Superconductors are poised to solve the economic alarm story of our age

Telegraph

30-04-2025

Humans are ever prone to Malthusian scares. The economic alarm story of our mid-2020s is exponential growth in power demand for AI and data centres, threatening to blow apart the world's energy system.
This fear has a broad hold on the energy elites, government planners and the global media. Some embrace it with fervour. This year's CERAWeek energy summit in Houston was a celebration of AI, the godsend that would keep the fossil fuel industry in rude good health for decades to come.
It ignores the radical breakthroughs under way in advanced superconductors, which have no electrical resistance and therefore dissipate almost no energy as the current goes round and round in a closed loop.
I suspect that superconductors will be the next Nasdaq darling once markets grasp the colossal (and benign) implications for the world economy.
Right now, exploding power demand for AI does indeed look terrifying. Data centres consume 26pc of total electricity in Virginia, the spillover effect of the US military complex.
They are breaking the grid in greater Atlanta, where power demand for processing has surged 12-fold since 2022. Data centres are overwhelming the Austin-Dallas corridor of Texas, ground zero in the global race for AI supremacy.
The International Energy Agency (IEA) estimates that the US economy will consume more electricity for data processing than for steel, cement, aluminium, chemicals and all energy-intensive manufacturing put together by 2030.
The rest of the world lags but data centres nevertheless soak up 21pc of Ireland's power. Britain's booming hub around Slough, Berkshire – Europe's largest – needs so much electricity that it is already clashing with Labour's housebuilding plans and its 2030 clean power target.
National Grid expects commercial power demand for British data centres to rise six-fold over the next decade.
But technology never stands still.
Tokamak Energy, in Britain's 'superconducting valley' near Oxford, is already moving the needle with high-temperature superconducting magnets that operate at 20 kelvins (minus 253C), a spin-off from its world-leading work on nuclear fusion.
These are made from ribbons of rare-earth barium copper oxide that are one to three microns thick. They superconduct at high enough temperatures (a relative term) to slash the cost of cryogenic cooling.
'It's absolutely transformational,' said Warrick Matthews, the Tokamak chief executive.
'We don't have losses from copper resistance or heat so it reduces energy use by 17pc in a data centre. We're in talks with several companies already.'
These super magnets can be used for all kinds of industries, from hybrid electric aircraft to the operation of wind turbines.
'Each nacelle on a 15-megawatt wind turbine has three to four tons of rare-earth magnets. Our magic tape can replace 99pc of those rare earths, reducing our reliance on supply from China,' he said.
Further in the future – but close enough to call into question the intoxicating forecasts of the world's gas drillers – is an entire redesign of the way data centres are built.
In crude terms, if you soak semiconductor chips in liquid helium at minus 269C and keep them cold, you may be able to stack them a hundred layers high in a 3D structure – impossible now because copper wires overheat. This could slash power use by orders of magnitude.
The global nanotechnology institute IMEC is developing a system that does exactly this. 'We think we can start building it for real in five years,' said Anna Herr, the scientific director.
'It could raise energy efficiency by a hundred times, and that covers the whole system, including the cost of cooling.'
Prof Herr said we may be looking at a future where data centres are no bigger than a fridge or even a shoebox, so small that they can be built wherever you need them.
Moore's law of semiconductor progress is that the number of transistors on a chip doubles every two years or so.
Superconducting transcends this ever harder task of miniaturisation, offering a way to do it with standard 28 nanometre chips that are relatively cheap.
Cooling helium to minus 269C costs money but it pays for itself once computing speed reaches the threshold of 10 petaflops, or one quadrillion 'floating-point operations' per second.
It is not easy science but the basic idea is that superconducting enables ultra-short pulses. 'We lose 70,000 times less energy with each pulse than conventional computing,' Prof Herr told me.
The world is now in an awkward phase. The computing demands of AI are doubling every six months. Average data centres consumed eight kilowatts per rack in 2022. Nvidia's latest GB200 chip needs nearer 120 kilowatts for training models such as ChatGPT.
This risks an immediate crunch for the global electricity system before the solution is ready.
The IEA says the US accounts for 45pc of global data centre demand, and China 25pc, making the AI revolution very fossil fuel-heavy. It threatens to smash the Paris and Glasgow accords on CO2 emissions and set off a renewed gas and coal boom.
American utilities plan a torrid expansion of gas-fired plants, despite a five-year waiting list for new turbines.
Gas is no longer just a 'bridge fuel' on the way to decarbonised power, said Freeport LNG's Michael Smith at CERAWeek. 'It's the base-load fuel necessary for us to create the electrons for AI.'
This sort of thinking lies behind moves by Shell, BP and others to drop the 'green stuff' and bet their corporate futures on a global gas boom lasting deep into the mid-century.
Donald Trump aims to go further as an ideological end in itself, to the point of forcing the closure of wind farms already far into development.
He has signed an executive order to 'turbocharge coal mining' and aims to double total power output, waving pollution curbs on toxic chemicals such as mercury and arsenic for good measure.
'We're ending Joe Biden's war on beautiful, clean coal once and for all. All those plants that have been closed are going to be opened,' he said.
The IEA says data centres worldwide consumed 415 terawatt hours (TWh) of power last year, or 1.5pc of global electricity output. That will double to 1,000TWh by the end of the decade. It could double again by 2035 under its 'lift-off' scenario.
So, yes, we have a big problem, but it is not as big as urban legend would have it. The IEA spells out in its 300-page report how AI itself will lead to a quantum leap in energy technology and savings, greatly mitigating the effect.
Few believe that China's DeepSeek-R1 matched American AI rivals on maths, coding and reasoning with just a quarter of the power. But its inference model probably uses 40pc less electricity than GPT models. That alone has large implications.
Chris Wright, the US energy secretary, says America and the world are going to need a vast expansion of fossil fuel-fired power to drive the AI revolution. 'Since the demand for intelligence is unlimited, so will be the demand for energy,' he said.
Actually, it won't. He is committing the sin of false extrapolation. The greater probability is that AI coupled with superconductors will in the end slash energy demand.