Why storage is key for AI data centres

There is no underestimating the importance of artificial intelligence (AI). From taking billions of jobs to creating trillions of dollars of value a year, AI will likely touch almost every area of life. And one area already feeling the effects is the global market for power. 

AI’s remarkable feats of analysis and creation require vast amounts of data to be processed. And as the data centres carrying out this work flourish, they are putting a strain on electricity grids.

In the US, for instance, data centres were already sucking up more than 4% of the nation’s electricity in 2023, and this level was due to triple by 2028. According to real estate services company JLL’s 2025 Global Data Center Outlook, data centre capacity is set to grow 15% a year up to 2027—yet will still lag demand. 

"Billions of dollars have been invested in AI over the last couple of years, driving demand for more data centre infrastructure," says the outlook. "Power infrastructure bottlenecks are a major impediment to data centre development,” it adds. 

“Power scarcity garners most of the headlines, but equally as significant are the extended timelines required to build transmission lines. These challenges will continue to intensify as the data centre sector expands rapidly into new geographies."

The AI power grab comes as grids face growing demand from the electrification of transportation and heating, not to mention challenges related to aging transmission, distribution and thermal generation infrastructure. And data centres cannot make do with any old power. 

Instead, they require round-the-clock electricity with good power quality and high reliability. And in many cases, the companies buying the power—which include household-name tech behemoths such as Amazon, Google and Meta—have corporate decarbonisation strategies to consider.

These characteristics have led to a race for firm sources of clean (or at least not-so-dirty) power. Iceland, for instance, is promoting itself as a data centre haven thanks to its abundant hydro and geothermal energy supplies. AI companies, meanwhile, are embracing clean power generation with investments in renewables.

Amazon, for example, was the world’s largest corporate purchaser of renewable energy last year. Microsoft is investing $9 billion in 5 GW of US renewables over in half a decade. And Meta is supporting the construction of 600 MW solar plant in Texas through a $900 million power purchase agreement (PPA) with developer Enbridge. 

The problem with investments in intermittent renewables is that they can only go so far in satisfying data centres’ need for continuous power. To supply the world’s data centres 24 hours a day would require a massive overbuild of generation and transmission capacity if done using wind and solar alone. It would not be practical. 

Hence, data centre operators are looking for alternatives. Clean baseload generation assets such as hydro and geothermal would fit the bill nicely but are highly dependent on geology and—outside of places such as Iceland—hard to come by. So, AI companies are searching further afield. 

One area that is taking off in the US is gas generation. This is clearly not clean—and may in fact be massively polluting if the full effects of methane losses are taken into account—but gas turbines could in theory operate with lower emissions if powered by clean hydrogen or coupled with carbon capture and storage. 

Another focus is nuclear power, which ticks both the baseload and low-carbon boxes. In some cases, data centre operators are signing PPAs with nuclear plant owners or even restarting mothballed reactors. 

But most announced projects involve small modular reactors (SMRs), an emerging technology that has yet to achieve commercialisation, and some tech giants are even exploring fusion, which has not even reached prototype stage yet. And here’s the rub.

Despite an executive order for at least three SMRs to start achieving criticality in the US by mid-2026, and a long list of eager customers and technology providers, an optimistic estimate for widespread SMR commercialisation would be the early 2030s at best. And there are reasons to suspect SMRs may never reach the mainstream. 

Fusion power plants, meanwhile, are unlikely become a reality before 2045, if ever. Even gas plants, which represent a mainstream technology and can in theory be built in around four years or less, face a seven- or eight-year delay for turbines. 

This means companies hoping to power their data centres using thermal power could be in for a long wait… something most AI businesses cannot afford in the current hyper-competitive market. There is an alternative, however. Developers can have a modern solar plant up and running in less than two years.

And to solve the intermittency problem, engineers at Abu Dhabi-based developer Masdar are planning something that was hitherto thought well-nigh impossible: tying solar to enough lithium-ion battery capacity for round-the-clock operations.

This feat is due to take place over the next two years in the United Arab Emirates and should see a 5 GW solar plant being backed with more than 19 GWh of battery storage, making it by far the biggest solar-plus-storage project in the world—and the world’s biggest AI supercomputer cluster, too. 

“This will transform renewable energy into baseload energy,” said Sultan Al Jaber, the chair of Masdar, in a report for the Financial Times. 

Not to be left behind, Saudi Arabia is also planning a massive data centre complex, called DataVolt, which will be powered by renewable energy. A press release from property developer Neom suggests green hydrogen could be used to back up wind and solar power at the 1.5 GW facility from 2028. 

Such projects are viable in Middle Eastern nations because of their high levels of solar and wind resource and their low cost of capital. But if batteries or hydrogen can be used to deliver baseload green power in the Middle East then it could be the beginning of the end for AI-linked thermal generation elsewhere. 

After all, AI companies do not care where they get their power from. And if the options are to have 100% clean electricity from established renewable and storage technologies in a couple of years or wait until the next decade for assets that carry a big carbon footprint or might not even work, which would you choose?