Canada’s $2-billion AI computing plan could wind up helping foreign tech giants

Canada’s $2-billion sovereign computing strategy is designed to help domestic companies access the necessary services to power artificial intelligence. But foreign companies could stand to reap much of the benefits – at least initially.

That’s because there are few options aside from U.S. companies such as Amazon, Microsoft and Google that dominate the market for the cloud computing services required to build and run AI models.

Prime Minister Justin Trudeau announced the sovereign computing strategy and access fund in April to bolster the Canadian AI ecosystem. To build AI models and run applications, companies need access to sophisticated computer chips such as graphics processing units (GPUs) and other equipment, known in industry jargon as compute. Demand for chips and access to the data centres that house them has surged as companies embrace AI, and as models themselves, particularly for generative AI applications, become larger and more resource-intensive.

Some companies have reported difficulty accessing compute from cloud providers, while costs have escalated. Academic researchers need access to high-performance computing services, too. But investment in public infrastructure has lagged other countries, and demand is outpacing supply.

As a result, some experts have warned that AI researchers and entrepreneurs could move abroad to gain access to affordable compute.

The federal government’s program is designed to alleviate these issues by helping researchers and companies that use AI access compute in the short-term, and build infrastructure in Canada over the long-term. The chips that power AI are increasingly seen as critical infrastructure and important for economic security.

Some Canadian companies say they can provide an alternative to the U.S. cloud giants, which are known as hyperscalers.

“Canada seems to like the idea of, well, let’s just go to Silicon Valley to solve our problems,” said Geoff Gordon, co-founder and chief executive of Denvr Dataworks in Calgary, which builds and manages data centres for AI workloads. “We should be thinking about being dominant and being experts in this space.”

The details of the program and how the government will allot the $2-billion over the five years of the initiative are still to be determined. Innovation, Science and Economic Development Canada (ISED) started a public consultation in June, and the period for submissions ended earlier this month.

A spokesperson for ISED did not answer specific questions from The Globe and Mail, but said the department is still developing the strategy and will publish a report on the findings of its consultation.

Some organizations that submitted feedback contend the government should subsidize the AI compute costs for Canadian companies as part of the program, arguing that would be the fastest way to benefit industry. Radical Ventures in Toronto, which has invested in a wide range of AI companies, said that short-term subsidies for compute – regardless of where the cloud providers are located – can bridge the gap until Canada builds up more domestic infrastructure.

“These things can’t be built overnight and they require a great deal of expertise,” said David Katz, a partner at Radical Ventures. “This is still a pretty nascent space, so I wouldn’t beat ourselves up over the fact that we don’t have incredible cloud AI operations in Canada yet.”

Mr. Katz said subsidies would be in place for only a short time and would help to spur more AI development in Canada. In turn, that would entice cloud companies to add infrastructure here to meet demand. “That’s a big first step towards domestic sovereignty,” he said. “In one to five years, it could be a very different situation.”

Indeed, not everyone is concerned about whether the cloud providers supporting AI development are foreign or domestic, just so long as Canadian companies can affordably access these services in the near-term. “If it allows hyperscalers to further provide such cloud capacity to Canadian businesses, that’s a win,” said Imran Ahmad, a partner at Norton Rose Fulbright who focuses on technology.

Toronto generative AI company Cohere, which recently raised US$500-million, has registered to lobby government about seeking funding through the program, according to the federal lobbyist registry, but declined to answer specific questions about its views.

“Rapid access to compute infrastructure could revolutionize Canadian and global AI, and provide a huge shot in the arm for our innovators to compete on the world stage. There is no time to waste,” chief executive Aidan Gomez said in an e-mail.

Adam Hendin, co-founder and CEO of Toronto-based AI cloud company Radium, said the government’s program should focus on local providers to reduce dependency on foreign firms. “If we’re going to support Canadian enterprise, it should be with Canadian infrastructure,” he said. Radium’s submission to ISED said that the government should subsidize the infrastructure assets of local cloud providers, likening the effort to government support for renewable energy projects.

Radium’s business plans do not depend on government support, Mr. Hendin emphasized, and the company has financing to build six new compute clusters, one of which will be in Canada. “The reason the Canadian government should be doing this with taxpayer money is the same reason we invest in our military, and our electricity grid,” he said.

Canada has some advantages for cloud computing, such as abundant renewable energy sources and a cool climate. But building a domestic cloud industry to service AI is far from easy. Data centres are capital intensive, and customers risk getting locked into the big cloud providers, making it harder for upstarts to win new business.

Hypertec Group Inc., based in Quebec, is forging ahead regardless. The company has plans to deploy about 100,000 GPUs in Canada and the U.S. by next year, a massive expansion that could cost upward of $6-billion. The effort will be financed through a combination of equity and debt, said Jonathan Ahdoot, who heads Hypertec’s cloud division.

Should the government offer compute subsidies, he said, there should be a “buy local” strategy to spend on domestic cloud providers. “That’s really what’s going to build up the cluster in Canada,” Mr. Ahdoot said. “If you just send capacity to somewhere in the U.S., you’re helping one side, but you’re not really helping the whole ecosystem.”

The fact that some cloud providers are already expanding in Canada to meet AI demand raises the question of just how necessary a subsidy is for private industry. Not everyone agrees that difficulty accessing compute is holding back Canadian companies, either. The Council of Canadian Innovators said in its submission to ISED that compute is “not a major barrier” for most growing companies.

Still, CCI recommended that the government buy compute access from commercial providers and grant blocks to Canadian companies. Building sovereign infrastructure will be necessary to protect sensitive data and service critical industries, according to the organization, which also said that Canada should work with allies such as the European Union to secure compute access.

Nick Schiavo, the director of federal affairs at CCI, is wary that the government’s compute strategy is the right program to support a domestic cloud industry, however. “Developing Canadian capacity from domestic firms is critical,” he said. “I just don’t think that the investment is significant enough to be able to do that.”

While AI infrastructure to service both researchers and companies is important, Canada is missing something in the middle, according to Ryan Grant, an assistant professor at Queen’s University, whose work focuses on extremely large computing systems.

He recommends that part of the compute fund be used to build a supercomputer through a public-private partnership. Companies could access the system when looking to commercialize a technology. “If you’ve come up with a new idea, then you need an innovation supercomputer to bridge that idea into a new product,” he said, adding that it could be useful for a whole range of applications beyond generative AI, including drug discovery, energy research, and disaster preparedness modelling, among others.

Building a supercomputer would also help Canada develop the talent required to operate and maintain such infrastructure, which is exceedingly complex, Prof. Grant said.

In its submission to ISED, Queen’s University said that a supercomputer could be housed on its campus.

Regardless of the details of the government’s compute strategy, one company that could benefit is Nvidia Corp. in California, whose coveted GPUs have become the workhorse of AI. It’s little wonder, then, that Nvidia CEO Jensen Huang is among the biggest proponents of sovereign AI.