David Shipley, founder of New Brunswick-based cybersecurity firm Beauceron Security, says the electricity and water required to support modern AI systems are far greater than most people realize. “AI takes incredible amounts of computing power,” said Shipley. “It requires electricity and it requires cooling. It can do a lot of things, but it comes with a big bill.” Shipley says large data centres that support AI tools can consume as much water as a mid-sized municipality. “A large AI data centre can use as much water as a 50,000 person city, particularly when they use evaporative cooling systems,” he said. He notes that in parts of the United States, where many centres are clustered in rural areas, groundwater resources are already under noticeable strain. While Canada is not currently home to many large-scale AI facilities, he said the environmental risks remain relevant as companies continue to expand. “The only area data centres may even make sense as they scale, is an area rich in water resources and hydroelectric power,” he said, pointing to Quebec as one example. “But even then, it requires a new level of thinking by public policy planners.” Shipley warns that if AI development continues unchecked, the technology’s carbon footprint could be equivalent to adding between five and 10 million cars to North America’s roads. He said that estimate does not include emissions associated with manufacturing the processors and servers required for AI workloads. At the same time, researchers say even smaller AI models require significant amounts of power. Tushar Sharma, an associate professor of computer science at Dalhousie University, said a medium-sized AI model can use the annual electricity of several homes. “Something as small as a medium-sized model uses the power of three American households per year,” he said. “Even the smaller models require a lot of energy, so now you begin to imagine how much the bigger models are consuming.” Cooling those systems, he adds, demands continuous flows of freshwater. “These big machines need to be cooled, and water is running constantly to dissipate that heat,” he said. But Sharma, whose research focuses on reducing AI’s environmental footprint, says solutions are already emerging. He says scientists are developing techniques to make models run more efficiently. For example, by “pruning” them. The method removes unnecessary signals from a model so the system performs fewer computations, reducing energy use. “This is just one example, but there are various optimization techniques companies can adopt,” he said. “Some organizations are also deciding where to deploy their models based on the type of energy available. Setting up a data centre in a cooler area, or in a region powered by greener energy, can make a major difference.” Sharma says wind power in parts of Europe is already attracting companies seeking cleaner energy sources for data processing. Other firms are exploring hydrogen-based solutions. At the user level, he says individuals and organizations should also be thoughtful about the size of the models they rely on. “We should decide whether we really need a very large model,” he said. “Sometimes even a smaller model is sufficient.” Both experts say public policy has not kept pace with the rapid growth of commercial AI tools. Sharma notes that Canada has no specific environmental regulations governing AI data centres, despite the technology’s accelerating electricity and water demands. Shipley argues governments need to hear from researchers, environmental experts and academics - not just major technology firms, before making decisions that could have long-term consequences. “We need to stop being so afraid that we’re going to miss out,” he said. “There are consequences to adopting this technology at scale, and it comes at a tremendous cost.” Sharma says only a small number of researchers in Canada specialize in the environmental impacts of AI. He hopes more companies will fund research aimed at reducing energy use and improving efficiency. “It will save money for companies, and it will help the environment,” he said. “Supporting this research is essential.” Both experts agree AI can be made more sustainable, however, doing so will require coordinated action from industry, researchers, users and governments.
