Today Cohere announced the initial close of a Series E targeting up to $5 billion. At full size the round would value the company at $25 billion and would be the largest private financing in Canadian history.

Who invested matters more than the amount. The round is led by Schwarz Group, the European retailer that runs its own sovereign cloud, and the Government of Canada is investing too; it has gone from funding the build-out of data centres for Cohere to owning a part of it.

In the last twelve months, sovereign AI has moved from a policy phrase to a budget line item. The shift is not really about technology. It is about what happens to data once it reaches a model, and which government can compel the vendor on the other end.

AI changed what handing over data means. A document sent to a software vendor used to sit somewhere. Sent to a model it gets read, processed and sometimes learned from, often on infrastructure the customer never sees. For a bank, a defence ministry or a hospital that distinction is a procurement question answered in writing: where will the data sit, and who controls the company processing it.

The demand is already in production

IDC put the question to 633 organizations in July, and 42% said they already use sovereign cloud to build or run AI, with another 38% expecting to within the year. The highest current adoption was actually in North America rather than Europe, at 57%, which should temper the instinct to read this as a story that only applies to markets outside the US. Gartner sizes worldwide sovereign cloud infrastructure at $80 billion this year, and the models and platforms layer above it at $64 billion, growing 63%.

Those are large numbers, but the more useful observation is where they are being spent. This is public administration, defence, banking, energy and healthcare, which means long procurement cycles, multi-year terms and high switching costs. Categories with those characteristics tend to consolidate around whoever can satisfy the requirements first, and they stay consolidated for a while.

The supply side reports the same thing. On its July quarter earnings call NVIDIA said sovereign AI revenue, sold largely through regional neoclouds, grew 35% sequentially and more than tripled year over year, with demand coming from Canada, France, the Netherlands, Singapore and the UK.

Why it is hard to serve

Sovereignty is not a single property a vendor either has or lacks. Morgan Stanley's public policy team argues that the debate moved past chips some time ago, and that every layer is now contested: the silicon, the cloud, the models, the data centres, the standards and the electricity underneath all of it.

No single company owns that whole stack. What a vendor can offer is a local contracting entity, the ability to run in environments the customer controls, and models tuned to local languages and regulation. Those take years to build, they cannot be bought with more compute, and they are what decides who gets to bid.

That also narrows the field, because most of the leading AI companies are American. The hyperscalers have responded with regional offerings such as AWS's European Sovereign Cloud, which keeps data and operations inside the EU. Those offerings settle where the data sits. But they cannot fully settle who controls the company processing it.

The reason buyers want an alternative has less to do with confidentiality than with continuity. In June the US Commerce Department required Anthropic to obtain a licence before its Mythos and Fable models could reach any foreign person. The company said the only practical way to comply was to disable both models for every customer worldwide. For anyone running production workloads the point had already been made: a model you depend on can be switched off by a government that is not your own.

Cohere is one of the companies built for it

Cohere's enterprise platform North is AI agents working alongside staff, data and the tools people already use. They search across a company's own information, draft documents and tables, and run multi-step tasks end to end. Agent Studio lets teams build their own agents. North 2, launched this year, added spend controls. Around all of it sits what Cohere calls a zero-trust framework, with tight access controls and a record of what the system did.

The part that matters here is where North can run. Cohere will operate it on its own hosted service, through its Model Vault, inside a customer's private cloud, or on the customer's own machines. That last option extends to air-gapped sites cut off from outside networks, which means a defence ministry never has to send its data anywhere.

Cohere now sells North into finance, healthcare, manufacturing and the public sector. The company noted in its release that annual recurring revenue grew 165% year over year through the third quarter, and that its customer base grew twentyfold over the past year to 300. Customers include RBC, TD Bank, Bell Canada, Fujitsu, PwC, RWS, the German naval shipbuilder TKMS, and Aston Martin Aramco.

Congratulations to the team

Congratulations to Aidan and the entire team at Cohere on what they have built, as well as the recognition that this capital raise signifies. With this, they are now in a position to serve governments and organizations worldwide and are well on the way to becoming the preferred model provider for sovereign AI.