Canada Is Building Europe’s AI Future. Europe Is Still in a Meeting.
Editorial Series | Sugau — Private AI & Cloud Repatriation
Two things happened in the same week in April 2026 that nobody connected.
France announced that 2.5 million government workstations would migrate from Windows to Linux — the largest sovereign technology declaration by a major European nation in a generation. And Cohere, a Canadian AI company most Europeans have never heard of, announced a $20 billion commitment to European AI sovereignty infrastructure.
One story was about Europe finally saying no to American vendor dependency. The other was about Canada stepping in to fill the void Europe created but cannot bring itself to occupy.
Read together, they tell you everything about where the next decade of global AI infrastructure is actually being decided — and who is deciding it.
What Canada Sees That Europe Doesn’t
Cohere was never trying to be ChatGPT. While OpenAI was chasing consumer headlines and Anthropic was navigating the ethics of frontier models, Cohere was quietly building enterprise relationships with regulated industries — banks, telcos, government contractors — that needed AI that ran on their terms, in their jurisdiction, under their control.
That positioning looked niche two years ago. Today it looks prescient.
Because the enterprise sovereign AI market — AI that runs where you tell it to run, governed by laws you actually operate under, without a foreign government’s legal reach into your data — is the fastest growing unsatisfied demand in global technology right now. And the primary source of that demand is Europe.
Europe has spent three years producing the most sophisticated AI regulatory framework in the world. The AI Act. GDPR enforcement with actual teeth. NIS2. Data sovereignty requirements written into procurement frameworks across member states. European institutions have built the legal and regulatory architecture for sovereign AI with extraordinary thoroughness.
What they have not built is the actual AI.
Canada looked at that gap and made a simple calculation: Europe has the demand, the regulation, the institutional will, and increasingly the budget. It does not have the supply. We can be the supply.
That is a $20 billion bet. It is also an obvious one — once you are outside the American tech industry’s habit of assuming it will supply everything by default.
The Specific Advantages No American Vendor Can Buy
The data sovereignty pitch for a Canadian vendor in Europe is not just marketing. It has legal and structural substance that matters in enterprise procurement.
The CLOUD Act is the American tech industry’s original sin in Europe. Passed in 2018, it requires American companies to hand over data stored anywhere in the world when served with a valid US government order — regardless of where the data physically sits, regardless of what the contract says, regardless of GDPR. A European bank storing data with AWS Frankfurt is not storing data under European jurisdiction. It is storing data under American jurisdiction with a German postcode.
Canadian companies are not subject to the CLOUD Act. Canadian data sovereignty law — PIPEDA and its successor frameworks — does not contain an equivalent extraterritorial reach provision. “Canadian jurisdiction” means something categorically different from “American jurisdiction” in a European procurement conversation.
This is not a subtle distinction. For a French ministry, a German hospital network, a Dutch financial institution, or a Belgian defence contractor — the difference between American and Canadian legal exposure is the difference between a compliant deployment and a non-compliant one.
Cohere can walk into those procurement conversations and say things that Microsoft, Google, and Amazon structurally cannot say regardless of how many data centres they build in Frankfurt.
That is the opening. And it is real.
Europe Is Still in a Meeting
Here is the uncomfortable question: why is it Canada filling this void and not Europe itself?
The capital exists. Europe’s sovereign wealth funds, the European Investment Bank, national development banks across member states — the institutional capital available for a serious European AI sovereignty play dwarfs $20 billion. France alone is redirecting €1 billion per year from Microsoft licences. Germany’s Sovereign Tech Fund model has demonstrated that targeted public investment in open technology infrastructure works. The money is not the constraint.
The talent exists. European universities produce world-class AI researchers. DeepMind was founded in London. Mistral was founded in Paris. The notion that Europe lacks the human capital for frontier AI is false — it lacks the conditions to retain and deploy that capital at scale.
The regulatory framework exists. The AI Act, for all its imperfections, is the most serious attempt by any jurisdiction to govern AI development with democratic accountability. Europe did not lack the policy sophistication to think about sovereign AI. It thought about it extensively and in great detail.
What Europe lacks is the one thing no committee can produce: urgency.
European institutional decision-making operates on a cycle that is structurally incompatible with technology markets. A procurement decision that takes eighteen months in Brussels takes eighteen days in San Francisco and eighteen hours in Shenzhen. By the time a European sovereign AI initiative has cleared its third impact assessment, the market it was designed to address has moved on.
The fragmentation compounds everything. There is no European CTO. There is a French CTO, a German CTO, a Polish CTO, a Dutch CTO — each operating in their own language, their own procurement framework, their own political cycle, their own definition of sovereignty. Cohere can sell one product to all of them. A European competitor would need to rebuild the go-to-market from scratch in every member state.
This is not stupidity. It is a structural problem that smart people built over decades for reasons that made sense at the time. But the outcome is identical to stupidity when the market is moving at AI speed.
The Window Is Not Permanent
Canada’s opening exists because of a specific geopolitical moment. Trump’s second term shattered the assumption of reliable American partnership in ways that cannot be unseen. European institutions that previously treated American tech vendor dependency as a minor governance concern are now treating it as a strategic liability. The anger is real and it has budget behind it for the first time.
But anger without urgency is just noise. And windows close.
There are two scenarios for the next five years.
In the first, Canada executes. Cohere builds out the sovereign AI infrastructure layer across European enterprise and public sector. A handful of other Canadian and European-adjacent vendors follow. The market consolidates around non-American supply chains that are close enough to American capability to be credible and far enough from American jurisdiction to be compliant. Europe gets sovereign AI — built by Canada.
In the second, Europe wakes up. The France Linux migration produces a generation of senior officials who understand that OS sovereignty was just the first layer. Mistral scales. A serious European sovereign compute initiative gets funded. The regulatory framework that Europe built to govern AI becomes the foundation for European AI industry rather than just American AI compliance overhead.
One of those happens first. The decade looks very different depending on which.
The honest answer is that Canada is currently executing and Europe is currently still in a meeting. That can change. European institutions have surprised before — usually when the alternative became undeniable rather than merely obvious.
The question is how undeniable it needs to get.
The Layer Below the Model
There is one more dimension that neither the Cohere announcement nor the France migration coverage is addressing directly.
Sovereign AI is not just about which company’s model you use or which country’s laws govern your data. It is about the full stack underneath the model — the compute, the orchestration, the networking, the storage, the security layer, the deployment pipeline.
You can run a Canadian model under European law on American cloud infrastructure controlled by American companies subject to American legal process. That is not sovereignty. That is sovereignty theatre.
The organisations that understand this are building private infrastructure — bare-metal compute in their own facilities, under their own control, running open-weight models that no single vendor can revoke access to. The Cohere bet and the France Linux migration are both pointing at the same destination. The path there runs through the full stack, not just the headline layer.
Canada is filling the void at the model layer. The void at the infrastructure layer is still open.
That is where the next decade of sovereign AI actually gets built.
Catalin Lichi is the founder of Sugau — a bare-metal Kubernetes consultancy specialising in sovereign infrastructure and private AI for regulated industries. The full stack, not just the headline layer.