Canadian AI: What kind, and who will own it

What is it we should fear about AI? That it will take our jobs? Overload our electricity networks and make our own energy needs more expensive? Put scarce freshwater resources at risk? Or is it that it will be unregulated, breach privacy laws and be used for a wide range of perverse purposes?

If Canadians are among the most distrustful of the AI revolution, it isn’t because they are uninformed or technologically regressive. Their reluctance is because there is no plan for Canadian AI guided by an overarching public interest.

To the contrary, the hundreds of AI data centre proposals in Canada are driven by an investor frenzy completely disconnected from Canadian needs or economic and social goals.

A Canadian Press freedom of information request from the federal government revealed that power demands from current proposals total more than 20 GW, a massive capacity comparable to the total power needs of all Canadian households, and 25 times greater than the roughly 850 MW the federal government has projected that known sovereign AI proposals will reach by 2030.

The vast majority of the AI data centre proposals are in Alberta, which has been luring U.S. AI investors looking for large industrial spaces and cheap energy.

The province has declared that AI data plants can be treated like oil sands or petrochemicals, with a “build your own natural gas electricity plant” component. That is what Meta is doing with its $13-billion project near Edmonton which broke ground in July. However, there is almost nothing Canadian about the Meta AI facility, which will be a cog in its global network, while it continues to hold Canada hostage over our laws requiring online platforms to pay for Canadian media content.

Canada’s AI needs are very different from those of Meta, the U.S. hyperscalers and the myriad investors wanting to get inside the AI bubble. Canadian sovereign and public interest AI has much smaller, affordable and sustainable infrastructure requirement.

Our needs are to ensure that our personal data cannot be accessed by U.S. companies under U.S. law. Canadian researchers and scientists, our public institutions and businesses, should have reliable, affordable, world-class computing without having to purchase it from a U.S. company. None of this requires Canada to go down the U.S. AI road.

Canadian sovereign AI is the starting point. Sovereign AI must be Canadian owned, with data processed, stored, transmitted and regulated in Canada. Two current Canadian projects — Telus (150 MW) and Bell (300 MW) — are the current proposals that meet the government’s own sovereignty definition, although Bell’s Fabric AI data centre will rent out to U.S. companies.

A more focused public interest AI is being proposed by Canada’s universities, research hospitals and not-for-profit affiliated institutes in the Digital Research Alliance of Canada (DRAC). Canada’s federal AI strategy calls for a publicly owned and operated supercomputer — SCIP. DRAC has proposed to lead a public sector project to build the supercomputer.

“Publicly-owned AI infrastructure delivers more for less, putting public interest over market forces,” DRAC argues in proposing to “integrate the new supercomputer with existing and emerging national research systems — data, software and services — to create a unified hub.”

Canadian AI does not have to compete with the U.S. frontier AI companies using their models. Open-source training models that are already competitive with those of frontier companies are being used and fine-tuned by public institutions in multiple countries. Inference models, fine-tuned on the supercomputer, can run on small, distributed servers that perform up to 85 per cent of practical AI uses with minimal energy and infrastructure.

The current AI question in Canada is how many data centres will be allowed, and where. Unfortunately, the answers to these questions will be determined by the lobbying power of AI investors and the inevitable guerrilla warfare against the data centres by angry and distrustful citizens.

Canada’s AI for All strategy promotes a rapid expansion of AI but does not answer the question of how many data centres. It offers a goal of sovereign AI but does not constrain the U.S. hyperscalers from dominating the Canadian AI landscape, nor does it prioritize public interest AI.

The better questions to ask are; what kind of AI does Canada want, what purposes it should serve, and who should own it. Crucially, who can Canadians trust to build a sovereign AI rooted in the public interest?

An OECD study says only about a third of Canadians trust government to protect privacy and fairness in AI regulation. In a KPMG study, barely half of Canadians have confidence in big technology companies, but 79 per cent have trust in universities and research institutions.

Before it’s too late, Canada needs to put a pause on foreign-owned commercial data centres and reboot Canadian AI policy, bringing public interest voices and our trusted public institutions to the forefront of investment decisions.

Fred Wilson is former director of strategic planning with Unifor. Robert Chernomas teaches economics at the University of Manitoba.