Can AI’s power problem become a clean-energy opportunity?

Artificial intelligence has triggered a new gold rush. This time, companies are competing not for land or minerals, but for access to electricity.

In many regions, data center developers are applying for connection to the electricity grid at a pace and scale that power systems have rarely experienced. New electricity infrastructure cannot usually be planned and built as quickly as a hyperscale (very large) data center.

This has made the time it takes to secure electricity supply one of the main constraints on digital infrastructure. This has also raised a serious concern: will AI absorb already-scarce clean electricity, extend the life of fossil-fuel plants, and make climate goals harder to reach?

These risks are real. But I believe the debate is often framed too narrowly.

The important question is not only how much electricity AI will consume. It is whether this enormous investment behind the AI boom can also help build the cleaner and more flexible power systems society already needs.

In a recent Perspective article, my coauthors and I describe this possibility through the “AI–energy storage nexus”.

The relationship works in both directions. Energy storage can help data centers gain faster access to electricity, reduce pressure on the grid, and better match electricity demand with renewable generation. At the same time, investments by large digital companies can accelerate energy storage innovation and deployment, creating benefits that extend beyond the data-center sector.

A congested highway to power

Connecting a large data center to the grid is like adding a major new entrance to an already congested highway. The road may have spare capacity for much of the day, but not during the busiest hour(s). New lanes and junctions may be required to resolve the traffic jam.

Electricity grids face a similar problem. Grid operators must ensure that a facility’s electricity connection can meet its highest demand, even if that level is only needed for a short time. That peak can determine the size of the connection and the need for wider network reinforcement.

Storage cannot remove the need for the highway, but it can reduce the surge during rush hour.

To illustrate this effect, we analyzed the electricity-use patterns of 96 real data-centers from UK Power Networks. We simulated batteries with different power ratings and storage durations and asked how much they could reduce the maximum electricity import seen by the grid.

For a typical data center, batteries reduced the maximum amount of electricity needed from the grid by roughly 10–15%. That could affect the size of the connection, reduce reinforcement needs, or allow a project to connect in phases.

But we observed a limit. Bigger batteries did not always produce proportionally larger benefits. Data centers often use large amounts of electricity at a fairly steady rate throughout the day. Once batteries have reduced the main peak in demand, adding more battery capacity may bring only modest further benefits.

Batteries can therefore make data centers more grid-friendly, but they are not a silver bullet.

A shock absorber between data centers and the grid

A battery can act like a shock absorber between a data center and the electricity system.

A shock absorber does not make a vehicle lighter or remove bumps from the road. It softens sudden movements. Energy storage plays a similar role by buffering changes in electricity demand before they reach the grid.

For a data center, storage can provide on-site services, for example, backup power, improved power quality, etc. At the grid interface, the battery can smooth rapid changes in demand, reduce electricity use, or supply power during system stress.

This is why storage-backed data centers should not be treated the same as traditional, inflexible large loads. Large technology companies operating AI and cloud data centres have advanced energy management systems, sophisticated computing networks, and strong technical capabilities. They also have a powerful incentive to secure access to electricity as quickly as possible.

That does not mean they should receive preferential treatment simply because they are wealthy. Instead, they could be asked to provide measurable benefits in return for faster access to electricity. This could include reducing demand at peak times, flexible demand, supporting the grid with batteries, or temporarily cutting electricity use when the grid is under pressure.

Running the compute when the grid is cleaner

At home, many people delay running a washing machine until off-peak hours. Some computing tasks can be managed in the same way, running when cleaner electricity is available rather than immediately.

Not all computing can, however, be delayed. Some tasks, such as providing navigation directions, must happen instantly. Others, including AI model training, can often be scheduled for times when more renewable electricity is available or shifted between data centers in different regions.

This flexibility complements battery storage. A battery can handle short and sudden changes, while workload scheduling can shift demand to better times or locations. Together, they can reduce peak demand on the grid, make better use of renewable energy, and reduce the amount of storage needed.

But simply changing when electricity is used is not always better for the climate. If the extra electricity comes from fossil-fuel power plants, emissions could still increase. To deliver real climate benefits, batteries and flexible computing must be supported by more clean electricity and better alignment between when renewable power is available and when it is used.

The other side of the nexus

Most discussions focus on what storage can do for AI. The reverse relationship may be just as important.

The new electricity gold rush is being led by some of the richest and most technically capable companies in the world. Their need for reliable, round-the-clock clean electricity can create early markets for long-duration energy storage and other emerging clean technologies.

These companies can help new storage technologies move from demonstration to commercial deployment. New trends show investment by Google in clean energy storage solutions (CO₂ -based energy storage and iron-air batteries) and securing electricity from the largest solar PV plus battery in the USA.

AI can also help improve energy storage by speeding up the discovery of new materials, predicting how long batteries will last, improving system design, choosing the best locations, and optimizing how storage systems operate. The resulting benefits could spill over to electricity grids, industries, communities, and renewable-energy projects.

Smarter rules, not simply faster approvals

In many jurisdictions, obtaining a grid connection still involves long queues and repeated studies carried out by scarce technical experts. Rather than responding with blanket restrictions or unconditional approvals, regulators should modernize connection rules around measurable system value.

Projects that reduce peak import, bring clean firm capacity, operate flexibly, or support the grid should be assessed differently from inflexible loads that impose the full cost of reinforcement on the system.

Policymakers could also consider removing barriers to co-located storage and clean generation, reward grid-aware computing, and support innovation in longer-duration storage.

AI’s electricity demand is real, and so are the risks. Instead of focusing on how to power AI, we need to think about how to turn today’s rush for electricity into opportunities for cleaner and more flexible power systems needed for tomorrow.

Read the full article published in Energy and Climate Change.

Note: This article gives the view of the author, and not the position of the IIASA Insights blog, nor of the International Institute for Applied Systems Analysis.