AI is Rewriting the Data Centre Rulebook, Moving Bottlenecks

At Data Centre LIVE: The London Summit, a  panel took stock of how Gen AI is colliding with the physical realities of building and running data centres. 

The discussion, “The AI Data Centre Debate,” brought together Lonnie Salmon, Senior Director of Jabil; Jean-François Berche, CTO of GreenScale; Alex Bennett, CEO of NTT Global Data Centres; and Jamie Allen, Head of Site Selection at Iron Mountain. 

Their verdict was the industry has never moved faster or been less certain about what comes next.

“The rules have been rewritten and continue to be rewritten every three months,” said Alex, capturing a mood shaped by surging AI demand, supply volatility and mounting public scrutiny.

Supply chains: from cost focus to end-to-end orchestration

Lonnie argued the era of just-in-time efficiency has given way to systemic orchestration. 

Energy availability now tops the risk list, followed by semiconductors. 

He flagged two- to three-year lead times for critical industrial gear at the site level –  delays compounded by geopolitics. 

One example raised onstage was a significant share of global helium supply coming from Qatar, and how bottlenecks can ripple into high-density semiconductor production. 

“If you don’t have the helium, then how do you produce the high-density memory, the high-speed processing?” Lonnie asked.

Alex added that internal execution is just as binding as external constraints. 

“Power and supply chain are massive constraints for our business,” he said. “But how we evolved as data centre operators and global businesses is just as important to examine.”

Net zero: ambition meets compromise

Reconciling the AI boom with climate goals is forcing hard choices. 

“There’s too much money in play,” Lonnie said, predicting uneven regulation as governments balance sustainability with inward investment.

Jean-François countered that purpose-built facilities remain the most efficient way to compute: “Data centres are big fridges and they’re super efficient. If you don’t like it, fine. 

“Let’s go back to the model where you have a computer room in your building… It’s going to be super inefficient.”

Alex pointed to concrete targets at NTT with net zero within its data centers by 2030, across offices by 2035 and through its supply chain by 2040. 

Meanwhile, Jamie distilled the operational reality to a single word: compromise. 

That can mean interim gas generation with carbon capture or paying for backup power while waiting years for a grid connection. 

“There’s not really much kind of in between,” he said.

Water: engineering solved it, perception hasn’t

Water use remains a hot topic, but Jean-François argued the industry “fixed water utilisation years ago,” citing closed-loop cooling and on-campus systems.

He contrasted data centres with recreational uses, saying: “Fourteen golf courses use more water than one data centre. We don’t talk about this.” 

While that reframes consumption, some estimates put a typical golf course at roughly 200 million gallons annually, underscoring how context matters.

Jamie called large-campus water constraints “a myth” for operators who have engineered around it with storage, private supplies and grey-water harvesting. 

When challenged on regional pushback in the United States, Jean-François acknowledged local complexity but stood firm, stating: “There is no data centre that is using water at the expense of the community.”

Power: generation isn’t the ceiling, interconnection is

The panel’s closing question focused on if the industry can marshal enough power for exponential AI demand.

For the UK, Lonnie’s view was clear: “The UK can generate more electricity than it ever needs. The problem is not generating the power. The problem is getting it from point A to point B.”

Permitting, planning and regulatory friction are the chokepoints, said Jean-François, who contrasted European timelines with the faster path in parts of the US. 

“You go to Texas, you do a hole in the ground, you get gas, you power generation,” he said. “Boom, done.”

Jamie agreed.

“We definitely have enough generation… It’s just so strenuous to build that route down to where we actually need that power,” he said. 

The takeaway

AI is accelerating every pressure point at once: supply chains, siting, interconnection, sustainability and public perception.

As Alex put it, the rules keep getting rewritten. 

For operators and their partners, the mandate is clear: orchestrate supply end-to-end, build grid connectivity as a first-class product, plan for net-zero realities, and communicate the true footprint of modern designs. 

In the AI era, execution is the differentiator.