Skip to content
Back to all essays
Policy & Economy

Canada Is Picking an AI Anchor at Every Layer

By George PuEssay8 min read

Published Updated

An AI-generated editorial photomontage maps connected Canadian AI model, compute, research and health infrastructure beneath Parliament Hill.
AI-generated editorial photomontage of the institutional layers described in this essay; not a documentary photograph.

Over the past 2 years, Canada quietly rebuilt the question of who controls its artificial intelligence.

Not in one announcement. That's the point.

There was never a single moment loud enough to make you look up.

The decisions arrived in fragments. A Christmas Eve letter. A contribution agreement with a file number. A press release at a university most people don't follow. An MOU with the important parts blacked out.

Each fragment was small enough to ignore.

Together they redrew the map.

I only saw the shape of it because I went looking - and I had to file access-to-information requests to do it.

I'll come back to why that's the most damning part.

The fragments

Here's the sequence, as plain as I can make it.

December 2024. Ottawa announces up to $240 million for Cohere to buy computing power at a new data centre in Cambridge, Ontario. On December 24, according to federal briefing documents released to me, the government sends Cohere the letter of intent. Christmas Eve.

March 2025. Around the handover from Justin Trudeau to Mark Carney, the deals get signed. Cohere's agreement is executed on March 10, and the government finalizes the $240 million that month. In mid-March, the deputy minister approves 5 more contributions, all effective April 1. I'll get to who they went to.

August 2025. Solomon and Cohere sign a memorandum of understanding. The deputy minister's approval memo describes Cohere as 'the only sovereign, cloud-agnostic large language model operating in Canada' and frames the MOU as a way to 'champion one of our champions'. The operative terms are blacked out.

November 2025. The compute money reaches universities. $42.5 million for a new AI system at the University of Toronto, hosted by SciNet and delivered through the Digital Research Alliance of Canada, announced at the Vector Institute and framed as an investment in sovereignty.

May 2026. The minister names 44 companies funded to buy compute, about $66 million in total. A program, announced. Not the architecture behind it.

What the fragments add up to

Each of those was covered, if at all, as its own small good-news story. A tech outlet. A university newsroom. A minister's quote about Canadian discovery.

None of them was wrong. None of them was the whole picture.

Here's the whole picture. Canada has been appointing anchors - one federally funded node per layer of the AI economy.

At the model layer: Cohere. $240 million, on an agreement that runs to 2049. I wrote about that one already - Cohere's own proposal shaped the program it now benefits from most.

At the research-compute layer: the Digital Research Alliance of Canada. About $83.6 million, the largest single contribution in the group, feeding the national compute that universities actually run on. The U of T money was delivered through it.

At the academic-institute layer: Vector, Mila and Amii. $19,728,023 each. The same number, 3 times. That's a formula, not a competition.

RecipientContributionAgreement runsPublic record
Digital Research Alliance of Canada$83,558,934April 2025 to March 2028Grants record
Unity Health Toronto (VITAL)$29,519,644April 2025 to March 2028Grants record
Vector Institute$19,728,023April 2025 to March 2029Grants record
Mila$19,728,023April 2025 to March 2029Grants record
Amii$19,728,023April 2025 to March 2029Grants record

That's 5 organizations and about $172.3 million, approved in March 2025, all part of the $2 billion Sovereign AI Compute Strategy.

You'll notice one name on that list that doesn't fit.

The health anchor

A single hospital network - Unity Health Toronto - received about $29.5 million for VITAL, which its grants record describes as 'a secure digital AI health infrastructure to leverage Canadian health data for trials, artificial intelligence, and a learning health system'.

When I first published this in May, that contribution had almost no public footprint.

A month later, it had a very public one. In June, Ottawa announced another $100 million to expand VITAL across Canada as part of its AI for All strategy. Solomon called it a 'cornerstone' of that strategy. The platform launched on June 23, starting with 160 hospitals in Alberta, Ontario and Quebec. Counting federal, provincial and institutional money, Unity Health puts total funding above $210 million.

Unity Health is a genuine leader in hospital AI. That isn't the question.

The question is what it means that Ottawa has made one hospital network the hub of a national clinical-data platform - one that will touch the health records of patients across 3 provinces, and counting.

It isn't Ottawa's only health bet. The same strategy put $100 million into a separate Health Sector Data Space. But VITAL is the one with the compute money, the hospitals and the head start.

'Anchor' is my read of the documents, not a government label. But the pattern matches Cohere at the model layer. Pick a node early. Fund it quietly. Build around it later.

The first $29.5 million arrived as a line in a contribution agreement. By the time it was announced from a stage, the decision had already been made.

This is not a conspiracy

The easy version of this story is a cover-up. It isn't one.

Nothing was hidden. Every contribution in this piece sits in the federal grants database. Parts of it were announced from stages, with cameras running.

That's what makes it strange.

It didn't need to be secret, because nobody was assembling it.

The pieces were scattered across newsrooms, disclosure portals and a minister's travel schedule. No one - not the press, not the opposition, not the public - put them in one place and asked the obvious question.

Is Canada deliberately picking one winner per layer, and is that what we want?

That isn't a cover-up. It's a visibility failure.

And a visibility failure on a $2 billion, decade-long restructuring of who controls a strategic technology is, in its own way, worse than a scandal. There's no villain to name and nothing to repeal.

It just happened, in the open, unwatched.

What it means on Monday

If you work in Canadian AI, the practical version is simple. Access to sovereign infrastructure increasingly runs through an anchor.

If you're an academic, through your institute or the Alliance.

If you're in clinical AI, through a hospital network in Toronto.

If you're a company, through Cohere at the top and the AI Compute Access Fund at the bottom. (My company is one of the Access Fund's recipients.)

If your name isn't on one of these agreements, you don't have a seat at the table. You have a landlord.

The choice every operator now faces is simple. Align with an anchor, or build at the layer the anchors don't reach.

What's built

The architecture is built. The agreements are signed. They run to 2028, 2029 and, for Cohere, 2049. The anchors are named.

None of it was put to a vote. Most of it was never really put to the public - not because anyone hid it, but because it was technical enough, fragmented enough and quiet enough that no one looked.

I looked. The least the rest of us can do, now that the map is drawn, is see it.

Receipts

Public records

Access-to-information records released to me

  • A-2025-00149 - the approval memo from Deputy Minister Philip Jennings to the Minister of Innovation, Science and Industry, plus the project and financial summary forms for each of the 5 recipients. All 5 are non-repayable contributions under the Advanced and Emerging Technologies Terms and Conditions.
  • A-2024-00994 - briefing for a meeting with Cohere
  • A-2025-01015 - briefing on the $240 million contribution
  • A-2025-00997 - the approval memo from Deputy Minister Schaan to Minister Solomon for the Cohere MOU
  • A-2025-01507 - the signed MOU

One small difference: the grants database lists Vector at $19,728,023, while the summary form in A-2025-00149 shows $19,728,024. I use the public figure.


Every fact in this piece links to its source or names the record it comes from. If I got one wrong, tell me and I'll correct it publicly, with a note at the top.

Share this:

New essays and letters, by email

Subscribe if you would like to hear from me when I have something worth sending. Everything here remains free to read.

Optional. Unsubscribe anytime.