Data Centre LIVE: Debating AI and the Data Centre Revolution

The data centre industry has long operated to a familiar playbook. Sites were selected for their proximity to network infrastructure, facilities were designed to last two decades and supply chains were optimised for cost and efficiency. 

The reality now is that artificial intelligence has torn that playbook to shreds. 

At the recent instalment of Data Centre LIVE: The London Summit 2026, four senior industry figures came together for The AI Data Centre Debate to discuss the challenges and opportunities created by AI's voracious appetite for compute, power and space. Their verdict: almost every assumption underpinning data centre development is being challenged simultaneously.

Alex Bennett, Global Strategy Realisation and Transformation Director at NTT Global Data Centers, set the scene. Operators had previously managed regional facilities of 10 to 20 megawatts. Then cloud computing arrived and changed everything. Now, customers are asking for 500-megawatt campuses.

“The rules have been rewritten,” Alex contends, “and continue to be rewritten every three months.”

More than simply commercial, the pressure is architectural, logistical and geopolitical – touching everything from semiconductor supply to planning permission, from cooling technology to carbon commitments.

A new logic for site selection

Jamie Allen, Head of Site Acquisition, EMEA at Iron Mountain, describes a shift in where facilities can realistically be built. Cloud-era development was constrained to within roughly 30-40 km of major cities, driven by the need for low-latency connections to customers.

“Over the last few years, the world has opened up,” says Jamie. “We're now looking at things 150 km from a metro that we never would have dreamed of being able to acquire and develop.”

However, that geographic freedom comes with a catch. The business case for a remote facility looks very different from one in a city centre. Urban data centres can be repurposed and sweated as assets for 20 years or more. Out-of-town AI campuses may need to pay back their investment in as little as eight to 10 years.

Jean-Francois Berche, CTO at GreenScale Data Centres, traces how the priorities driving site selection have shifted over time. In the early days of cloud computing, the overriding question was network access. Then, as AI emerged, the focus moved to power.

“The way site selection is going to change moving forward,” Jean-Francois begins, “is we're going to go where you have permission to build. We're going to live in the permission first – are we welcome in this area? Otherwise, you're going to be in trouble.”

It’s an observation that reflects growing resistance in some communities to large-scale data centre development, particularly in Europe, where planning rules are complex and public scrutiny is intense.

Strengthening supply chains

At Jabil Procurement and Supply Chain Services, the challenge of matching supply chains to the pace of AI development falls squarely on the shoulders of Lonnie Salmon, Senior Director of Supply Chain. He argues that the traditional focus on price and efficiency is no longer fit for purpose.

“Within this new environment of AI,” Lonnie says, “the need for orchestration across the whole environment becomes paramount. The speed to realise the capacity is what's driving the industry.”

The bottlenecks are multiple and interconnected. Energy is the most visible constraint, but semiconductor supply is also under severe pressure, with around half of industry capacity now structured around high-density memory. Industrial equipment at site level carries lead times of up to three years.

“Demand is outstripping supply on all levels,” continues Lonnie. “It's not price and it's not efficiency – it's collaboration across the whole supply network, end to end, which is paramount.”

Avoiding obsolescence

The pace of hardware evolution creates a particular dilemma. AI chips are advancing so rapidly that a facility designed today may be partially obsolete before it opens. 

Jean-Francois describes the challenge through the lens of how the industry used to think about data centre lifespans. A traditional commercial cloud facility – what the industry calls an availability zone – might be financed over 15 to 20 years, with hardware refreshed once during that period. AI infrastructure operates on a very different timescale.

“We're not building things where the hardware lasts more than three, potentially five years,” Jean-Francois asserts. “You have to look at the data centre in terms of Lego bricks. What can you break down into components that, at some point, you pull out and replace?”

Training versus inference

The distinction between different types of AI workloads adds another layer of complexity. Training and inference have very different infrastructure requirements.

From inside the building, Jamie notes, a training facility and an inference facility look broadly similar: both feature high-density racks, liquid cooling and dense internal connectivity. The differences become apparent when you step outside.

“Inference is below one millisecond from a latency perspective,” explains Jamie. “You're going to be in the metro, 30 km out. With ML training, you're up to 20 milliseconds of latency and you can be 150 km from a metro.”

In practical terms, training campuses can be located in more remote, lower-cost locations where large power supplies are accessible. Inference facilities need to be close to end users, driving a predicted return to urban development.

Net-zero considerations

The environmental implications of AI-driven data centre growth are among the industry's most contentious debates. Utilities in several markets have kept fossil fuel power plants running specifically to meet surging demand from data centres – a direct tension with the sector's net-zero commitments.

Alex is direct about his company's position. NTT Global Data Centers has committed to net zero within its data centres by 2030, within offices by 2035 and across its supply chain by 2040.

“It's not just a sustainability consideration,” says Alex. “It's an economic consideration as well. It underpins commercial deals and also our values as a business.”

Lonnie takes a broader view, arguing that regulation will ultimately determine how – and where – AI infrastructure lands. Countries and regions are already diverging in their approaches, creating what he describes as a landscape of haves and have-nots.

The water ‘myth’

Water use is another area where public perception and industry reality are frequently at odds. Jean-Francois pushes back firmly on the idea that data centres are depleting community water supplies.

“Data centres fixed the water utilisation problem years ago,” he argues, “Nobody stops their car every 20 miles to add water to the engine, because you have a closed-loop system. That's what we do in data centres.”

Jamie agrees, pointing to on-campus water storage, private water supplies and grey water harvesting as standard practice on modern large-scale campuses.

“I genuinely think it's one of those myths that are out there in the industry,” he adds. “Anyone in the large-scale campus business knows it isn't a design problem anymore.”

Construction headaches

The question of construction methodology is equally pressing. Jamie outlines three broad approaches currently in use. Traditional stick-built construction – where a facility is assembled entirely on site – is largely obsolete.

Modular construction, using standardised prefabricated units, offers speed but limited flexibility, and Jamie is clear that it works best only under specific conditions.

The sweet spot, in his view, is prefabricated construction, where components are manufactured off-site and assembled on location. This reduces on-site labour, shortens timelines and allows late-stage configuration.

“Most of the labour is attributable not to the site but to the factory where it's created,” Jamie notes. “You're coming in at the commissioning stage and really connecting the pieces together – and it allows you to make changes closer to go-live than otherwise wouldn’t be possible.”

Jean-Francois agrees that flexibility must be baked into every element of a facility's design. At GreenScale, every component is examined for how it might be reconfigured as technology evolves.

“You have to ask what you can literally ‘Legoise,’” he says. “Because honestly, we don't know what the next GPU generation is going to bring – and then we're back to the drawing board.”

Jamie adds: “The only kind of fungible thing you can really do at the moment is to plan a data centre to an envelope that you know will work from a prefabricated perspective, and that allows you to change the pods later on. That's the only way to really catch this technology shift.”

Can we power the AI age?

The closing question — whether the industry can actually generate enough power to meet AI's exponential demand — drew candid responses. 

Lonnie notes that the UK is technically capable of generating more electricity than it needs. The problem is getting it from where it is produced to where it is needed.

"We've got legacy infrastructure — cables overhead and in the ground — that are past their due date,” says Lonnie. “The infrastructure won't allow that to happen.”

Jean-Francois points to planning and permitting as an equally significant constraint in Europe, contrasting the regulatory complexity of the continent with the relative speed of development in markets such as Texas.

Meanwhile, Jamie tempers expectations about the pace at which demand can realistically be met. Even a 200-megawatt campus stretches the limits of available construction labour. The idea of gigawatt-scale campuses – discussed seriously by some hyperscalers – remains, for now, largely theoretical.

“I think we need a bit of a natural slowdown,” concludes Jamie. “But we're seeing this massive shift to inference and I think we're all going to be talking about getting back to the metro soon, building whatever we can to serve this latency-sensitive demand.”

The AI gold rush, it seems, is not about to end. But the industry building the infrastructure required to sustain it is being forced to think harder, move faster and plan further ahead than ever before.