AI and the coming power surge

In my role as a principal client partner for Energy clients, I’ve seen first hand how electricity demand has stayed pretty steady (in advanced economies) even as the digital service economy grew, cloud computing took hold and digital devices transformed daily life. The efficiency gains, from technologies like LED lighting through to hyper-scale data centres, largely offset new consumption keeping it broadly flat through the late 2010 and into the 2020’s. For the utilities sector, whilst demand forecasting was nuanced and intricate, it was in a word, manageable.

However, that era is  over. The world is now scrambling for power amid a twin surge in demand. 

The first element of this is the expected one: the drive to electrify transport, heat, and industry in pursuit of net zero. The second, far more unexpected, is the artificial intelligence boom. It’s a software revolution whose appetite for electricity is pushing energy systems to their limits. 

Technology and code are colliding with an industrial story about steel, copper and the physical constraints of a grid built for our grandparents generation.

Let’s explore in more detail, the coming change.

—

The surge we saw coming: climate change & electrification

The first surge was deliberate and expected. To try and confront climate change, the strategy has been clear: electrify everything!

Cars, heating, and industrial processes are shifting from burning fossil fuels to drawing power from the grid, which can then be decarbonised with renewables and nuclear energy.

According to the International Energy Agency (IEA), the global stock of electric vehicles is set to more than triple by 2030. Heat pumps, the electric successor to the gas boiler, are also expected to triple in number by the end of the decade.

This represents a massive, planned expansion of electricity use, yet one for which grids were not originally designed. The real shock to the system, however, came from elsewhere and is about to go stratospheric.

The surge we did not see coming: artificial intelligence

For years, data centres were a story of efficiency. Even as streaming and cloud usage expanded, total electricity demand remained remarkably flat.

Generative AI has broken that equilibrium. The true cost of artificial intelligence is not only the energy required to train a model, like GPT-5, but also the power consumed by every subsequent question it answers, known as inference. Each time an AI drafts an email, writes code, or composes an image, it consumes vastly more energy than a standard web search.

The numbers are striking. The IEA now projects that global electricity use by data centres will more than double to around 1,000 terawatt-hours within the decade, with AI the dominant driver. That’s roughly equivalent to the entire annual electricity consumption of Japan, a country with 124 million people.

The pressure is also highly local. In Ireland, data centres consumed 22 per cent of metered electricity in 2024, up from 5 per cent in 2015. In Virginia, the world’s largest data centre market, the market leader, Dominion Energy, reports that data centres account for approximately 26 per cent of its electric load. The new gold rush is not merely for land, but for land with a very large power plug.

The illusion of a bubble

Sceptics might argue that AI’s energy hunger will fade when today’s froth subsides. That misreads the demand. The persistent demand comes from the largest platforms Microsoft, Google, Amazon and Meta, and their enterprise clients embedding AI into core operations. If a start-up fails, the workload does not vanish; it migrates to a hyperscaler. The electrons are still required. Not to mention the demand from the general public, where hundreds of millions of people regularly use AI systems.

Innovation versus infrastructure

A more serious counterargument focuses on innovation. Similar alarms were sounded in the early 2000s, when analysts warned online video would overwhelm the grid. Instead, global electricity demand remained almost perfectly flat for two decades.

Innovation will again help. Nvidia’s Blackwell platform targets up to 25-fold gains in energy efficiency for AI inference versus the prior generation, a reminder that we are still early in the hardware curve. Software will also evolve from giant general-purpose models to smaller, specialised ones that achieve more with less power. Yet even with such advances, one bottleneck remains: the grid itself.

The physical bottleneck

The electricity grid is a 20th-century marvel, but it is ill-suited to 21st-century demands. In the UK , a long queue for connections has delayed projects by years, prompting reforms intended to cut waits that in some cases stretched towards a decade. In the United States, federal and industry assessments note that around 70 per cent of transmission lines are at least 25 years old and much of the network is operating beyond its intended life

This is now as much a political and regulatory challenge as a technical one. As Stripe co-founder John Collison recently argued in The Irish Times, countries must rediscover the will and capacity to build.

Building smarter demand, not just bigger supply

Supply must expand, but demand must get smarter. As technologists, we should treat inefficient code as environmental debt. Systems need to route workloads to regions with the cleanest, cheapest power in real time, schedule deferrable tasks away from peak hours, and optimise the data we share with AI systems. Recent disclosures suggest median LLM prompts often consume materially more energy than a web search, though newer models and hardware are narrowing that gap in some cases, another reason to prioritise efficiency throughout the stack.

The new energy race

The AI genie will not return to its bottle. The task is not to halt progress, not that we even could, but the goal is to power it responsibly by matching digital ambition with physical capacity.

The next great technology race will not be won by algorithms alone. It will be decided by those who can build grids as quickly as they build models and who remember that the cleanest watt, or byte, is the one never consumed.

—

I hope this article offered a fresh perspective on the future of energy.

At Waracle, we’re passionate about bringing intelligent, intuitive digital products to the energy sector. Our deep industry expertise makes us the perfect partner for ambitious Energy and Utilities clients across the UK.

Get in touch with our team to find out more.