As artificial intelligence transforms industries, economies and the nature of work itself, one question looms large: How will we power it all?  

AI data centres are power hungry.  

“People equate search engines like Google with AI queries, but it is not the same at all,” explains Michael Welland, associate professor of engineering physics.  

“Unlike search engines, which simply point users to sources, AI systems synthesize information before delivering results – a process that uses significantly more power.” 

AI needs power in two ways. First, there’s the initial training of data, which requires a massive one-time energy investment. Then there are the ongoing queries, which are the daily questions and tasks that now account for about 80 per cent of the power used. 

A basic Google search might use about 0.3 watt-hours. An AI query can range from 0.3 to 3 watt-hours, depending on the complexity of the query, the depth of research and the AI model used.   

“Tasks like deep research or image generation can take up to 40 Wh. That is enough to boil a cup of water,” says Welland.  

With energy demands rising, it makes sense that big tech companies, like Google, Microsoft, Meta and Amazon are exploring nuclear power as a long-term solution.  

“AI data centres don’t sleep. They run around the clock and that means they need a constant, stable supply of electricity. That’s where nuclear power stands out,” says Markus Piro, associate professor of engineering physics.  

“Unlike solar or wind, which are intermittent, nuclear provides what we call baseload electricity: Steady, reliable power 24/7. For AI companies with net-zero commitments, that low-carbon reliability is especially attractive.”  

Currently, most of the focus on AI and nuclear is in the U.S., where energy security is a growing concern.  

“Most AI centres are in the U.S., and the companies operating them are looking to partner with nuclear providers to power them. That’s not the conversation in Canada – we don’t have anywhere close to the number of AI centres here,” says Piro.   

But it’s an evolving situation.  

A recent report from the Electric Power Research Institute (EPRI) shows how fast energy demand is growing.  

Training a single frontier model is projected to require four to 16 gigawatts of power by 2030, which is enough energy to power millions of homes.  

“These models are driving the need for massive, stable energy sources, which is why nuclear power is increasingly part of the conversation,” says Welland.  

“Without changes to how we generate and deliver electricity, the growth of AI could outpace our ability to support it.”