Nuclear energy is qucikly being recognised as a complementary method of energy generation to renewables. It provides constant baseline power that can easily step in when intermittent methods like solar or wind can’t meet demand. Now, a new generation of smaller, modular reactors could be used to power\u00a0a new generation of eye-wateringly big data centres. <\/p>\n\n\n\n
Abundant nuclear energy could see the next generaiton of campuases built at never-before-seen scales. It’s a potentially necessary step for the industry, as operators look for ways to host burgeoning artificial intelligence (AI) workloads.\u00a0<\/p>\n\n\n\n
Back in 2021, the Ukrainian government needed cash. Maybe someone thought it would be useful to have some liquid funds to hand. You know, in case they needed to buy several billion dollars worth of bullets, bombs and drones some time in the near future. And what better way to get rich quick in 2021 than get into the crypto game? Maybe they just wanted in on the Doge Coin hype.<\/p>\n\n\n\n
As part of its efforts to create new revenue streams for the country, the Ukrainian government decided to invest heavily<\/a> in building some of the largest data centres the region had seen. The sites would cap out somewhere between 250 MW and 500 MW\u2014almost unthinkably large at the time. The plan: use them to mine bitcoin. A lot of Bitcoin. Today, there’s a lot about the story that feels, well, really dated. A peacetime Ukraine, thinking of a 250 MW data centre as “unthinkably big”, and a massive crypto-mining scheme. I have lived a thousand lifetimes since these concepts were new. <\/p>\n\n\n\n One aspect of the story, howver, puts the whole scheme ahead of its time. Had their construction not been derailed by Putin\u2019s invasion a year later, the massive data centres would have been powered by neclear energy.<\/p>\n\n\n\n Built next door to one of Ukraine’s four biggest nuclear generators, the Government\u2019s hope was that \u201cthe data centres would help use idle load and reduce the burden on transmission systems\u201d.<\/p>\n\n\n\n Now, it\u2019s uncertain if the facility will ever be completed. But the idea of using nuclear energy as a source of cheap, clean electricity to fuel increasingly power-hungry data centres has, at the very least, outlived the crypto boom.\u00a0<\/p>\n\n\n\n Driven by the increased need for capacity amid the AI craze, data centre operators are scrambling for ways to cost-effectively and sustainably fuel a new generation of facilities.\u00a0Of course, in a lot of places, it’s not going great; the US is extending the lifespan of coal plants previously destined for the scrapheap, and opening new natural gas fired facilities at a rate we haven’t seen since the 80s. But, at the same time, the country is also reopening decommissioned nuclear plants<\/a>. <\/p>\n\n\n\n At a time when AI data centres are demanding an order of magnitude more electricity than cloud or colocation sites (so much so that sweetheart deals between grid companies and big tech are resulting in mass blackouts and service disruptions in the midwest, as Meta and its ilk scrape the bottom of even that cheap, abundant barrel), it’s clearly not a case of how <\/em>they get the power, but how much can we get? <\/em><\/p>\n\n\n\n “A normal data centre needs 32 megawatts of power flowing into the building. For an AI data centre it’s 80 megawatts,” Chris Sharp, CEO of Digital Realty told the BBC<\/a> in February 2024. \u201cOur industry has to find another source of power.\u201d\u00a0Since then, that figure has risen steadily, and the average AI data centre now sits comfortably in excess of 100 MW. <\/p>\n\n\n\n The increasing demand for clean energy in the midst of an already sluggish green transition from fossil fuels is pushing the data centre sector into an energy crunch. Expanding capacity while moving towards net zero seems logically impossible.\u00a0<\/p>\n\n\n\n Renewables like solar and wind provide huge amounts of power some of the time. However, when the wind dies down and the sun sets, that power goes away. Data centres need a baseline source of reliable, abundant, cheap power. <\/p>\n\n\n\n Increasingly, that seems to be nuclear. This is especially true as small modular reactors (SMRs) grow in popularity. <\/p>\n\n\n\n These smaller, modular reactors are gaining popularity in countries with ageing large scale nuclear capabilities like France and Belgium. <\/p>\n\n\n\n Not only that, but countries tat never built out nuclear infrastructure to a great degree are also exploring the possibility of SMRs. The UK, for example, is strongly supporting SMR adoption. There are plans in place to invest \u00a320 billion over 20 years to develop a fleet of power plants covering up to 25% of the country’s electricity needs. Leading this effort is Rolls-Royce, which aims to deploy 16 of its 470MW SMR generators across the UK. <\/p>\n\n\n\n If SMRs can provide abundant clean electricity, what\u2019s to stop you bolting one to a data centre?\u00a0<\/p>\n\n\n\n In the US, Microsoft recently put out feelers<\/a> to hire a \u201cprincipal program manager\u201d to lead the company\u2019s nuclear energy strategy. In the country\u2019s data centre capital, Virginia, Green Energy Partners (GEP) recently moved a step closer to building a campus powered by SMR technology. The 641 acre plot could eventually house 19 data centres powered by \u201cfour to six\u201d SMRs. <\/p>\n\n\n\n Commercially available SMRs probably won’t hit the market in the US until the end of the decade. Nevertheless, early stage designs recently approved by the US Nuclear Regulatory Commission have already caused a stir in the country\u2019s data centre market. The reactor certified for use in the US by the Nuclear Regulatory Commission generates approximately 50MW of power. The facility also uses a light-water design, requiring a facility about a third of the size of a standard power plant. <\/p>\n\n\n\n The amount of readily available power from SMRs could be a vital answer to the looming question of AI. The technology is already putting a heavy finger on the scales of data centre power consumption. The graphic processing units used to host generative AI consume four times the power of cloud servers. At the current pace of adoption, McKinsey estimates the power needs data centres in the US will jump from 2022’s 17 GW to 35 GW by 2030. <\/p>\n\n\n\n \u201cThat’s going to put such a strain on resources, particularly on power infrastructure, let alone data centre capacity infrastructure,\u201d Dominic Ward, chief executive of data centre company Verne Global, told Reuters<\/a><\/em>.<\/p>\n\n\n\nNuclear powered data centres? <\/h3>\n\n\n\n
SMRs and data centres: a perfect match? <\/h3>\n\n\n\n
AI power consumption goes nuclear<\/h3>\n\n\n\n