“It’s really about creating a digital replica and making sure you keep this digital twin up-to-date to address any issues.”

This combination of technologies, with AI acting as a responsible ringmaster, is crucial to cutting costs and saving time. 

Studies suggest that smart construction tech can trim 10–20% off delivery times, but it also pays dividends in day‑to‑day operations, as DeepMind famously demonstrated back in 2016, when its AI reduced Google’s data centre energy bill by 40%.

“The key aspect for [owners and contractors] is predictability,” says Almagor. “They care less about innovative ideas concerning emerging tech; they want to make sure that the technology they’re utilising is predictable. 

“A connected data environment, together with AI…provides what we call a ‘single source of truth’ that leads to efficiency, to confidence in the data, and reduced risk.”

Will the Hype Outrun the Hardware?

The elephant squatting in the corner, however, is the encroaching threat of the AI bubble bursting.

Already, big projects from industry heavyweights are stalling. Last year saw Microsoft abandon a $1 billion data centre project in Ohio, citing a “strategic investment review” driven by overspending on AI infrastructure that spooked investors, while Oracle lost a major backer for a $10 billion data centre being built for OpenAI.

Even in the US, where AI is contributing more to GDP than any other country, data centre cancellations almost quadrupled last year, with at least 25 major projects scrapped, compared to six in 2024.

The reasons for this collapse are myriad, but one issue is unmistakable: the data centres being built right now may be obsolete before they’re finished. 

Next-generation AI racks, with dense GPU clusters and coolant-filled manifolds pushing the weight limits of traditional designs to their limits, while the extreme thermal and electrical demands of the latest chips may require fundamentally different architectures.

Recommended reading

• Data Centres Are the World’s Hottest Property Investment
• How Badly Will AI Data Centres Be Impacted By Power Shortages?
• Are Traditional Data Centres Fit For Purpose?

Labour shortages are also becoming one of the most destabilising forces in the data centre buildout, with operators competing for the same limited pool of electricians, plumbers, HVAC specialists, and engineers, with the US alone facing a shortage of 130,000 electricians and 150,000 construction supervisors by 2030.

The result is a growing backlog of half‑finished sites waiting for specialist teams to become available, with planned data centres already running an average of 8.5 months behind schedule.

Because upskilling and training new talent takes time, technology will have to fill the gaps, which Almagor argues means a digital-first execution is essential to ensure that the construction of data centres is still the “safest part of the equation” when it comes to AI investment.

“The capability gained by adopting this type of technology remains valuable regardless of the short-term market sentiment, so even in the case of a bubble, it can be deployed in another project.

“It might not be a successful data centre at the end, but still be a successful construction project.”