The artificial intelligence (AI) revolution may be transforming industries, unlocking efficiencies, and revolutionising daily life — but it’s also causing an unanticipated strain on the world’s power systems.
According to a sobering new report from the International Energy Agency (IEA), the explosive growth of AI technologies is fuelling a dramatic surge in electricity demand, with data centres emerging as the chief culprits.
What began as a race for innovation is fast becoming an energy arms race. As tech giants such as Microsoft, Google, and Amazon compete to build the largest, fastest, and most capable AI systems, they are simultaneously erecting vast new server farms to support these ambitions. These data centres — the beating heart of AI — consume prodigious amounts of electricity, much of it to keep the machines cool enough to operate.
The IEA estimates that global electricity consumption from data centres, cryptocurrencies and artificial intelligence could more than double by 2026, jumping from 460 terawatt-hours (TWh) in 2022 to over 1,000 TWh. That is more than the total electricity demand of Japan, the world’s third-largest economy.
And yet, the public debate around AI has barely touched on its energy footprint. Much of the discourse has centred on existential risks, automation fears and regulatory frameworks — not the sheer amount of power required to teach a machine to write a poem or generate a synthetic video.
The irony is palpable: AI was supposed to herald a smarter, greener, more efficient future. It still might — in theory. In sectors such as logistics, transport, and manufacturing, AI-driven optimisation promises significant energy savings. Algorithms can shave fuel use, reduce waste, and streamline production lines. But those benefits risk being dwarfed by the voracious power needs of the systems themselves.
Consider this: a single ChatGPT conversation may seem innocuous, but each query triggers computations spread across hundreds — sometimes thousands — of graphics processing units (GPUs) housed in energy-hungry data halls. Training large AI models consumes even more. According to some estimates, developing a single large language model (LLM) can use as much electricity as 100 American homes consume in a year.
This phenomenon, known as “rebound effect,” is not new. Technological efficiency often leads paradoxically to greater overall consumption — the more efficient a process becomes, the more it is used. The promise of AI as an energy saviour may well fall into the same trap.
Governments and regulators are only beginning to grapple with the implications. The IEA warns that electricity grids in many advanced economies are already under strain from electrification, the phase-out of fossil fuels, and the integration of intermittent renewables such as wind and solar. Add surging AI demand into the mix, and the risk of grid instability looms large.
In the United States, power utilities in states like Georgia and Virginia have raised concerns about local grids being overwhelmed by new data centre projects. In Ireland, data centres already consume 18 per cent of all electricity — and the regulator has introduced a moratorium on new connections. Meanwhile, in the UK, projects to connect data centres in West London have been delayed due to insufficient grid capacity.
It’s a geopolitical issue, too. AI supremacy is becoming a strategic goal for global powers, and electricity is the silent engine behind it. Countries with abundant, reliable, and cheap power — often fossil-fuel-rich states like the United States, China and Saudi Arabia — may gain a competitive advantage. Europe, with its ambitious decarbonisation targets and less elastic grid, could be left struggling.
So what is to be done?
The IEA calls for a two-pronged approach: efficiency and transparency. AI developers must adopt more energy-efficient architectures, from chip design to software optimisation. Data centres should be built where renewable energy is abundant and stable — in Iceland, for example, or Northern Canada — rather than on overstretched urban grids. Policymakers must also insist on better reporting of energy use from tech companies, who are often secretive about their true consumption figures.
Above all, the conversation must shift. As we charge headlong into the AI age, power — not code — may become the ultimate limiting factor.
The future may well be written in ones and zeroes. But it will be powered, quite literally, by megawatts and gigawatts.
Main Image: By Varistor60 – Own work, CC BY-SA 4.0, https://commons.wikimedia.org/w/index.php?curid=59368531
