Individual data-centre projects exceeding 1 GW. That was one of the figures that stayed with me after the COP31 Women in AI & Energy Transition Dialogue in Türkiye. For perspective, Denmark’s peak electricity demand is around 7 GW, meaning a single data-centre project could require power equivalent to almost one-seventh of the country’s peak demand.

 

Perhaps the most obvious question is: where should all that power come from?

 

It is a question that absolutely needs to be answered. But the scale of demand also raises a broader energy-market question: how do we integrate data centres efficiently into the power system?

 

AI and cloud computing are adding significant new electricity demand at a time when power systems are already being reshaped by electrification and growing volumes of intermittent renewable generation. That creates an obvious challenge for grids and generation. But sustained demand growth can also help underpin investment in new energy infrastructure.

 

Bringing those two developments together requires us to think about data-centre energy as an optimisation problem. There are two sides to that equation: how the electricity supplying a data centre is sourced and managed, and how the data centre itself can interact with the energy system.

 

Managing the residual position

A large data centre needs significant volumes of electricity and, critically, a high degree of reliability. Renewable PPAs can form part of the answer, but signing a PPA does not make the underlying energy equation disappear. The generation profile of a wind or solar asset will rarely match a data centre’s consumption profile exactly, leaving a residual position that still needs to be managed.

 

Rather than looking for a single contract that solves the entire requirement, the energy solution can combine different instruments. Depending on the project and market, that might mean renewable PPAs alongside shaped or firm supply, battery storage, wholesale market access and active portfolio optimisation.

 

The larger the load, the more material that residual position can become. It is also why energy has to be considered early in the development of a data centre. Grid capacity, connection, available generation and the structure of the power supply are not separate procurement questions. They are all inputs into the same optimisation problem.

 

Optimising the other side of the equation

If we look at a data centre purely as a fixed load, the challenge becomes one of finding enough generation and grid capacity to serve it. But parts of that load can potentially be flexible, and we are already seeing what that can look like through our work with Google in Belgium.

 

When the Belgian electricity system is under pressure, Google can shift non-urgent computing tasks from its local data centre to other times and locations. Centrica Energy’s optimisation algorithm can dispatch that flexibility alongside battery storage at the site, allowing both computing load and physical storage to respond to conditions in the electricity system.

 

The important point is that demand-side flexibility does not require compromising a data centre’s core operations. It is about identifying the computing load that can be shifted in time or location, while protecting critical workloads and operational reliability.

 

At the scale at which data centres are now being developed, even a portion of flexible demand can become meaningful to the wider power system. A data centre does not necessarily have to be treated solely as a large, inflexible load. Where operational requirements allow, computing load, battery storage and market participation can turn parts of that demand into a source of flexibility.

 

Designing both sides together

This was what made the discussion at the COP31 Women in AI & Energy Transition Dialogue particularly interesting. It brought together people working across two industries that increasingly depend on one another, but have historically approached infrastructure from very different starting points.

 

What’s clear is that the energy question around AI and data centres is bigger than simply finding enough electricity. We need to get much better at optimising the relationship between digital infrastructure and the power system.

 

On the supply side, that means combining generation, storage, markets and different contractual structures to manage the residual positions created by very large loads. On the demand side, it means understanding where data centres can move beyond being passive consumers and provide flexibility back to the system without compromising their core operations.

 

With individual projects now reaching the scale of large power stations, getting that interaction right increasingly matters not only for the economics of the data centre, but for the energy system around it.