The Catch-22

Michael Geist, the Canada Research Chair in Internet and E-commerce Law at the University of Ottawa and one of the country's foremost authorities on digital policy, identified the structural problem in December 2025.^[5] Canadian digital sovereignty, Geist argued, faces a paradox: smaller companies might avoid U.S. legal jurisdiction but lack the resources for competitive infrastructure, while larger firms could build viable alternatives but remain vulnerable to U.S. court orders because of their American ties.

The location of the infrastructure provides little assurance that domestic law and policy will apply to Canadian data.^[6]

• Michael Geist

The Treasury Board of Canada acknowledged the same point in its own assessment: "using a Canadian supplier or storing data in Canada does not guarantee data will be outside the jurisdiction of foreign courts."^[7]

The acknowledgement exists; the investment proceeded anyway. Canada's own assessment says geography does not guarantee jurisdiction.

Privatisation with a Canadian Flag

Joseph Wright of Derby, "An Iron Forge" (1772). Tate Britain, London. The forge is on home soil. The question is who holds the hammer. Public domain.

The Professional Institute of the Public Service of Canada (PIPSC), the union representing 75,000 federal science and technology professionals, titled its analysis of the Cohere deal "Privatisation with a Canadian Flag."^[8]

The criticism was specific: Cohere generates 90 percent of its revenue outside Canada.^[9] Its data centre is operated by an American company subject to American jurisdiction. Its models, while developed by Canadian researchers, rank #37 on the LMSYS Chatbot Arena^[10] – below free, open-weight alternatives from DeepSeek (China), Meta (United States), and Mistral (France) that any Canadian institution can deploy on its own hardware without paying rent to anyone.

When asked to name specific government projects that justified the investment, AI Minister Evan Solomon could not provide a single example.^[11] The promised public registry of government AI projects, meant to ensure accountability, had not been delivered.

Meanwhile, PIPSC noted, Canada's 20,000 in-house IT professionals – the people with the skills and security clearances to build sovereign AI infrastructure on government-owned hardware – continued to be sidelined.^[12]

The union's framing was blunt: this was not sovereign AI. It was outsourcing with national branding.

The $7 Billion Question

Cohere is valued at $7 billion.^[13] It has raised $1.54 billion across seven funding rounds from investors including NVIDIA, AMD, Oracle, Salesforce, and Canada's own pension fund (PSP Investments) and development bank (BDC).^[14]

The valuation reflects Cohere's enterprise positioning: 85 percent of revenue comes from private deployments where data sovereignty matters.^[15] Clients include Oracle, Fujitsu, RBC, and LG. The company's annual recurring revenue hit $240 million in 2025, up 287 percent year over year.^[16]

The models tell a different story.

Cohere's flagship, Command A, ranks #17 out of 38 on the Artificial Analysis Intelligence Index for open-weight non-reasoning models.^[17] Its output pricing ($10 per million tokens) is the second most expensive on the market. Its inference speed (37.6 tokens per second) is nearly 40 percent slower than the median. Its licence (CC-BY-NC 4.0) prohibits commercial use without a separate agreement.^[18]

DeepSeek-R1, an open-weight reasoning model released in early 2025, matches or exceeds Cohere's best benchmarks. It is free. It runs on a laptop. It requires no licensing agreement, no API subscription, and no American cloud provider to operate.^[19]

The $7 billion valuation is not for the technology. It is for the contracts. The question is what happens when the contracts come up for renewal and the clients discover they can self-host better models for less.

What Germany Built Instead

Johannes Vermeer, "The Geographer" (c. 1668-69). Stadel Museum, Frankfurt. A man studying the world through instruments he built, maps he drew, in a room he controls. Public domain.

Detlef Schmuck, the managing director of TeamDrive Systems in Hamburg, has built his entire business on a principle that Canada's sovereign AI strategy has not yet absorbed: "If the server operator can read the data, the data is not secure. And if the data is not secure, the competitive advantage it represents is already gone."^[20]

TeamDrive provides zero-knowledge encrypted file synchronisation for organisations that will not trust their data to American cloud providers. The keys are generated locally. They are never transmitted to the company's servers. Neither TeamDrive's administrators nor its infrastructure can read the files they store and sync. The company processes data exclusively within the Federal Republic of Germany, under German and EU data protection law.^[21]

The GAIA-X consortium, now in its implementation phase with more than 180 data spaces, is building a federated architecture for European cloud infrastructure designed to reduce dependence on American and Chinese providers.^[22] The International Criminal Court in The Hague announced in late 2025 that it was replacing Microsoft Office with OpenDesk, an open-source office suite delivered by the German Centre for Digital Sovereignty.^[23]

The European approach differs from Canada's in a way that matters: it distinguishes between location and control. A server in Frankfurt operated by a German company under German law is sovereign in a way that a server in Ontario operated by an American company under American law is not. The distinction is not geographic. It is jurisdictional.

The Enshittification Warning

Cory Doctorow, the Canadian-born author and digital rights activist, delivered a keynote at OCAD University in Toronto in November 2025 titled "Digital Elbows Up: How Canada Can Become a Nation of Jailbreakers, Reclaim Our Digital Sovereignty, Win the Trade War, and Disenshittify Our Technology."^[24]

Doctorow's argument was characteristic in its directness. AI firms, he warned, will intentionally degrade their products to recoup large infrastructure expenditures – the same pattern he has documented across every platform that achieved market dominance through initial generosity.^[25] The process he calls "enshittification"^[26] follows a predictable arc: first the platform is good to users, then it extracts value from users to attract business customers, then it extracts value from both to maximise shareholder returns.

"Canadians get the rights that foreign companies believe they deserve," Doctorow told the audience. His prescription: seize not just the means of manufacturing, but the means of computation itself, back from what he called techno-oligarchs.^[27]

Canada's Cohere investment does the opposite. It routes $240 million of public money through a Canadian company to an American cloud provider, creating a dependency that will deepen as the infrastructure scales and the switching costs compound.

What $240 Million Could Have Built

The Centre for International Governance Innovation (CIGI) published an analysis in early 2026 arguing that open source represents Canada's "third path" in AI – neither dependence on American hyperscalers nor the impossible expense of building a domestic rival to OpenAI.^[28]

For $240 million, Canada could have had:

A well funded development and fine-tuned open-weight models optimised for Canadian government workloads – bilingual (English and French), trained on Canadian legal and regulatory corpora, deployable on government-owned hardware with no foreign jurisdiction exposure.^[29]

Additionally, they would have had money to:

Build GPU clusters at existing Canadian institutions (the Vector Institute, Mila, the National Research Council) for training and hosting open-weight models on Canadian-operated infrastructure – not American-operated infrastructure that happens to be located in Canada.^[30]

And:

Created a national model registry where Canadian institutions share, evaluate, and deploy open-weight models without commercial licensing restrictions – the AI equivalent of what the Canadian Digital Service already does for open-source government software.^[31]

Then, they could have:

Invested in the 20,000 PIPSC members who already have security clearances, institutional knowledge, and the skills to build and operate AI infrastructure on government-owned hardware – the workforce the Cohere deal bypassed.^[32]

Each of these alternatives, just on their, preserves actual sovereignty: Canadian-operated infrastructure, Canadian-controlled models, Canadian jurisdiction. None of them requires routing public money through a company whose data centre is operated by an American firm subject to American law.

The Moat

Canaletto, "The Stonemason's Yard" (c. 1725). National Gallery, London. The workshops are separate. The trades are separate. The masters are separate. The city gets built because the work stays local. Public domain.

A moat is a defensive structure. It works by creating a barrier that attackers cannot easily cross. In business strategy, a moat is a structural advantage that competitors cannot replicate – a patent, a network effect, a switching cost, a regulatory capture.

The moat is a data centre in Ontario operated by a company in New Jersey. The CLOUD Act crosses it without getting wet. Canada's "moat" is not a barrier. It is a line drawn on a map.

The sovereignty that Canada is looking for – the assurance that Canadian data stays under Canadian law, that Canadian AI infrastructure serves Canadian interests, that the public investment compounds inside the country rather than flowing outward – already exists in the form of open-weight models, self-hosted infrastructure, and the legal frameworks that other countries are building to ensure jurisdiction matches geography.

DeepSeek-R1 runs on a laptop. Llama 4 is open-weight and commercially licensed. Qwen3 has toggleable reasoning modes. Mistral Large 3 ships at 675 billion parameters with 41 billion active. Every one of these models can be downloaded, deployed, and operated on Canadian-owned hardware, under Canadian law, without a single dollar flowing to an American cloud provider.^[33]

Sage.is AI-UI is built on this premise. AGPL-3 licensed, self-hostable, model-agnostic. The platform connects to any model – cloud or local, proprietary or open-weight – through an interface the organisation controls.^[34] Sage is a small platform without the enterprise scale of Cohere or the government contracts of CoreWeave. The architecture is the argument: sovereignty is not where the server sits. It is who controls the server, under whose law, with what recourse.

Canada's moat is a mirage. The water was never there. The crossing was always open. The only question is whether the country will notice before the next $240 million flows through it.

Disclosure

This article examines the structural relationship between Canada's Sovereign AI Compute Strategy, the Cohere investment, and the jurisdictional implications of the U.S. CLOUD Act. The author has no financial relationship with Cohere, CoreWeave, or the Government of Canada. Sage.is AI-UI is a product of Startr LLC and is referenced as one example of a sovereignty-first architecture. Michael Geist and Cory Doctorow are quoted from published works and public speeches; neither was interviewed for this article.

The views expressed are those of the editorial board. Sage.is AI-UI is a product of Startr LLC. The author has no financial relationship with Cohere, CoreWeave, or the Government of Canada. Michael Geist and Cory Doctorow are quoted from published works and public speeches; neither was interviewed for this article. We welcome their response. Full disclosure and transparency is a feature, not a bug.