Canada’s push for cloud and AI sovereignty is driven by the tense geopolitical climate, but concerns can be addressed without the country building and owning all the technology, say tech and policy experts.

Moreover, the federal government’s $2.3-billion AI strategy isn’t sufficient funding to move the needle on the AI infrastructure built, owned and operated in Canada, they said during a webinar by the Ottawa-based Centre for Canadian Innovation and Competitiveness.

Other countries, such as France and Germany, use data centres, cloud services and AI models provided by U.S. hyperscalers, but have implemented local laws and rules to ensure control and protection of sensitive data.

“A lot of people are worked up. There's a lot of policy tailwinds. There's a lot of fear,” said Graeme Harrison (photo at right), co-founder and CEO of Calgary-based Simply Silicon, which builds AI data centres and networks.

“There's uncertainty. It's compounded by the current geopolitical situation,” he said.

Much of the policy action and narratives around digital sovereignty are focused on it being an opportunity to foster a local supply chain that can create economic activity inside of Canada,  rather than importing U.S.-based cloud and AI inference services, he said.

“There's been a lot of gasoline poured on the fire by people who are lobbying for either contracts with government or . . . subsidies from government,” Harrison said.

Banks, for example, are being lobbied to use Canadian-owned AI inference infrastructure, he noted.

“Look at the tremendous amount of domestic lobbying that has gone on to encourage these types of sovereignty discussions over the past three years by players who are now lining up to get money out of the various pots of money that the AI strategy announced.”

Yet the federal strategy’s funding isn’t even at the scale necessary for seed capital to build large-scale AI infrastructure, Harrison noted. San Francisco-based AI developer Anthropic’s last funding round, for example, was $85 billion just for one round of financing.

“I don't believe that the AI strategy is being done to effectively execute on a program of, ‘We're going to build stuff with federal money that's going to move the needle.’ I don't see the potential of how it could possibly move the needle there at that small of a scale,” Harrison said.

Webinar moderator Lawrence Zhang (photo at right), head of policy at the Centre for Canadian Innovation and Competitiveness, said “sovereignty has become sort of a magic word.”

Sovereignty is being used to describe several different issues or problems all at once, he said.

They include data security; resilience or keeping systems running; legal jurisdiction or who can compel access to data; domestic ownership of infrastructure; and industrial policy or building Canadian firms and capturing value.

“Those are all real things, but they're not the same problem and they don't all have the same fix,” Zhang said.

“And when you bundle them all under one word, you end up just reaching for the thing that feels like you can hold it in your hands the most, which is to build it here, own it here, put the servers in Montreal just for the sake of it,” he said.

“And it turns out that's a pretty weak proxy for the thing that you actually care about, which is control [of the data and infrastructure].”

Relying on foreign AI service providers raises several concerns

Josh Tabish (photo at right), senior director for Canada at the Chamber of Progress, said people have raised concerns about the so-called “kill switch scenario.”

“What happens to us if the providers of technology that we rely on decide that Canada isn't an ally anymore, and they order their companies to stop working with us?”

Canadians worry about the impacts on critical infrastructure and on health care and other sectors in that scenario, Tabish said.

There are also concerns about foreign access to data under other countries’ laws. People worry that the U.S. Cloud Act, for example, compels U.S.-based technology and communication companies to turn over data under their control to American law enforcement, even if that data is physically stored in another country.

“While these concerns aren't entirely unfounded, I do think the risks are, in most cases, overstated,” Tabish said.

The solution proposed by many advocates of cloud and AI sovereignty is that Canada needs to get off of U.S. technology, increase Canada’s digital independence, and rebuild the tech stack in Canada to ensure we have made-in-Canada options.

“What I worry about in these conversations is that we often over-index on the location of data, where it's stored, where the servers are, which country it's operating in,” Tabish said.

“And we spend a bit less time talking about what kind of security or controls are in place for that data. Is it encrypted? Are there strong contracts in place to help protect that? These are some of the issues that I think deserve a lot more discussion.”

Jaxson Khan (photo at right), CEO of strategy consulting firm Aperture AI and a senior fellow at the Munk School of Global Affairs & Public Policy at the University of Toronto, said a lot of the sovereign AI debate is occurring during a tense relationship with the U.S., and the Canadian economy’s historical dependence on the U.S.

“I think the desire, whether it’s emotional or economic, to try and reduce some dependency is sensible. It makes sense to have that reaction when you're feeling threatened.”

However, Khan and Sean Mullin, senior fellow at the Munk School, said in a report that there’s an over-emphasis on AI sovereignty when it comes to reducing reliance on the U.S. and safeguarding sensitive Canadian data.

In France and Germany, companies are using Microsoft and Google technologies, but have implemented local laws and rules that compel the U.S. hyperscalers to abide by those countries’ jurisdictions and avoid the “kill-switch” scenario or being compelled to provide data.

“We call this jurisdictional sovereignty, where you can ensure that outlet infrastructure effectively [follows] the local rules,” Khan said.

Khan and Miller’s report also noted that when data involves national security or is at a high-security level, it makes sense for Canada to build infrastructure that’s domestically owned and operated. But most Canadian companies and organizations aren’t going to require that level of protection or independence.

Tabish said this approach means “really taking a kind of clear-eyed and risk-based approach to the data we have and then taking measures that are available right now in practical and pragmatic terms to protect that.”

Tools like encryption can help protect Canadian data so that, even if the U.S. government asks for the data to be turned over, “it’s a scrambled mess and they can’t make any sense of it,” he said. “Similarly, there are procurement and contracting measures that governments like the Government of Canada can take to protect that information.”

 Canada needs access to AI models and cloud services to stay competitive

Khan said he’s less worried about the kill-switch scenario than about the loss of availability and access to key services like hyperscale data servers and the cloud.

As AI models become more advanced, “sometimes I do think about deep prioritization or dequeuing. What if Canadians simply don't get access to the highest-quality latest models, either due to . . . a national security [policy] in the U.S. or China?”

If this means Canada doesn’t have competitive industrial capabilities, then the country is digitally sovereign in name only because it would still be highly dependent on other countries’ technology, Khan said.

So AI sovereignty isn’t having 100-percent Canadian-made, designed, owned and operated infrastructure, he said. Rather, “It is, ‘Are we free from coercion?’”

Tabish said AI models and services offered by the U.S. and other countries are getting safer, more resilient and cyber-secure. “We need to ensure that the market stays open to those so that Canadian businesses and consumers can have access to those while we scale up our own sector to try and catch up.”

Harrison said there are opportunities for Canadian providers of data centres and cloud and AI services. Simply Silicon has a 12-megawatt flagship data centre in Calgary, but demand for the facility is currently about 60 MW – about half of it from Canadian players, he said.

“I think there is substantial demand to oversubscribe a dedicated Canadian infrastructure base for [AI] inference here,” to provide very specific services.

He said he believes that in time, the market in Canada will rise to fill that domestic demand “with players who can offer domestic infrastructure that isn't just headline-oriented” and preoccupied with sovereign AI.

However, to realize that market, government needs to reinforce the right of the private sector to build infrastructure in Canada, Harrison said.

He pointed to TELUS’ planned data centre project in downtown Vancouver. Within a week, 750 people showed up to protest the project and called on the B.C. government to stop it.

“They're calling for all sorts of different stakeholder engagement mechanisms that are provided by the nature of our constitutional order to exert their democratic right to squash infrastructure projects,” he said.  

But if every person along the route of the Canadian Pacific Railway – Canada’s first national infrastructure project started in 1867 – had been given a right to object legally to that rail line’s construction, “it would have never been built and B.C. would have never joined Confederation,” Harrison said.

“We wouldn't have the country called Canada today. There comes a point when you want to build things, you need to build things for the necessity of the national order, for the nature of Confederation . . . that overrides the need for individuals to be given a right to express via legally binding stakeholder processes attempts to stop them.”

If Canada tapped its abundant energy and water resources to build AI infrastructure, the country could become “the world’s powerhouse of AI inference,” Harrison said.

“The problem is today we're living in this very fragile and baroque order, where any little change is met with very stiff and very vocal and aggressive resistance and that makes it impossible to build things,” he said.

Creating the federal Major Projects Office was a good step but it’s not enough, he added. “We need total reform of what it means to build things in Canada.”

How can the federal government help Canada’s AI industry?

Kahn said if the federal government’s AI strategy was truly structured toward creating a “domestic walled garden,” the funding would be much higher – closer to $5 billion to $10 billion.

Most of the money in the federal AI strategy is really to support AI adoption by businesses, which is lagging in Canada, he said. “And most of the policy measures are actually oriented on [improving] trust and literacy.”

Canada, with its early lead in AI researcher, probably should have commercialized on the AI opportunity far more than any other middle power in the world, “but we haven’t done so,” he said.

Khan said he worries less about the AI strategy funding and “more about the larger business economic environment that we’re creating and operating. Are we actually changing how we do that?”

Harrison said that given Ottawa’s $2.3-billion commitment to its AI strategy, “clearly the Canadian federal government is not making a bet that actually matters or they wouldn’t minimize the amount of resources put behind it.”

“People are terrified of AI,” with Canadians having among lowest trust in AI for countries worldwide, he noted.

“I think that there's a worry that if politically you go all in on this, that all that's going to happen is you're going to get voted out because everybody's going to hate you. So to that extent, I think it's a very challenging scenario to be in,” Harrison said.

He said if the federal government really wants to help Canada’s AI industry, it should eliminate the capital gains tax.

“We need to send a signal to the rest of the world that this is the best place to come and build a business so that we can pull in a gross talent base that is able to start to leverage all of the great stuff we've got in our country that currently isn't being leveraged, power resources and energy resources being a primary one,” Harrison said.

Tabish pointed out that the federal government says sovereign cloud and AI is a high priority, and that encryption is an important part of securing and protecting Canadians’ data.

Yet Ottawa is pushing forward Bill C-22, which “threatens to weaken encryption that we all rely on businesses, users to keep our sensitive information safe,” he said.

Bill C-22 “also proposes extraordinary powers to build interception and surveillance capabilities into the networks of cloud providers and AI providers and other large network operators in Canada that can then be exploited by foreign adversaries or bad actors, because once that technical vulnerability exists, the door doesn't just open for the good guys, it opens for the bad guys, too,” Tabish said.

The federal government could help accelerate AI adoption by showing real-world examples of the technology’s possibilities, he said.

Alberta’s Ministry of Technology and Innovation has found ways to save millions of dollars by using AI to manage a major infrastructure database, he noted.

“At the federal level, at the risk of sounding critical, I haven't seen the cool case studies yet,” Tabish said.  

“So lead by example. Do cool things with AI and share them with Canadians so that they can feel like they can trust it and they can do it, too.”

R$