As pressure mounts to bring new AI capacity online faster, Matt Salter, Head of Data Centres at Onnec, argues that getting infrastructure right first time is becoming more important than ever.
A few milliseconds should not bring a data centre to its knees.
Yet over the summer, we’ve seen just this. Millisecond voltage drops, power and cooling failures have knocked out data centre services for multiple hours and even days.
The incidents haven’t been publicly linked to rushed construction. But they show how little margin data centres have for a design flaw, wherever it comes from. And that margin is about to get thinner, as Europe gears up to triple its data centre capacity within the next five to seven years to meet growing demand for AI and cloud services.
That capacity needs to come online quickly. But if the rush to build means facilities need reworking, upgrading or fixing soon after go-live, the industry risks trading speed for higher costs, more disruption and unnecessary use of equipment and materials.
The pressure to build before the picture is clear
AI is forcing data centre operators to move faster. Recent research from Onnec found that 92% are accelerating build timelines, while 75% say speed-to-market pressure is forcing design decisions before infrastructure requirements are fully understood.
An AI data centre is a tightly connected system. More compute means more power and heat. Higher densities change cooling requirements. Growing network demand affects fibre, cabling routes and physical space. Make one decision too late and the consequences can ripple through the rest of the design.
The pace of change in AI hardware only raises the stakes. What looks AI-ready when a project starts may no longer match what customers need by the time it opens.
Operators cannot wait for perfect certainty. But nor should designing on the fly become an accepted part of moving quickly. The challenge is making better-informed decisions earlier, while designs are still flexible and changes do not mean expensive rework elsewhere.
When commissioning becomes the place to claw back time
The pressure to build fast becomes even harder to manage towards the end of a project.
Planning delays, grid connections, supply chain issues and construction overruns can all eat into the programme, while the go-live date remains fixed. Testing and commissioning then become an obvious place to claw back time. But this is where a single fault could turn into a 15-hour outage or three days of degraded connectivity.
Already, 45% of operators say accelerated timelines are reducing the time available for testing, commissioning or quality assurance. And 43% say infrastructure is having to be upgraded or remediated after go-live. That shifts risk from construction into operations, where problems are harder and more expensive to resolve.
Commissioning is where operators find out whether backup power responds correctly, cooling recovers as expected and multiple systems behave properly together when something fails. If those answers only emerge once customers are live, fixing them means working around operational equipment, finding maintenance windows and potentially replacing hardware.
Saving days before go-live is little consolation if it creates weeks of remediation afterwards.
Get the right expertise in earlier
The solution is making better use of the time operators have.
That means bringing the right infrastructure expertise into a project earlier, before key decisions around power, cooling, cabling and connectivity are locked in. These systems increasingly depend on one another, so spotting constraints early can prevent costly redesign or remediation later.
Experience of large, high-pressure projects can also help teams maintain speed without sacrificing quality. Clear processes, resource planning and continuity through commissioning can enable a cleaner handover and reduce problems later.
AI hardware, rack densities and connectivity requirements are evolving quickly, so facilities need enough flexibility to adapt without major rework every time requirements shift.
And bringing that expertise in earlier does not mean slowing delivery down. Prefabrication and repeatable designs can reduce complexity on site, while maintaining continuity through commissioning helps ensure that what was designed, what gets installed and what is ultimately handed over remain aligned.
The point is not to make every project more complicated. It is to solve the difficult questions earlier and get things right the first time, when problems are still easier – and cheaper – to address.
Build for day two, not just day one
Europe is about to embark on an extraordinary data centre build-out, and operators are being asked to deliver capacity at a pace the industry has not seen before. Simply telling them to slow down is not realistic.
The challenge is to make sure the pressure to deliver today does not create problems that have to be repaired tomorrow. That means planning for change, protecting the parts of the programme that prove resilience, and addressing potential weaknesses while they are still relatively straightforward to solve.
Day one is go-live. Day two is everything that follows, and it is what operators are being judged on. That is what speed-to-market should really mean. It’s not just about reaching go-live faster, but reaching it in a state that lasts. Anything else just delays the real cost.
