The money is moving faster than the engineering.

€176 billion in cumulative European data centre investment is expected between 2026 and 2031. Power capacity is forecast to grow from just over 10 gigawatts to nearly 15 in two years. AI workloads are driving demand for facilities that consume electricity on the scale of small cities.

Governments are competing to attract investment. Planning regimes are being reformed. Grid connection queues are being restructured. The UK has designated AI Growth Zones. The Nordics are emerging as the fastest-growing data centre region in Europe.

The capital is flowing. The ambition is real.

But as with every major infrastructure cycle, commitments are being locked in well ahead of the engineering.

Why the Nordics

The attraction is clear.

Traditional data centre hub – London, Dublin, Amsterdam and Frankfurt – are running into physical limits. Grid connection waits now stretch to seven or even ten years in some markets. In Ireland, nearly €6 billion worth of projects sit stranded: land secured, permits in place, but no power available.

The Nordics offer what those markets increasingly cannot. Abundant renewable energy. Cooler climates that reduce cooling costs. Available land. Grid capacity that has not yet been fully consumed.

For AI workloads – continuous, power-intensive and heat-generating – these are structural advantages – and in some cases, beginning to look like constraints elsewhere.

Equinix’s acquisition of atNorth for $4 billion underlines how seriously the market is moving. One gigawatt of secured capacity across multiple Nordic countries, with facilities already configured for high-density AI workloads.

The investment case is compelling.

But strong investment cases have a habit of shifting bottlenecks rather than removing them. Power may be available. That does not mean everything else lines up behind it.

The Moving Parts

The difficulty is not that any single element is impossible.

It is that everything important sits outside your direct control.

Power connections depend on grid operators with their own constraints and sequencing.

Planning approvals depend on local authorities and, increasingly, on communities with strong views about energy use and land impact.

Supply chains for specialist equipment are global and already stretched. Cooling technology for AI-density workloads is evolving faster than most operators can deploy it. Tenant demand rests on hyperscaler commitments tied to AI adoption rates that have not yet been tested at scale.

Each of these moves to a different timetable. Each is driven by different incentives.

The program assumes they will align.

They rarely do.

What looks like a coherent plan is often a tightly coupled system of interdependent assumptions. The numbers work. The sequencing appears logical. The risk register is complete.

But the system itself has not been tested under real conditions.

And that is where programs begin to drift.

The Capacity Trap

The organisations delivering these programs are not starting from a blank sheet of paper.

They are already running live platforms. Existing data centres. Existing customers. Existing uptime commitments. Network operations. Maintenance cycles. Capacity upgrades already in progress.

The new build is layered on top.

The same teams. The same specialists. The same leadership bandwidth.

On paper, the organisation appears fully resourced.

In practice, the work is carried by fragments of people – stretched across competing priorities, moving between programs, and never quite available when it matters.

Dependencies begin to slip. Sequencing weakens. Coordination across grid operators, planning authorities, suppliers, contractors and vendors becomes reactive rather than controlled.

At that point, success no longer depends on the elegance of the original design.

It depends on something much more practical.

The ability to orchestrate under constraint. To surface problems early. To make trade-offs before they become irreversible.

Most organisations are not set up to do this well.

The Assumptions Underneath

There is a deeper layer that deserves attention.

The scale of current investment rests on a chain of reinforcing assumptions.

Hyperscalers commit on the basis of projected AI demand. Developers secure land and power based on those commitments. Investors fund construction against those projections.

Each step reinforces the last.

But the underlying demand curve remains unproven at this scale.

If AI adoption accelerates as expected, the investment will look prescient.

If it evolves differently – slower, geographically uneven, or shaped by different technical architectures – some of those commitments will prove difficult to unwind.

By the time that becomes clear, capital will already have been deployed.

And the room to adjust will be limited.

Where This Leaves Us

The data centre boom is not primarily a technology story or an investment story.

It is an execution story.

The technology works. The capital is available. The demand – today – is real.

The question is whether these programs can be delivered under conditions where most of the critical dependencies sit outside direct control.

Power. Planning. Supply chains. Cooling. Demand.

Each one is someone else’s problem – until it becomes yours.

And by the time that happens, the room to adjust is usually much smaller than expected.

The investment case is already priced in.

The execution risk is not.

About the author

David Hilliard is founder of Mentor, specialists in strategic program execution.

You can call him on 0118 359 2444 or email david.hilliard@mentoreurope.com.