Compute Capacity

Canadian AI compute capacity spans three distinct tiers: hyperscale cloud regions operated by Amazon, Microsoft, Google, and Oracle; federally-funded academic high-performance computing operated by the Digital Research Alliance of Canada; and a growing colocation and sovereign-compute layer serving startups, enterprises, and government buyers who want Canadian-domiciled infrastructure.

At the hyperscale tier, AWS, Microsoft Azure, and Google Cloud all operate Canadian regions anchored in Toronto and Montreal, with AWS having added a Calgary region to its footprint. Each carries the standard suite of AI accelerators — NVIDIA H100, H200, and more recently B200/Blackwell GPUs, Google TPUs, and Trainium/Inferentia on AWS. Capacity allocations to Canadian customers compete with global demand, which means scheduled GPU availability has been a recurring bottleneck since 2023.

Academic compute runs through the Digital Research Alliance's national systems. The older generation — Cedar, Graham, and Béluga — has been retired and replaced by a new fleet including Fir (SFU), Rorqual (ÉTS/Calcul Québec), Nibi (Waterloo), and the Trillium system. Institute-specific clusters at Mila, Vector, and Amii sit alongside. The federal Pan-Canadian AI Compute Strategy, first announced in late 2024 at roughly $2 billion and expanded to approximately $2.4 billion through 2026, committed significant capital toward expanding this tier, with a fraction earmarked for a sovereign compute capability available to Canadian-headquartered firms. In April 2026, ISED opened intake for the AI Sovereign Compute Infrastructure Program — the procurement vehicle that will actually commission new Canadian supercomputing capacity.

Sovereign compute has become the defining policy question. The argument for building dedicated Canadian AI supercomputers — rather than renting hyperscaler capacity — turns on data residency, export-control exposure, and strategic autonomy. The counter-argument centres on cost per FLOP, speed to market, and whether domestic operators can actually secure near-term GPU allocations in a market where the largest US hyperscalers have locked up most of Nvidia's forward production.

Power and location now drive where new capacity lands. Hydro-rich Quebec and Manitoba dominate new colocation builds, and the Alberta government has moved aggressively to position the province as an AI data centre hub using natural-gas power and cold climate. Expect land acquisition, grid connection queues, and municipal planning approvals to become the binding constraints on capacity growth through the late 2020s.