Kevin Yin is a contributing columnist for The Globe and Mail and an economics doctoral candidate at the University of California, Berkeley.
This essay is part of the Prosperity’s Path series. In a time of geopolitical instability and a shifting world order, the challenges facing Canada’s economy have only gotten more visible, numerous and intense. This series brings solutions.
For anyone not whittling down Claude tokens every day, the significance of artificial intelligence computational resources can seem remote.
But the fact is that AI and the hardware that supports it are quickly becoming general-purpose inputs into science, business, national defence and health care. The productivity gains are already showing up as more accurate health-risk detection in hospitals, faster software development and a lower barrier to solving complex research problems. Canadians need dependable infrastructure to store, process and protect the data behind those uses, and the expertise to develop and use these tools.
More than attracting data-centre investment, the objective is to secure reliable access to computing power, greater jurisdiction over data, and stronger domestic capabilities in data centre development and operation.
Normally economists care more about the location of capital than its ownership because a factory must be located where production occurs. But data centres are different. Computational services can often be supplied from abroad, while many of the economic benefits we care about come from learning how to build, manage and improve them (what economists call an “innovation spillover”). Much like oil, they are also becoming a strategic chokepoint, because storage and computation are increasingly used across the entire economy.
The Trump administration’s recent decision to restrict access to Anthropic’s frontier AI models drives home the strategic point. Just as we have seen how Iran can cripple energy markets by blocking the Strait of Hormuz, the United States has shown that it is willing and able to cut off access to AI tools when its interests dictate.
Open this photo in gallery:
The Equinix data centre on Parliament Street is one of six facilities the global digital infrastructure company operates in the Toronto area. As AI quickly becomes embedded in science, business, national defence and health care, Canada will need the hardware that supports it.Fred Lum/The Globe and Mail
As AI becomes embedded in the basic functions of our society, Canada cannot afford to allow essential services or the accompanying domestic innovation to depend entirely on the decisions of foreign governments. While we cannot control how the U.S. governs access to its AI models, we must do what we can to reduce our dependence.
Across Canada, the fight against artificial intelligence goes offline
Still, data centres impose real costs, and proposed projects have understandably faced backlash across Canada. They raise energy prices for local residents, create noise and provide far fewer permanent jobs than many other large capital projects. The profits are hard to tax, especially when the value being produced is tied to intellectual property held elsewhere. And the broader benefits that could otherwise motivate public support – greater technological capability and national resilience – have not been adequately realized thus far.
Polling this year from the Angus Reid Institute suggests Canadians broadly see the need for domestic AI infrastructure. But more than two-thirds oppose having it built near their own communities. The problem is fundamentally about externalities: the costs are local, while the potential benefits are both diffuse and only partially captured by Canadians.
Just as we do not expect private companies to consider national needs in their production decisions, we cannot reasonably expect individuals to be happy paying higher electricity bills for what are at present only weak innovation and sovereignty gains. The question is, how do we address this misalignment of incentives to get those data centres built?
Open this photo in gallery:
When he was appointed the country’s first Minister of AI and Digital Innovation last May, Evan Solomon said his priorities included building sovereign data centres to power AI technology.Rich Lam/The Canadian Press
To start, our policy framework needs to ensure that Canadians actually receive the strategic and innovative benefits that justify data centre construction in the first place.
Since these are externalities, they are not naturally delivered by standard foreign investment agreements and a more active industrial policy is justified. Federal tax credits and subsidies should favour projects with Canadian-headquartered ownership stakes, joint ventures with Canadian firms, or enforceable commitments to allocate priority compute to Canadian researchers, startups and public institutions.
Canada should continue to welcome global firms that want to build here. But it should also ask what Canadian workers, companies and institutions learn from these projects, and how much control over our data and processing we have in times of geopolitical disruption.
The objection that conditions on ownership and participation will deter foreign investment is far weaker here than in other sectors. This is because the benefit Canada seeks from data centres is different from that of a factory – foreign-owned factories can still create substantial domestic employment and production. If the reasons for supporting data centres are instead primarily to boost our innovative capacity and secure sovereignty over our computational resources, then domestic participation is not a nice-to-have. It is itself the point.
Next, local residents should not have to pay higher electricity bills for infrastructure that they are not using directly and that does not employ many of them.
Open this photo in gallery:
Following the announcement that Telus would move forward with multiple large-scale AI data centre projects in B.C., hundreds took to the street in Vancouver to protest.
Open this photo in gallery:
Protestors cited environmental concerns regarding the high energy and water demands of data centres, a lack of community consultation and an overall rejection of AI’s impact on art and culture.DARRYL DYCK/The Canadian Press
Data centre owners should be required either to provide their own power or to finance equivalent new supply and grid capacity. This supply matching would need to occur at a reasonably high frequency – it should match what the data centre uses at all times. Matching total electricity use over a month or year is not enough if residents still have to pay much more during peak-demand hours. The approval process needs to be strict enough to ensure that tech companies cannot avoid these requirements through clever accounting.
Versions of this idea are already emerging as a response to rising data centre electricity demand, and they should become the baseline. As Manitoba Premier Wab Kinew’s recent rejection of the province’s data centre proposal (which would have had its own natural gas turbines) shows, “bring-your-own-power” does not eliminate all opposition. But it would change the bargain so that communities are not being asked to subsidize the electricity needs of distant users.
Finally, host communities need to become visible fiscal winners.
Municipalities should not be forced into bidding wars against one another by offering local tax breaks to attract data centres. If subsidies or tax credits are justified, they should come from the federal government and be available on a consistent national basis. This way, local governments can keep the full property-tax base created by the data centre and, if necessary, receive additional federal transfers for public goods conditional on accepting large projects. It is easier to tolerate a bit of noise when you know that it is paying for roads and libraries.
Canada also needs a more credible approach to multinational profit-shifting because existing solutions are imperfect. Alberta’s 2 per cent levy on computer hardware is easy to administer but commits two cardinal sins of efficient taxation: It taxes intermediate inputs and singles out one industry.
Some legal experts have argued formula-based profit apportionment could better divide IP-based tax revenue across jurisdictions where companies operate, but this would require careful design and international co-operation. Making progress on these issues would help ensure that communities share more visibly in the benefits of hosting data centres.
The backlash against data centres is understandable. But the answer is not to block construction. It is to design a better bargain.