Dispatch · Tech sovereignty
Canada promised to be master of its own house. The pitch was sovereignty. The receipts run on American silicon, Florida headquarters, and funding amounts that are notional and subject to negotiation.
On June 4, 2026, Mark Carney stood at a podium in Toronto to launch Canada's national AI strategy entitled "AI for All." The pitch was sovereignty, the good clean homegrown kind, free of foreign entanglement. Carney referred to the concept at least ten times by name. For comparison, compute got seven mentions, and intelligence, a lowly one.
The Prime Minister told the room that Canada would be "masters of our own house" in the virtual world. Though it turns out that that house is mostly sublet.
Starting with the money, which no one could quite agree on, we were bombarded with figures ranging from $2 billion to $200 billion, depending on whether you were counting spending or daydreaming about growth. Independent tallies that looked at the existing programs being folded into the announcement landed north of $3.5 billion.
The ambiguity seems intentional. It may be the whole point.
It uses Carney's Build-Partner-Buy framework: build in Canada first, partner with allies if you can't build, and buy from abroad only, in Carney's words, after exhausting these options. Sovereignty as a waterfall, cascading politely from domestic to foreign.
The trouble is that the receipts run the waterfall backwards.
The largest near-term commitment goes to Cohere, which got up to C$240 million to build a data centre with CoreWeave, an American firm, on Nvidia chips. In other words, the flagship sovereignty play is American hardware running American silicon with a Canadian invoice stapled to the front.
Carney bragged in the same speech that Canada is one of only four countries with its own large language model. True. Though he failed to mention that the compute to run it largely isn't ours.
The honesty as usual is in the fine print. The Compute Access Fund is described in its own terms as a subsidy so Canadian firms are "not entirely dependent on American hyperscalers."
Not entirely. That qualifier is the most accurate word in the whole strategy, and it's doing its accurate work two clicks below the headline where nobody reads.
Quantum tells the same story in a different costume. Canada's National Quantum Strategy is backed by $360 million over seven years, against Germany's roughly $2.8 billion CAD and France's €1.8 billion. The government's own consultations warned that the funding "may be insufficient to achieve the goals," which is the polite federal way of filing a complaint against yourself.
D-Wave, founded in British Columbia, now answers to Florida. Xanadu merged with a Nasdaq shell and arranged to raise up to US$300 million from a firm tied to the Trump family. Canada keeps the labs and exports the upside.
Budget 2025 added $334.3 million and reclassified it from the science budget to the Defence Industrial Strategy. The funding rationale just migrated to whichever department had political appetite that fiscal year.
Carney opened the speech by quoting Pope Leo XIV on artificial intelligence: never has humanity had such power over itself. The complication is that the power, the compute, the chips, the capital, the founders, mostly sits somewhere else.
The word "sovereign" turns an industrial subsidy into a geopolitical virtue years before any silicon is poured. The compute strategy concedes its dependence in its own grant language. The quantum strategy filed its own insufficiency in the consultation record.
We will be masters of our own house, eventually, notionally, and subject to negotiation.