AI boom could move huge data centres into the desert as power grids struggle

The global race for artificial intelligence could see vast new data centres built in deserts as electricity grids struggle to provide enough power for the booming technology.

AI requires enormous amounts of electricity to run the tens of thousands of powerful computer chips, known as GPUs, behind services such as ChatGPT.

But experts warn the rapid growth of AI is beginning to run into a much more basic problem: finding enough electricity to power it.

The International Energy Agency expects electricity consumed by data centres worldwide to more than double to around 945 terawatt hours by 2030, slightly more than Japan uses today.

Power consumed specifically by AI-optimised data centres is expected to more than quadruple.

And the strain is already beginning to tell. The IEA estimates electricity grid constraints could delay around one in five data centres currently planned for construction around the world by the end of the decade.

Now clean energy infrastructure company GIGATONS believes part of the answer could be to stop trying to bring ever greater quantities of electricity to data centres and instead build more of the data centres where energy is already abundant.

Its solution is strikingly simple: send the GPUs to the desert.

Vast solar farms, battery storage and dedicated microgrids could allow AI data centres to generate much of the power they need close to where it is consumed, rather than waiting years for conventional electricity grid connections.

Toddington Harper, founder and CEO of GIGATONS, said: “Instead of waiting years for power to reach the GPUs, we can take the GPUs to where energy is abundant.

“By directly powering AI infrastructure through very powerful microgrids, we can dramatically reduce the time from deployment to the first token.”

The economics are becoming increasingly attractive.

The global average cost of electricity from new utility-scale solar farms fell to just $0.043 per kilowatt hour in 2024, according to the International Renewable Energy Agency, while battery storage project costs have fallen 93 per cent since 2010.

The US Department of Energy has also identified microgrids as a potentially faster way of providing reliable electricity to data centres than waiting for conventional transmission and distribution networks to be expanded.

The idea essentially reverses the traditional model.

Instead of transporting huge quantities of electricity over long distances to giant computing facilities, companies could increasingly locate computing power close to cheap and plentiful renewable energy and transmit the resulting data through fibre networks.

Or, as GIGATONS puts it: “Don’t transport the electrons. Transport the data.”

The company is already pursuing the idea at scale.

GIGATONS announced a global partnership with energy technology giant Schneider Electric in July, with an initial development pipeline approaching one gigawatt of AI computing infrastructure, including major projects in Abu Dhabi and Australia.

Its GIGABLOCK technology combines renewable generation, battery storage and resilient power infrastructure specifically designed to meet the huge electricity demands of AI computing.

The company does not claim every data centre should suddenly move into the desert. Such locations bring their own challenges, including cooling equipment and securing sufficient water.

But it argues the AI race is rapidly becoming an energy race as governments and technology companies compete to secure both computer chips and the electricity required to run them.

Mr Harper said: “The global AI race will increasingly be determined by two things: access to GPUs and access to power.

“Bring those two together and you create a faster route to sovereign AI capacity and, ultimately, a faster route to tokens.”