Inside the Energy Challenges Facing AI Data Centres

The AI boom has created a strange paradox for the data centre industry.

On one hand, AI promises to optimise everything from grid forecasting to renewable storage and real-time energy balancing.

On the other, the infrastructure powering those ambitions is consuming electricity at a staggering pace, leaning too heavily on gas generation just to keep up.

That tension was explored on the “Energy and AI. Who’s working for who?” fireside chat on Day 2 of Data Centre LIVE: The London Summit.

Led by Dr. Ben Krikler, Head of Energised Futures and Director of Research & Innovation at Centrica, and moderated by Data Centre Magazine's Ben Craske, the session unpacked the tangled relationship between AI innovation and energy infrastructure.

And according to Ben Krikler, the answer to whether AI is helping or hurting the grid is simple: both.

“It’s clearly a hungry beast,” he said, pointing to projections that AI-related energy demand could quadruple over the next decade.

"At the same time, AI could become one of the energy sector’s most powerful optimisation tools, improving forecasting, supporting decarbonisation efforts and helping operators make smarter infrastructure decisions in real time."

Could DC microgrids change everything?

One of the most intriguing developments discussed during the session was Centrica’s work on DC-connected microgrids.

Centrica's Ben Krikler described them as a potentially transformative shift for digital infrastructure efficiency, revisiting the century-old “DC versus AC” debate with fresh urgency.

The logic is straightforward: much of the technology inside modern data centres, from batteries to solar and compute systems, already operates natively on DC power. Converting between AC and DC repeatedly creates inefficiencies.

According to Ben, switching systems within certain boundaries to DC could improve efficiency by as much as 15-20%.

“It’s a sizable benefit,” he explained. The conversation also focused heavily on flexibility.

In reality, flexibility could become one of the defining tools of future grid management.

The ability to shift workloads between locations, delay AI training tasks or dynamically reduce consumption during periods of grid stress may eventually allow operators to secure smaller grid connections and accelerate deployment timelines.

Critically, Ben stressed that flexibility does not necessarily compromise critical operations.

“There’s lots of evidence,” he noted, “that there’s no impact on critical loads.”

Rivers, glaciers and tectonic plates

Perhaps the most memorable moment of the fireside came when Ben explained the mismatch between AI innovation and infrastructure development through his own vivid analogy.

AI, he argued, behaves like a river: fast-moving, unpredictable and constantly changing.

Infrastructure development is more like a glacier, moving slowly over years and requiring enormous investment certainty.

Regulation, meanwhile, resembles tectonic plates.

“They don’t move at the speed you’d like,” he said.

The result is an industry attempting to build long-term infrastructure for workloads and technologies that may not yet exist.

That challenge is now colliding with mounting pressure to connect data centres faster than ever before.

Ben noted that grid connection requests linked to data centres have surged dramatically in recent months, as operators are turning toward on-site generation to bypass bottlenecks.

Is the industry risking a perception problem?

While much of the discussion focused on technology, one of the session’s more unexpected themes was behavioural science.

Centrica’s Energised Futures team combines energy research with behavioural analysis to better understand how communities respond to emerging infrastructure.

For the data centre industry, PR and public perception represents one of the sector’s biggest long-term risks.

“There’s a very big risk that we damage the perception of this whole industry,” he warned, if communities associate AI infrastructure with gas-powered generation and rising energy strain.

He argued that operators do not necessarily need massive outreach programmes to improve public understanding.

Instead, simple engagement, like working with schools, local businesses and communities to explain what data centres actually do, could significantly improve acceptance.

The broader concern is that if operators build fully “islanded” data centres disconnected from surrounding communities, they risk deepening public mistrust.

Ben instead advocated for greater integration, whether through local heat reuse, energy-sharing initiatives or wider community engagement.

Delivering net-zero with AI

Despite the challenges, the session ended on a notably optimistic note.

Ben pointed toward AI’s growing role in areas like battery chemistry, fuel cell innovation and energy storage optimisation as one of the sector’s most exciting frontiers.

“If AI can help us with storage,” he said, “that could have huge potential.”

The future may lie not only in smarter buildings, but in “energy communities”: ecosystems where homes, EVs, businesses and data centres dynamically share energy locally, instead of simply exporting everything back to the grid.

It is a vision where AI doesn't just act as a power-hungry tenant, but as an active participant in balancing the system around it.

Whether the grid can evolve quickly enough to support that vision, however, remains the industry’s biggest unanswered question.