By Prof. Phil Hart , Chief Researcher, Renewable and Sustainable Energy Research Center, Technology Innovation Institute (TII)
In recent months, climate analysts and media outlets have raised alarms about the rapid rise in energy demand from AI, warning that if left unchecked, it could derail global climate targets. As AI scales at unprecedented speed, questions are emerging not just about what it can do, but what it will cost, in energy and emissions. TII’s Chief Researcher for its Renewable and Sustainable Energy Research Centre, Professor Phil Hart, takes a closer look at what the data actually tells us.
In recent months and years, AI’s power appetite has become a headline magnet. Headlines like ‘AI may kill net zero by 2050!’ reflect growing concern that AI’s resource needs could outweigh its benefits, especially if powered by fossil fuels. That concern is understandable. But before we jump to conclusions, it’s worth stepping back and separating signal from noise.
So, how much energy does AI actually use?
It’s a tricky number to pin down. AI runs inside data centers, and separating out the exact percentage of power used directly by AI from total server load isn’t straight forward. The IEA reports that in 2024 global data centers energy use was about 1.5% of total demand, which equates to about 415TW of power.
Yep, I agree, that’s a whole bunch of power, and with an estimated 11,000 data centres out there, it adds up fast. Some researchers have estimated that AI absorbs a good proportion of that, about 60%, whilst others say it’s closer to 2%. Carbon Brief suggest that AI uses between 5% and 15% of data centre energy demand, putting it at about 0.075-0.3% of global power, which feels like a reasonable middle value range to hold in our heads for now.
So globally, the numbers might seem modest, but zoom in, and things look different.
In the USA, which hosts about 45% of the world’s data centres (nearly 3x the next biggest region), they already consumer about 4% of the country’s power. In Virginia alone, approximately 25% of all state electrical power goes to data centres, and it’s estimated to be more than 10% in North Dakota, Nebraska, Oregon, and Iowa. This points to a key concern; while the global impact might be relatively modest currently, the local impact can be quite significant.
Looking to the future, the IEA predicts that data centre demand could double by 2030, with some forecasters suggesting AI alone could push global electricity demand to 4% or more. Based on various models, by 2050 we could see AI-driven demand reach anywhere from 350 TWh to over 2,000 TWh. That's a wide range, and it is almost impossible to forecast accurately in such a rapidly changing landscape - but the one surety is growth.
But the real wildcard? How much of this demand can be offset by efficiency gains powered by AI itself.
Energy demand is only half the story. The real issue is how we generate the power AI needs.
If AI is powered by clean energy, whether renewables with storage, nuclear or geothermal, its climate impact could be minimal. Better yet, it could become a force for good, driving investment in clean energy, lowering costs, and even repurposing waste heat into local heating networks. That’s a win–win.
The alternative? Relying on fossil fuels - predictable dispatchable power, yes, but also polluting. Scale that up to meet growing AI demand, and the tech becomes the climate villain critics warned us about.
So where do hyperscalers stand?
To date, they’ve been fully behind the first option, building some of the world’s greenest data centres and setting bold net zero goals. But they’re also racing to dominate AI, and that means rapid scale. Data centre build cycles run on 2–3 year timelines. Power infrastructure? Often a decade or more. That disconnect poses a real risk.
In the race to scale, power pragmatism may well win out over principle. If clean energy isn’t readily available, will hyperscalers settle for whatever gets the job done fastest? And if one shifts course, does that open the floodgates for others to follow in trading climate commitments for competitive edge?
There’s a lot riding on how this plays out. AI is already reshaping the world. The question now is whether it powers that transformation cleanly, or at the cost of the very future it promises to improve. Frankly, how committed are they, individually and as a group, to their climate targets? I guess we will wait and see, but it may prove to be one of the defining sustainability questions of the next decade.