Canada Bids for First Leadership-Class Supercomputer

Two Canadian universities have signed a memorandum of understanding to jointly bid for what would be the country's first leadership-class supercomputer, a system aimed at the global top 10. But the federal funding program they're counting on hasn't opened yet, and the gap between ambition and allocation remains wide.

Queen's University in Kingston, Ontario and Simon Fraser University in Burnaby, British Columbia signed the MOU on March 26, committing to a joint proposal under Canada's AI Sovereign Compute Infrastructure Program. The plan: a top-10 system at Queen's in Kingston and a top-25 system at SFU, operated as a distributed national platform serving academia, government, and industry.

As of late 2024, Canada was the only G7 nation without a Top 50 supercomputer. Its highest-ranked system, the Telus Sovereign AI Factory, sits at 78th on the Top500 with 22.74 petaFLOPS. It's an HPE-built cluster running NVIDIA H200 GPUs over Quantum-2 InfiniBand. For a nation that considers itself a serious player in AI research, that ranking is hard to explain away.

The people behind the bid

Ryan Grant, an associate professor and head of the CAESAR lab (Computing at Extreme Scale Advanced Research), spent nine years at Sandia National Laboratories before returning to Queen's. His group of 18-plus researchers, one of the largest exascale-focused labs in the world, developed software specifications now used in all of the world's most powerful supercomputers. "We're toolmakers and tool improvers," Grant told the Queen's Gazette. "We build the big tool that others use."

Then there's Ian Karlin. Until late 2025, Karlin was a principal engineer at NVIDIA and technical lead on El Capitan - the world's number-one supercomputer at Lawrence Livermore National Laboratory - along with two upcoming US systems, Doudna and Mission. He left NVIDIA to join Queen's as an assistant professor in December 2025.

"I was thinking about moving back into research... it seemed like the right time," Karlin told BetaKit. The article noted he joins a growing list of US researchers moving to Canada amid federal research funding cuts and political pressure on universities south of the border. Karlin has since said that "very senior, very big names" have expressed interest in following him north.

Grant and Karlin claim they are the only two people in Canada with direct experience designing and procuring next-generation supercomputers. Whether or not that's precisely true, it's hard to argue with the resumes.

SFU brings operational credibility

SFU brings more than geographic balance. The university operates Canada's largest public supercomputing facility through the Cedar Supercomputing Centre, which serves over 24,000 researchers and industry partners through the Digital Research Alliance of Canada. Cedar's flagship system, Fir, runs 165,888 CPU cores and 640 NVIDIA H100 SXM5 GPUs with more than 50 petabytes of high-performance storage.

More notable is SFU's energy efficiency. Cedar operates at a PUE of 1.07, meaning the facility uses only 7% more energy than the computing equipment itself. The industry average sits around 1.56. For a proposal that will inevitably face questions about power consumption and sustainability, that track record matters.

"By partnering with Queen's, we're bringing together the expertise, talent, and the national-scale facilities needed for a sovereign platform that Canadians can trust," SFU VP Research and Innovation Dugan O'Neill said in the announcement.

The funding question

The proposal hinges on federal money. Canada's 2024 budget committed $2 billion over five years to a Sovereign AI Compute Strategy. Of that, up to $705 million is earmarked for the Sovereign Compute Infrastructure Program - the pot Queen's and SFU are targeting. An additional $700 million is allocated to an AI Compute Challenge aimed at mobilizing private-sector investment, plus $300 million for an AI Compute Access Fund and $200 million for near-term upgrades to existing infrastructure.

Bell Canada signed its own MOU with Queen's in January 2026 to support the Kingston facility and expand capacity at SFU. Dan Rink, president of Bell AI Fabric, called the partnership an effort to "support Canada's leading research institutions by providing the infrastructure needed to operate at national scale."

But the Sovereign Compute Infrastructure Program hasn't opened for proposals yet. It's expected sometime in 2026, and the Queen's-SFU bid will compete against other institutions and consortia for a finite pool. An MOU is a statement of intent, not a funding commitment.

Kingston's location - roughly equidistant from Toronto, Montreal, and Ottawa - is strategically sensible for a national facility. Whether $705 million is enough to build a credible top-10 system is another question. The machines currently occupying that tier cost billions and require sustained operational budgets that dwarf initial procurement.

What to watch

This bid is worth tracking for two reasons beyond the hardware specs. First, it's a test of whether Canada can convert its 2024 budget rhetoric into actual infrastructure. The money was announced two years ago; no major system has materialized yet. Second, the talent pipeline flowing north from the US - embodied by Karlin's move - could accelerate if American research funding continues to erode. Grant's CAESAR lab is already one of the world's largest exascale research groups, and it's growing.

None of this is guaranteed. A signed MOU, a federal program that hasn't launched, and a top-10 target that no Canadian institution has ever hit... there's a lot of distance between here and an operational exascale-class machine. But Canada's supercomputing deficit is real, the team has credible experience, and the money exists on paper. The question is whether Ottawa will write the check.