I’ve been working in AI for years – there’s one big problem no one is really addressing

Far from simply powering growth, could AI spell out doom and gloom for climate change – even accelerate it? Emily Sheffield investigates

At the heart of the fanfare over the government’s unveiling of its AI Opportunities Action Plan – 50 recommendations to supercharge the UK as a new AI superpower – were three central planks: the creation of AI “growth zones” in the UK, the huge expansion of government-owned AI computing capacity, including a new supercomputer; and the creation of an AI Energy Council.

But amongst the hype surrounding AI’s transformative powers for national growth and productivity that our prime minister rightly stressed, there was scant attention paid to the hurdles. Namely: the rapacious quantities of energy this hungry technology requires and how that fits globally with our goals to lower emissions.

There was no roadmap for how AI could develop alongside Ed Miliband’s highly ambitious net zero pledges. Could AI even accelerate climate change?

To be fair, this is the task set by their new Energy Council. Without access to cheap and reliable power, AI firms will not choose the UK to invest. And what is consistently underestimated among the enthusiasm for AI is the vast quantities of energy AI sucks up. One academic paper said that training a large language model would produce as much in carbon emissions as New York City does in a month.

What we need to be addressing far more comprehensively are the energy needs of this thirsty technology – this will also have a transformative effect. And is the world ready? Certainly, the UK is not.

Further afield, the speed of growth in AI is such that we have not yet caught up by creating a global measurement of its likely energy use and how fast it may spiral – the concern being it will outstrip the creation of green energy, placing yet more pressure on our current energy sources. What is without doubt is that demand is going to ramp up, but the result of that on carbon emissions remains a total unknown.

The National Grid expects demand from commercial data centres could represent up to 6 per cent of Britain’s electricity demand by 2030, up from about 1 per cent today. The use of large language models (LLMs) requires powerful servers in data centres, needing constant powering and cooling.

The International Energy Agency warned last year: “The combination of rapidly growing size of models and computing demand are likely to outpace strong energy efficiency improvements, resulting in a net growth in total AI-related energy use in the coming years.”

Change is afoot: one example, the EU is about to pass laws that require all but the smallest data centres to publish their energy consumption. But only a joined-up focus on AI-related energy consumption will ensure the emission challenges become clear.

So, it is good news the dedicated AI energy council chaired by the science and energy secretaries is focusing its work with energy companies to “understand the energy demands and challenges which this technology does present”. The aim is to direct support to the government’s mission to become a clean energy superpower – mainly by developing small modular nuclear reactors. Amazon, Google and Microsoft have all announced their investment in their own nuclear energy supplies.

This is where the UK’s immediate problem lies: it will be a decade before we have any modular nuclear reactors up and running. And given the history of local resistance to nuclear energy on the doorstep, it’s going to take a lot of will. Four major US tech companies committed to investing $6.3bn (£5.1bn) in Britain’s data centres.

The government now has to deliver – overriding resistance and our predilection for failed infrastructure plans. Given they’ve already commissioned a six-month review on top of Monday’s AI proposal, I remain pessimistic. Our lack of grid capacity will place the UK’s AI ambitions and energy transition goals at risk.

And other countries will meanwhile power ahead of us. The Middle East has the means, will and control. In the UAE, they are building vast solar energy resources alongside giant data centres in their deserts. And developing their own government-funded LLMs – first Falcon, now Jais through GS4.

There is little proof – yet – that the hyped focus on AI innovation, powered by billions of dollars of investment, is being matched with equal excitement in creating the green energy to supply that AI growth.

That’s not to say it’s being ignored. The biggest cloud companies have net zero pledges and are using AI as a tool to drive down carbon emissions. Machine learning has already been used at DeepMind to reduce Google’s data centre cooling bill by an incredible 40 per cent.

The future is heading towards agentic AI – but its impact on climate change still lies in the balance.