A Concentration Number That Actually Means Something
On June 2, 2026, the Canadian Anti-Monopoly Project (CAMP) published Parting Clouds: Creating a Competitive Marketplace for Compute, finding that Amazon, Microsoft, and Google's parent Alphabet together control 85% of Canada's public cloud market — Amazon at 42%, Microsoft at 31%, Google at 12% — well above the roughly 66% global average the same three firms hold worldwide. Two days later, on June 4, Prime Minister Mark Carney launched AI for All, Canada's national AI strategy, promising a domestically built "public AI supercomputer" and investment in "sovereign compute infrastructure at scale — resilient, sustainable, and under Canadian governance." The timing wasn't a coincidence CAMP missed: its report explicitly frames cloud concentration as a problem Ottawa needs to solve before it spends billions building the AI economy on top of it.
The report's warning is not manufactured. Cloud is genuinely foundational infrastructure now — for banking, health records, and increasingly for AI training and inference — and a market where three firms hold this much share, this far above the global norm, deserves scrutiny on its own competition-law merits, independent of any geopolitics.
Steelmanning the Concern
CAMP's strongest argument isn't the headline number; it's the mechanism. Vendor lock-in through proprietary APIs, opaque and complex billing, egress fees that penalize customers for leaving, and bundling of adjacent services are textbook ways an oligopoly forecloses competition without ever needing to collude. The report also cites a specific, checkable figure: compute expenses can run as high as 80% of a young AI startup's costs, meaning whoever controls compute pricing effectively controls who gets to build AI companies at all. That claim surfaced in the Competition Bureau's own January 2025 consultation, What We Heard: Artificial Intelligence and Competition, where stakeholders flagged that AWS, Azure, and Google Cloud's control of the "input layer" — compute, data, and increasingly talent — creates durable barriers for challengers like CoreWeave or open-source alternatives.
The sovereignty argument also isn't hyperbole dressed as policy. CAMP writes that "dependence on a handful of U.S. hyperscalers is a sovereign risk as well as a competition problem," pointing to "rising tensions in U.S.–Canada relations" and Washington's demonstrated willingness to use technology access as leverage. Given the past year of tariff disputes and immigration-driven talent flight — Canada has been an explicit beneficiary as skilled tech workers flee H-1B uncertainty in the U.S. — a government building critical AI infrastructure on entirely foreign-controlled cloud rails is taking on real counterparty risk, not just a competitive disadvantage.
Where the Prescription Gets It Right
CAMP's most useful recommendation is also its least disruptive: mandate interoperability and portability as a condition of federal cloud procurement and of any subsidy tied to the new AI strategy's sovereign-compute buildout. This is proportionate regulation in its best form — it doesn't ban anything, doesn't cap market share, and doesn't require picking a domestic champion. It simply lowers the switching costs that currently trap customers, which is precisely the mechanism CAMP itself identifies as the actual harm. As CAMP policy analyst Curtis McCord put it, the fix is requiring vendors to build interoperable systems, not simply changing which companies get the contracts — otherwise Canada risks building "maplewashed dependencies" that recreate the same lock-in under a domestic flag. The Competition Bureau also already has the legal tools to open a formal market study under sections 7, 125, and 126 of the Competition Act — a process it has used only four times since 2007 (fintech, beer, self-regulated professions, generic drugs) and, per its own statement to reporters, has never applied to cloud. That gap is worth closing on the merits alone.
Where It Overreaches
Two of CAMP's other asks deserve more skepticism. First, designating a "dedicated industry regulator with technical oversight capacity" for cloud risks duplicating the Competition Bureau's existing mandate and layering a new compliance regime onto a sector that is, by CAMP's own numbers, growing at nearly 17% annually — friction that lands hardest on the smaller Canadian cloud entrants the report claims to want to help. Second, the call to use the 2026 CUSMA renegotiation to strip out "constraints on digital market regulation" conflates trade policy with competition enforcement. Reopening a treaty chapter that Canadian tech exporters also rely on for U.S. market access, in service of a domestic antitrust goal that the Competition Bureau can pursue unilaterally, invites retaliation risk for a marginal procedural gain.
"Rising tensions in U.S.–Canada relations, the intermingling of Big Tech interests with U.S. government power, and the demonstrated willingness to use technology access as geopolitical leverage have forced Canada and others to confront an uncomfortable reality: dependence on a handful of U.S. hyperscalers is a sovereign risk as well as a competition problem." — CAMP, Parting Clouds
The Proportionate Path
The better sequence is narrower than CAMP's full menu: the Competition Bureau should open the cloud market study it already has authority to conduct, and Ottawa should attach binding interoperability and portability conditions to federal procurement and to any subsidy dollars flowing through AI for All's sovereign-compute commitments. That addresses the actual switching-cost harm CAMP documents, without a new regulator or a trade fight that has little to do with the underlying competition problem. Canada doesn't need to choose between an open AI economy and a competitive one — it needs the enforcement tools it already has, pointed at a sector they've simply never been used on.