Why the shift toward on-device AI puts Ottawa’s $925M data centre bet in question
In my last post, I looked at the case for Canada’s AI data centre build-out. The main argument is that we need to build these centres to remain competitive in an economy transforming around AI and to keep our data sovereign. If the government doesn’t intervene to subsidize them, Canadians will lack “access to the compute they need,” and the centres would be built and controlled by US companies.
In this post, I want to set aside the data sovereignty issue and outline an argument that isn’t receiving much airplay at the moment in debates about the need for new data centres. It goes to the heart of whether we need to build new centres to enjoy the full benefits of AI.
The counter-argument is that access to the compute we need to enjoy many if not most of the benefits of AI is rapidly becoming much less expensive and will soon be accessible at a far lower cost economically and environmentally.
The future of AI, the argument goes, is not one in which the models we interface with most often are situated in the cloud and reliant upon large, energy-hungry data centres. Instead, it’s one in which most of the processing is done on board our personal devices, like our phones or laptops, on a company’s local server, or a rack of servers at a university or an internet service provider.
And this isn’t a matter of pure speculation. The evidence pointing in this direction is mounting. Though, to be clear, we will still need larger models and data centres for some things. The question is how many.
May 26, 2026
The economic benefits are modest, the environmental claims are unverified, and the strain on grids and water is real
On Sunday, hundreds of people marched through the streets of Vancouver to protest two new AI data centres Telus is building in the city — which has just announced water restrictions with more expected in June.
Ottawa has pledged some $925 million over five years to build “sovereign AI data centres” across the country, and this month it named Telus as the first builder, to expand one facility in Kamloops and develop the two new ones in Vancouver. The case for a build-out rests on a belief that Canada needs more domestic compute or it will stay dependent on foreign infrastructure for something vital. But the case is weaker than it looks, and rests in part on an assumption about the future of AI that may turn out to be false.
AI Minister Evan Solomon says the goal is to ensure that “Canadian innovators, researchers and businesses have access to the compute they need,” while keeping “Canadian data… on Canadian soil.” We need far more domestic processing capacity to stay competitive as AI transforms the economy, and if we fail to build it ourselves, Canadians will become more dependent on foreign entities for some of our most critical infrastructure.
There are reasons to question whether the current rush to subsidize and build these centres is good policy. The economic benefits seem limited once the centres are built. They place a further strain on electricity and water systems that are already stressed. And many of the environmental claims the companies make can’t be easily verified.
More crucially, the push to build a network of large, power-hungry AI data centres assumes that most of the AI we will be using in the years to come will still rely for the most part on large models we access in the cloud. I’ll offer reasons to be sceptical of this in my next post. This one lays out the build-out itself, and how much about it is still unsettled.
May 21, 2026
With major tech companies threatening to leave Canada and US lawmakers weighing in, the government is signalling it will amend the bill. But the constitutional questions won’t end there.
Opposition to Bill C-22 , the lawful access bill currently in second reading before Parliament, has intensified sharply in recent days.
More than a dozen major companies — Apple, Meta, Signal, and several VPN providers among them — have raised concerns about key provisions, with some threatening to withdraw services from the Canadian market if the bill passes as drafted.
Two members of the US Congress have written to the Public Safety Minister warning that certain powers could compromise the privacy of American citizens and national security.
Civil liberties groups and privacy experts have published open letters and op-eds , and the dissent on social media has been equally loud.
Meanwhile, the RCMP and the Canadian Security Intelligence Service have been on a media blitz bemoaning Canada’s lack of a lawful access regime and making the case for why the bill needs to pass.
The proposed legislation covers a range of new lawful access powers, but the two provisions drawing the strongest response are those dealing with encryption and bulk metadata preservation. Critics fear the bill would permit secret ministerial orders compelling companies to introduce vulnerabilities into their systems, and mandate the retention of metadata that would capture nearly every Canadian’s movements for up to a year, without any individualized suspicion.
I teach and write on section 8 of the Charter, which protects against unreasonable search or seizure, and the concerns about encryption and metadata raise serious constitutional questions.
Before getting to those, it’s worth taking stock of where the bill stands politically and what changes are likely before it reaches third reading.
May 19, 2026
The Bank of Canada offers a helpful snapshot of AI adoption, job losses, and early productivity gains
In Ottawa last week, the Deputy Governor of the Bank of Canada, Michelle Alexopoulos, offered an illuminating snapshot of how AI is affecting employment and productivity across industries in Canada.
I posted a few details about her speech on LinkedIn on Friday, but there are a few other points here that I think will be of interest to readers.
Alexopoulus poses a big-picture question about AI at the outset. Will it prove to be a general-purpose technology, like the steam engine or the computer, that “reshaped entire economies and societies”? Or will it remain for many of us in the short to medium term limited to certain purposes, with a marginal impact on the workplace as a whole?
The early data on the uptake of AI in Canada points in the direction of a more pervasive transformation, but one that may unfold gradually. While some 12% of companies were using AI last year, up from 3% in 2022, they’re adopting it unevenly, with only 1.5% of businesses in accommodation and food services using AI, but more than 30% of finance and insurance firms doing so.
The Deputy Governor anticipates that “some jobs will be replaced by AI. New jobs will emerge, and others will be transformed,” offering this helpful analogy:
… when computers were first introduced into offices, some jobs vanished, like office typists and switchboard operators. New jobs were created, like entire IT departments. And other jobs changed — analog tasks were digitalized, and workers learned to use computers.
Alexopoulos makes three more specific observations about what the data tell us about how AI is currently affecting the workforce.
May 14, 2026
Most of our closest allies do not force telecos to retain everyone’s metadata for up to a year without oversight
Last week, when I appeared before the Standing Committee on Public Safety and National Security on Bill C-22, a question kept coming up from members on both sides of the table: how do other countries handle this? Do our closest allies require electronic service providers to retain the metadata of nearly everyone in the country for a lengthy period, without grounds or individualized suspicion, as the government is proposing to do here?
With one exception, no, the Five Eyes partners have not gone where Bill C-22 proposes to go.
The European experiment
For context, the European Union passed a similar law in 2006 and the European Court of Justice struck it down in 2014. In the Digital Rights Ireland case, the Court acknowledged the general interest in making sure that data is preserved to help investigate and prosecute serious crime, including terrorism. But it held the impact of bulk metadata retention on privacy to be “wide-ranging,” “serious,” and disproportionate. The data retained under the law was not limited to investigating “serious crime.” Police could access it without a warrant. And the periods of data retention bore no connection to the possible usefulness of the data for investigations.