Canada’s place in the history of artificial intelligence runs deeper than the country’s recent investment announcements. Long before ChatGPT made AI a household subject, Geoffrey Hinton at the University of Toronto, Yoshua Bengio at the Université de Montréal and Richard Sutton at the University of Alberta were developing ideas that became foundations of the field. Hinton and Bengio shared the 2018 Turing Award for their breakthroughs in deep learning, while Sutton shared the 2024 award for his foundational work on reinforcement learning. Hinton’s Toronto team, including Alex Krizhevsky and Ilya Sutskever, also developed AlexNet, the 2012 system that demonstrated the potential of neural networks in image recognition and helped accelerate the current AI era. Canada followed this scientific leadership with an early institutional bet. In 2017, it launched what the Canadian Institute for Advanced Research (CIFAR) describes as the world’s first national AI strategy, supporting research institutes in Toronto, Montreal, and Edmonton and strengthening Canada’s ability to attract leading researchers. That inheritance still matters, although the character of the competition has changed. Advanced algorithms now depend on a physical foundation of specialized chips, enormous amounts of electricity, cooling systems, and the facilities that bring them together. Canada helped establish the intellectual foundations of modern AI. Its next challenge is converting that legacy into lasting technological and economic capacity. What a data centre actually does The physical capacity behind AI sits in data centres: highly engineered facilities containing rows of powerful servers, storage systems, and networking equipment. AI-focused centres connect thousands of specialized processors—most commonly graphics processing units, or GPUs—so they can divide a demanding calculation and work on it simultaneously. These machines carry out two broad stages of AI. During training, they process enormous datasets to build and refine a model. During inference, a trained model applies what it has