Canada takes up to $1.6B stake in Cohere for sovereign AI

Governments are done just regulating AI vendors. Now they buy them. On October 29, Prime Minister Mark Carney said Canada will make an equity investment of up to $1.6 billion in Cohere, the Toronto AI company, in a Government of Canada release. The pitch is sovereign AI: models that businesses and governments can run with control over where sensitive data sits and who can touch it.

What actually happened

The money is equity, not a grant. Canada takes a stake and shares in any upside. It pays out in stages, with regular reporting on Cohere's progress. The release calls it a commercial investment made alongside private investors NVIDIA, Bell and Schwarz Group.

Cohere's side of the deal, per the release:

• At least 1,000 new Canadian technical jobs by 2030
• At least $1.7 billion in new computing infrastructure in Canada by 2030
• Canadian incorporation and headquarters stay put
• Continued work with the Canadian AI Safety Institute

The release also notes Cohere's recent merger with Germany's Aleph Alpha. AI Minister Evan Solomon framed it as keeping Canadian companies from leaving to scale. CTV News, via The Canadian Press, reported the same up-to-$1.6B figure.

The government did not publish the full round size or Cohere's valuation, and press reports disagree on both. We left those numbers out. The release also uses plain "$" throughout. In a Canadian federal release, read that as Canadian dollars.

Why it matters for your business

"Sovereign AI" is now a product feature you can ask for. Cohere sells to businesses and governments that need to know where data is processed. If you serve healthcare, legal, finance or public-sector clients, someone will soon ask you which country your AI calls go through. Have the answer written down.

Ownership changes vendor risk, both ways. A government shareholder makes Cohere harder to kill and easier to trust for regulated buyers. It also ties the roadmap to job and compute targets that have nothing to do with your use case. Treat any single model vendor as a dependency you might swap.

The model layer is fragmenting by geography. US labs, Chinese labs, and now state-backed Canadian and European players. If your app hard-codes one provider, a client's residency requirement can force a rewrite. An abstraction layer and a per-client routing rule are cheap now and expensive later.

Sources: Government of Canada, Office of the Prime Minister, CTV News / The Canadian Press.