---
The idea of harnessing the immense tidal power of the Severn estuary has been around for a long time. Backers of the current proposal for a semi-circular barrage at Minehead claim it could generate an impressive 2.5GW of electrical power for 120 years. This represents roughly two thirds of the projected output of nearby Hinkley Point C nuclear power station, but for twice as long, and at one third of the cost. It could power two million average homes for over a century.
Yet a recent article promotes the project not in these terms, but as a way of "harnessing renewable energy for the UK's AI boom". Apparently this is now the country's most pressing electricity requirement. Running and cooling endless racks of powerful computers certainly takes a lot of power. Estimates vary, but it has been calculated that data centres already consume over four percent of the electricity generated in the UK, and that this will treble to fourteen percent by 2035, largely due to increasing use of so-called artificial intelligence.
It's often said that that firing up ChatGPT uses around ten times as much energy as just Googling the same question. This is broadly true, but since Google results pages now themselves include AI-generated "summaries", it's not that simple anymore. Other search engines (not to mention books) are available. But general internet browsing, especially for shopping, increasingly triggers background AI systems. More broadly, much of the power is consumed in "training" AI models, before any normal person ever sees or uses them.
The current UK government sees AI and its associated infrastructure as a magical engine of economic growth, and is frantically prostrating itself before those who want to build it here. Planning regulations are being amended to ensure that power, land and water are swiftly made available wherever the big US corporations want them.
The Magic Power Tree
All this is pushing hard against the decarbonisation of electricity generation. Companies such as Google, Meta, Amazon, Microsoft, and Apple acknowledge they are not on course to hit emissions targets set before the AI boom took off. Their preferred get out of jail free card, enthusiastically endorsed by compliant governments, is to go nuclear.
In the US, decommissioned nuclear plants, including at Three Mile Island, are being brought back online to power data centres. New ones are also being commissioned. Even under Biden, the US Energy Department was citing AI as a key driver of a proposed tripling of US nuclear capacity:
Not to be outdone, Trump has instead ordered a quadrupling (from 100GW in 2024 to 400GW by 2050), to be achieved in part by raising radiation limits and loosening regulation.
The UK is also relaxing rules on nuclear power station siting. During Trump's visit last September, Keir Starmer hailed a new partnership agreement with the US as bringing "a golden age of nuclear that puts both countries at the forefront of global innovation and investment". The press release also included a revealing perspective from the US team:
One key project will be "the UK's first nuclear-powered data centre", at Cottam in Nottinghamshire. This will use SMR-300 "small modular reactors" supplied by US company Holtec. Licensing for this new reactor type will be fast-tracked, as the new "partnership" requires UK regulators to accept US safety assessments rather than conducting their own from scratch. It is unclear how this will affect their consideration of the fact that SMRs are likely to produce proportionally more radioactive waste than existing larger reactors.
Invasion of the Data Centres
Despite the unseemly rush, new nuclear cannot possibly arrive fast enough to feed even existing "hyperscale" data centres, which are already hungry for power. Some routinely deploy ranks of back-up diesel generators, and some operators want to build on-site gas power stations. Yorkshire's Drax biomass power station, Britain's largest single carbon emitter, has spotted the opportunity and plans to host a vast data centre within its own grounds, ensuring consistent and maximised demand for its controversial product.
Of course the UK is only a small part of the picture. Vast AI data centres are supposedly going to be ubiquitous. Globally, the quest to reduce cooling bills means many are already located in extreme Northern latitudes. Antarctica may be next. Ambitious plans are afoot to site them (and their power supplies) deep under the oceans, in orbit, and even on the Moon. More prosaically, there is considerable interest in building data centres on remote military bases where the requisite security already exists, and scrutiny is minimal. This may well already be happening: few people would know.
Beware the Bubble
Most available information about the future trajectory of AI is hype, designed to further boost companies' already astronomical share prices. This is a classic bubble, and it will inevitably burst. When it does, some big players may collapse, but others will probably be bailed out by corrupt and captured governments. The 2008 rhetoric of "too big to fail" is likely to return, followed by escalating discontent which will in turn be ruthlessly weaponised by the "anti-elite" right-wing populists who enabled the whole mess in the first place.
This probably means that some of the wackier ideas about where to put data centres will never come to fruition. But the bursting bubble will not lead to AI going away. It will continue brutally reshaping modern life, from education, to commerce, to medicine, to the arts, to warfare, solving some technical problems but also centralising control in ever-fewer hands, and leaving destroyed livelihoods and broken social bonds in its wake.
Datapocalypse?
Some claim that profligate misapplication of AI may even destroy the very data it feeds on. Old steel, dating from before the lunatic era of atmospheric nuclear testing, is prized for use in scientific instruments because it is not contaminated by elevated background radiation. Pre-AI data may come to be similarly precious, because it was carefully created by real people and is unpolluted by the AI-generated slop that makes up an ever-increasing percentage of internet content.
There is no plan. Corporate self-interest and market forces are operating in their usual catastrophic way. Silicon Valley revels in moving fast and breaking things, but only because it makes so much money by doing so. Around the world, polling shows that despite the hype, people are on average significantly more concerned than excited by AI. No-one has a clue where all this will lead, other than to massive profits for the broligarchy, and perhaps to some very big problems when the power goes off.
Profitability is arguably the worst possible measure by which to evaluate new technologies and decide how to use them. Just as with nuclear, responsible and transparent public-interest governance of AI would produce very different policies and priorities. It is surely not beyond the wit of humanity to establish a principle that powerful and resource-intensive technologies should be deployed only where they will actually achieve something useful.
Another sensible rule might be that no AI system should ever be permitted to pass itself off as a human being, or as any other kind of sentient creature. Artificial Intelligence is a deliberately misleading and hyperbolic term. No matter how great the computing power, there's no intelligence involved, just very advanced calculation. AI computer networks crunch unimaginable numbers in order to give the questioner an answer that looks like one a real human being might come up with, given sufficient time and data. That's quite a feat (when it works), but it has nothing to do with intelligence, sentience, or consciousness.
Do we humans perhaps also just produce answers that look like real ones? In a sense this is exactly what socialisation does to us. But though we may sometimes doubt each other's intelligence, this is a matter of degree. Our status as living biological beings is not in doubt. We know what a person is. Conscious robots (and dilemmas about how to treat them) make great sci-fi, but they do not walk among us. Healthy scepticism remains the intelligent response.
---