Canada touts 'sovereign AI' while quietly running Palantir inside defence and policing

Ottawa’s new ‘AI for All’ strategy casts the federal government as the ‘strategic anchor customer’ for homegrown Canadian AI, aiming to lift AI adoption among Canadian businesses from roughly 12 percent today to 60 percent by 2034. Al Vigier, founder of Vancouver decision-support firm Caseway (and thus an interested party, as he acknowledges), argues the pitch collides with reality: the government is already a major AI buyer, but it buys American and buys in secret. He points to an undisclosed 2020 Department of National Defence contract with Palantir’s Canadian arm that started at $14.4 million and, through more than a dozen amendments, ballooned to about $44.4 million committed and roughly $46.8 million spent — plus a separate $3.7 million contract that only came to light when a Conservative MP pushed for AI spending figures. The Ontario Provincial Police have used Palantir’s Gotham platform since 2015. These data-fusion and decision-support tools, Vigier notes, are exactly what the strategy says Canada should own.

His sharper critique is structural. The strategy’s headline funding — $500 million for equity stakes in promising firms, $700 million for compute, a Trusted AI Certification program, and a health-focused ‘missions’ program — contains no actual purchase orders. He reads that as telling: governments reach for grants, equity, and certification schemes precisely when their procurement machinery makes simply buying from a small domestic vendor difficult. Even the one procurement promise is routed through a new Office of Digital Transformation rather than a mandate to cut cheques to Canadian vendors. Vigier warns that state equity stakes can turn founders into wards of the state and scare off allied investors, and he questions launching the missions program in health — among the most privacy-bound and slowest-procuring parts of government — rather than in higher-consequence, already-audited domains like courts, public safety, or defence.

Vigier’s ask is not subsidy but transparency: make supplier lists usable for firms under 50 people instead of treating security questionnaires as moats for incumbents, set a hard floor requiring each department to spend a fixed share of its AI budget on Canadian-controlled products, hold those systems to the auditability the Palantir contracts never received, and publish the receipts quarterly. Until Ottawa buys Canadian AI as openly as it has long bought the foreign kind, he argues, ‘AI’s first customer’ is a slogan attached to a country that has quietly been someone else’s customer all along.