Sovereign AI Is Not Just About Keeping Data in Canada

Two “sovereign AI” announcements landed 48 hours apart this week, from the same government, and they don’t actually agree on what the word means. On October 29, Canada committed up to $1.6 billion to Cohere, calling it an investment in “secure, sovereign AI.” On October 30, Canada joined Europe’s supercomputing network, calling that sovereign AI too. Read closely, and the two deals are making almost opposite arguments about where sovereignty actually lives — which makes this a good moment to pull the word apart into the pieces it’s usually used to paper over.

What actually happened

October 29 — Ottawa buys into Cohere. The federal government is investing up to $1.6 billion in Cohere, Canada’s homegrown large-model company, as part of a Series E round targeting up to $7 billion at a $35.7 billion valuation. NVIDIA, Bell, and Germany’s Schwarz Group are co-investing alongside the government, on commercial terms. In exchange, Cohere committed to: keeping its headquarters and incorporation in Canada, creating at least 1,000 high-paying technical jobs by 2030, investing at least $1.7 billion in Canadian computing infrastructure by 2030, and continuing to work with the Canadian AI Safety Institute. Prime Minister Carney’s own framing is the most useful sentence in the whole announcement: sovereignty means “our ability to build, adopt, and govern AI on our own terms,” giving organizations “greater control over where sensitive data is stored, how it is used, and who can access it.” Notice he’s already listing three different things in one breath.

October 30 — Canada joins Europe’s supercomputers. At the EU-Canada summit, Canada became a full Participating State in the European High-Performance Computing Joint Undertaking — reciprocal access to a network of 14 supercomputers, 10 quantum computers, and 19 “AI factories” across Europe, with Canadian researchers and companies now eligible to use them, and vice versa for European users on Canadian infrastructure. The joint statement’s language is notably not about walls: it talks about “reciprocal access to respective digital markets, including cloud and AI services,” “interoperability,” and “technological neutrality” — the opposite instinct from “keep everything inside our own border.”

The same word, two different bets

Pull apart what each deal is actually buying, and they sit at opposite ends of a spectrum that “sovereign AI” usually flattens into one idea:

1. Data residency — where the bytes physically sit. The easiest question to answer and the one most vendors lead with (“your data stays in-country”). Neither deal is really about this.
2. Ownership — who holds the legal/equity stake. This is what the Cohere deal is mostly buying: a literal ownership position in the company building the model, with governance terms attached.
3. Operational control — who can actually access, audit, or shut off the thing day to day, independent of who owns it on paper. Equity doesn’t automatically grant this; it’s a separate, harder-won property.
4. Capability independence — whether the whole stack can keep running on its own terms if an outside party changes its mind. This is the hardest one, and it’s where the Cohere deal’s own structure quietly undercuts its branding.

The Cohere deal is real progress on (2) and a genuine attempt at (4) — but NVIDIA, a foreign hardware vendor Cohere (and basically everyone in this industry) depends on for the chips underneath all of this, is literally one of the co-investors in Canada’s flagship “sovereign AI” company. That’s not a criticism of the deal; it’s evidence that even a G7 government’s best-funded shot at AI sovereignty doesn’t mean self-sufficiency top to bottom. It means buying a controlling, governed position inside a stack that’s still, by necessity, interdependent. The EU-HPC deal is almost entirely about (3), pointed the opposite direction from residency — it’s reciprocal access, not exclusion, and it says so explicitly (“technological neutrality,” “open markets”).

Why this matters at a much smaller scale

Almost nobody reading this is negotiating a $1.6 billion stake in anything, but the same four categories apply when a 30-person company picks an AI vendor. “Is my data stored in-country” is the question most sales conversations answer first, because it’s the easiest one to answer with a checkbox. It’s also, per this week’s news, the one a national government treated as table stakes, not the finish line. The harder, more consequential questions — who can actually see and act on your data day to day, and what happens to your operation if that vendor changes its terms, gets acquired, or raises prices — are exactly the ones residency alone never touches.

The honest caveat

This is a snapshot taken two days after the announcement, not a verdict on whether it worked. Cohere’s Series E is only at an “initial close” against a $5-7 billion target; the 1,000-job and $1.7 billion infrastructure commitments are promises dated to 2030, not delivered facts. Worth a genuine revisit in a year or two to see which of these commitments actually landed — government-backed strategic bets have a long history of underdelivering on the job-creation line specifically, and this is exactly the kind of claim that deserves checking rather than taking on faith because a press release said so.

What this means practically

The next time an AI vendor’s pitch leads with “your data never leaves the country,” treat it as the answer to the easy question, not the whole one. The readiness-audit version of this: ask who can actually act on your data today, what happens to your operation if the vendor’s terms or ownership change tomorrow, and how much of your actual capability depends on a layer you don’t control at all. Canada just spent $1.6 billion learning that ownership and control are different line items. Worth getting that distinction for free, before writing the check.