A 320-megawatt “AI gigafactory” in the Greater Toronto Area sounds like the kind of big, confident move Canada says it wants. It also sounds like the kind of project that can quietly turn into an expensive mess if we pretend electricity, land, and politics don’t matter.
The basic news is simple: HIVE Digital Technologies says it plans to build an industrial-scale AI data center outside Toronto, with 320 MW of utility capacity. The pitch is that this would strengthen Canada’s AI infrastructure and support “sovereign AI” — meaning more of the computing power and data control stays inside the country instead of being rented from data centers somewhere else. At full build-out, the facility is expected to host more than 100,000 GPUs.
If you like the idea of Canada being more than a customer of other people’s technology, it’s hard not to like this. Compute is power now. Not metaphorically — literally. If you don’t have enough of it, you don’t just fall behind on flashy chatbots. You fall behind on drug discovery, manufacturing, logistics, climate modeling, defense, and even basic public services that are being rebuilt around machine learning. The countries and companies with reliable access to compute get to move faster and pay less for experimentation. Everyone else becomes dependent.
So yes, “sovereign AI” has a real point. If Canadian universities, startups, hospitals, and government teams all have to line up behind foreign cloud capacity and foreign rules, you’re basically outsourcing the future. And once your systems get built on someone else’s infrastructure, switching later is painful and slow. Dependency has a way of becoming permanent.
But the thing I don’t love about these announcements is the implied simplicity: build a giant box, fill it with GPUs, and boom — national advantage. That’s not how it works.
First, 320 MW is not a cute side project. That’s a serious chunk of power, and power isn’t just something you “secure.” It’s negotiated, priced, routed, permitted, and fought over. If this pulls clean electricity away from other needs, people will notice. Imagine you’re running a factory that actually makes physical goods and you’re told your expansion has to wait because a data center needs the capacity. Or imagine you’re a municipality dealing with housing pressure, transit issues, and aging water systems, and the big new plan in town is… a warehouse full of computers. You can call it the future, but you still have to live next to it.
Second, “sovereign” can become a marketing word if the benefits don’t actually land here. A facility in Canada doesn’t automatically mean Canadians get priority access, fair pricing, or any real say in how it’s used. If most of the capacity ends up effectively reserved for the highest bidder — say a foreign company training models for global products — then Canada gets some jobs and some tax base, but not the strategic upside people imagine when they hear “sovereign AI.”
And I’m not even saying that outcome is evil. It’s just the default path of big infrastructure: it serves whoever can pay, unless someone sets rules and sticks to them.
There’s also a more uncomfortable point: stuffing “over 100,000 GPUs” into one place is not just ambition, it’s concentration risk. A single massive site becomes a single massive target for outages, supply issues, cooling failures, and security problems. If you actually care about resilience, you don’t build one throne and hope it never wobbles. You build capacity in a way that can take hits.
On the other hand, I can already hear the pushback: this is exactly what we need, because scale matters. A lot of the best AI work is limited by access to big clusters. And Canada has spent years celebrating its AI talent while letting the most valuable layers of the stack drift elsewhere. If someone is finally willing to put real steel in the ground, maybe we should stop nitpicking and let them build.
That’s a fair argument. But it only holds if the project doesn’t become one more case where public excitement covers for private upside.
Because the stakes are real. If this works the way it’s being sold, Canadian researchers and companies get faster access to compute, less exposure to foreign policy swings, and more ability to set standards around sensitive data. If it fails — or if it “succeeds” but mostly serves outside demand — we risk locking up power, land, and grid attention for years while telling ourselves we bought sovereignty, when we mostly bought a big electric bill.
Picture two near-future scenes.
In one, a Canadian health network wants to train models on domestic patient data with strict controls. Having local capacity makes that actually doable without a maze of cross-border rules and uncomfortable compromises. That’s a win you can feel.
In another, local residents are told energy costs are rising, and grid upgrades are being prioritized, but they’re not sure why. Meanwhile, the shiny AI facility hums along, and the real gains — the best-paid roles, the key contracts, the leverage — sit somewhere else. That’s the kind of resentment that kills good projects and poisons the politics around AI for a decade.
So I’m torn in a very specific way: I like the ambition, and I don’t trust the framing. “Sovereign AI” can be a serious strategy, but only if we’re honest about tradeoffs and clear about who gets access, under what terms, and what happens when the grid gets tight.
If we’re going to build something this big, what concrete guarantees should exist so that Canadian public interest wins don’t end up being an afterthought?