Smaller AI startups are getting crowded out in the compute race. The feds want to ensure they’re at the table.
While the Government of Canada’s planned US Qualified Small Business Stock-style tax incentive has hogged much of the post-Budget 2026 spotlight, the feds also announced a notable new sovereign AI computing commitment on Thursday.
“Almost nobody has capacity available at any price.”
Ali Asaria, Transformer Lab
Prime Minister Mark Carney’s Liberal government has promised to “enable Canadian AI companies to access 500 [megawatts] of sovereign AI compute through new financial tools, such as backstops and offtake agreements, to mobilize private capital and expand compute capacity.” The feds have previously leveraged similar mechanisms to help get carbon capture and mining projects off the ground.
In an interview with BetaKit, Sean Mullin, who previously served as lead economic advisor to the Prime Minister and helped design and launch Canada’s $2-billion sovereign AI compute strategy, said this proposal appears geared towards addressing the “chicken-and-egg, Catch-22 problem” facing smaller domestic AI startups seeking access to compute in Canada.
While the exact nature of this plan appears to still be under development, the feds appear poised to pull together pools of these companies and provide guarantees that the compute they have committed to buy from these data centres will be purchased as a means of de-risking them.
Canadian startups crowded out
Data centre builders require huge commitments to secure the financing they need to break ground on new projects. But demand for the AI computing power these hubs provide is so high that the companies developing them currently have their pick of the litter in terms of buyers.
The feds’ 500-megawatt (MW) target would be a big step up compared to the country’s existing AI processing power. According to a recent Boston Consulting Group report, Canada currently has 300 MW of AI data centre capacity available. Work to drastically boost that number is well underway, including by domestic telecom giant Bell, which aims to build Canada’s largest data centre in Saskatchewan with a 1.2-gigawatt (GW) capacity. Mullin noted that these plans are based on the assumption of a huge increase in AI demand.
“Countries around the world are scrambling to build and attract as much compute as possible … If Canada manages to set the terms on this appropriately, it could be a strong competitive advantage.”
Jaxson Khan, Aperture AI
That 500 MW could be spread across multiple projects over time. “It would be quite a lot right now, but it may be only a small share of demand by 2028, 2029, and so I think it’s a reasonable target,” Mullin said.
At the moment, given the choice between selling future compute to a more established foreign firm with a deep balance sheet and strong credit rating or to a more nascent, cash-strapped domestic startup, many data centre developers look to be choosing the former because they are viewed as less risky by lenders, thus crowding out smaller Canadian AI companies.
“What the [Canadian] government is articulating here is maybe they can help solve this market failure by helping make this market,” said Mullin, who now works as senior fellow at the University of Toronto’s Munk School of Global Affairs & Public Policy.
Demand for AI compute in Canada could soon begin to grow even faster. During the US midterm elections last week, the Democrats seized control of the House and the Senate from Republicans, while voters approved nearly every data centre moratorium on the ballot. This spurred a big tech-stock selloff as experts predicted these shifting political winds to both slow and increase the cost of data centre construction south of the border.
Anthropic, Meta, and Microsoft have already turned to Canada for AI processing power, and these federal election results could lead more major US tech firms to look north.
“If that were to happen … that would make the need for this type of intervention stronger,” Mullin said.
In his budget speech, Finance Minister François-Philippe Champagne listed AI infrastructure as among the projects “that will power our future.”
“Countries around the world are scrambling to build and attract as much compute as possible … If Canada manages to set the terms on this appropriately, it could be a strong competitive advantage,” Aperture AI CEO Jaxson Khan, who also works at the Munk School as a senior fellow and previously served as a senior policy advisor for Innovation, Science, and Economic Development Canada.
Cautious optimism from Canadian tech
Canadian AI entrepreneurs and investors whom BetaKit spoke with were also happy about the feds’ latest move to expand domestic access, which they think could benefit Canadian AI startups “enormously” if thoughtfully implemented. However, some reserved final judgment until they see what it looks like in practice.
“We’re excited to see the federal government backing sovereign AI compute and the momentum behind a fully Canadian AI stack,” Ideogram co-founder and CEO Mohammad Norouzi said. “Access to compute is one of the biggest challenges for AI companies building in Canada, and this is a significant step toward expanding capacity here at home.”
Norouzi added that Ideogram “would love to bring a meaningful portion” of its compute spending to Canada.
Ali Asaria, co-founder and CEO of Kitchener-Waterloo AI research startup Transformer Lab, said that, at the moment, securing access to the graphics processing units typically used to train and deploy AI models “is close to impossible.”
“Almost nobody has capacity available at any price,” Asaria said. “The vendors who do will only rent to you if you commit to three years or more and pay 25 to sometimes 50 percent up front. Even once you hand over a multimillion-dollar deposit, they tell you they’re going to go buy the hardware and aren’t sure when it’ll be online. It might be six months.”
For startups lacking the balance sheets of major firms, Asaria said that process is simply “a non-starter.” While Transformer Lab is willing to pay for the compute it needs on a monthly or annual basis, he said the company and others like it “can’t finance the hardware on behalf of the data centres and then wait half a year to use it.”
“Ottawa helping data centres finance those purchases might sound crazy, but it might be crazy enough to work,” he added. “Demand is so high that the government’s risk is close to zero.”
More on #Budget2026
• What’s in Budget 2026 for Canadian tech?
• Ottawa earmarks $176 million to establish online super regulator
• Real-Time Rail payments system to launch Monday, feds say
• Canada to implement QSBS-style tax incentive and “level playing field” with US
• The feds plan to merge BDC and EDC. What might that mean for Canada’s largest VC investor?
Unified.to co-founder and CEO Roy Pereira and Tailscale general manager of sovereign AI Patricia Prewitt also welcome the feds’ plan.
“Making compute more accessible and affordable for smaller Canadian companies through financing new data centres is an excellent public investment,” Prewitt said. “We look forward to seeing who has access as a key metric of success.”
Mullin echoed the latter assertion, noting that “deciding on a fair, transparent way to allocate that compute will be important.”
Allen Lau, co-founder and operating partner at frontier tech-focused Toronto venture capital firm Two Small Fish Ventures, is cautiously optimistic about the proposal, as well. He thinks it could accelerate domestic AI data centre development and improve access for Canadian AI startups “enormously,” but emphasized that “the devil is in the details.”
Lau favours the idea of using financial backstops and offtake agreements to attract private capital rather than relying solely on federal funding. He expects the biggest immediate beneficiaries of such an approach to be the firms developing and operating large-scale AI data centres and the infrastructure and energy providers supporting them.
“To me, the critical question is whether this will actually give Canadian AI companies access to competitively priced compute,” Lau added. “Building 500 MW of capacity in Canada is one thing. Making it accessible and affordable to startups is another. We need both.”
Lau said he is interested to see how the Government of Canada intends to structure these agreements and determine pricing and access, as “those details will ultimately determine how much this initiative moves the needle for Canada’s AI ecosystem.”
Feature image courtesy Microsoft.
