Datacentre energy demand set to go nuclear
In this guest post, Patrick Smith, field CTO for Europe, Middle East and Africa at Pure Storage, examines the toll that growing AI use is taking on datacentre energy demands, and finds out what can be done to address it
Throughout the past decade, datacentre power consumption has remained relatively stable thanks to efficiency gains acquired through technology innovation.
That all changed with the rise of artificial intelligence (AI). According to the International Energy Agency (IEA), global electricity consumption by datacentres is projected to more than double by 2030, with AI playing a major role. As an example, a single ChatGPT query consumes nearly 10 times more energy than a Google search.
Goldman Sachs projects global datacentre power demand will reach 55GW by early 2025 and climb to 84GW by 2027 — a rise of more than 50% in just two years. By 2030, capacity could reach 122GW. Power density is increasing too, with usage per square foot projected to grow from 162kW to 176kW by 2027. AI-specific hardware is a major part of this jump.
Can power generation keep up?
The growing appetite for energy in datacentres is colliding with an aging and inefficient legacy infrastructure. In the US, nearly half of the country faces elevated risks of power outages in the next decade, according to the North American Electric Reliability Corporation. That’s due to retiring power plants, delays in new construction, and surging demand.
Globally, the picture is similar. It’s estimated that over half of current power plants in the UK, and EU are expected to shut down by 2040, even as electricity demand is forecast to increase by almost 80% between 2022 and 2050. This is driven by a desire to move away from reliance on fossil fuels such as coal.
Is Nuclear the answer?
Given this looming gap, nuclear power is emerging as a key player in the future energy mix — both to meet demand and to support decarbonisation efforts. The IEA expects nuclear power’s share of electricity generation to nearly double by 2050.
The IEA reports that currently, 63 nuclear reactors are under construction worldwide, with a combined capacity of more than 70GW. However, most of this activity is in Russia and China. In contrast, G7 countries have seen declining investment and continued retirements of aging facilities.
SMRs: A nuclear fit for datacentres
One promising development is the small modular reactor (SMR) — compact nuclear fission reactors that produce up to 300MW, with microreactors as small as 10MW. SMRs are designed to be prefabricated, easier to deploy, and quicker to bring online than traditional plants.
SMRs could be colocated with datacentres to provide dedicated, carbon-free electricity. While conventional nuclear plants take 8–10 years to build, SMRs could be deployed in just 2-3 years. That’s still slower than datacentre construction, but it’s a step in the right direction.
The challenge? SMRs remain largely unproven at scale, with very few in commercial operation – though this may change rapidly in the coming decades as demand intensifies.
The need for immediate efficiency wins
While nuclear power may help solve long-term capacity problems, datacentres also need immediate solutions to curb energy use. This means drilling down on every area of energy consumption in the datacentre from networking, to compute, to data storage, and all of the associated cooling and power requirements. Only by utilising the most energy-efficient solutions across the entire datacentre can businesses hope to mitigate the impact of AI, and energy shortfalls.
One part of the answer could be phasing out inefficient, legacy spinning hard disk drives (HDDs) in favour of all-flash data storage. Flash is significantly more energy-efficient than HDDs, which still account for much of today’s datacentre storage. Unlike HDDs, flash storage has no moving parts and consumes five-to-10 times less power as a result.
While any flash storage is more efficient than HDDs, advanced flash storage systems can offer even greater benefits. Commodity SSDs, often capped at around 20–30TB per drive, can only go so far in reducing energy usage. Some storage manufacturers now offer high-density flash modules designed to work as part of integrated systems. These modules minimise onboard components like RAM and rely on system-level optimisation. The result? They are up to five times more energy-efficient than standard SSDs, have 85% lower carbon emissions, and three-to-six times better reliability compared to both SSDs and HDDs.
As AI workloads grow and datacentre energy demands surge, there’s no silver bullet — but multiple strategies will be needed. Nuclear power, particularly SMRs, could help bridge the supply gap in the long term. In the near term, more efficient datacentre technologies such as all-flash storage offer a clear and immediate path to greater efficiency.