Live
Policy

Budget 2026’s biggest AI promises aren’t in its fiscal tables

Illustration for the prompt/power story: Budget 2026's biggest AI promises aren't in its fiscal tables

Tucked into Chapter 6 of Canada’s 2026 federal budget is an estimate that CANChat, the government’s in-house generative AI tool, frees up somewhere between 15,000 and 60,000 hours of public servants’ time every week. Remember that range. It’s a fourfold spread, and it’s a fair guide to the rest of the document.

Abstract illustration: five long violet bars stacked like rows in a table, each ending in a short glowing pink segment, with an empty dashed outline of a longer bar floating above the stack, on a dark purple gradient.
Five years of booked AI money. Only the pink ends are new since June, and the biggest promise has no row at all. Illustration: prompt/power

Finance Minister François-Philippe Champagne tabled Budget 2026 on Thursday, Nov. 5, and the AI headlines came quickly. CP24 led with CA$1 billion for a digital overhaul of the federal public service. BetaKit and others went for a pledge of 500 megawatts of “sovereign” AI compute, and for an equity stake of up to CA$1.6 billion in Cohere.

Those are not the same kind of number. Some are booked in the budget’s fiscal tables; others are cash-basis figures that won’t match those tables. One has no dollar figure at all.

So we went to the budget plan itself, Chapter 5 and Chapter 6, and sorted the AI, compute and public-service digital items line by line. Figures are in Canadian dollars. US-dollar figures are approximate, at the Bank of Canada’s Nov. 5 rate of 1.3989.

The public-service money

Chapter 6, titled “A more efficient government,” creates Digital Transformation Canada, a new organization that the budget says “will bring together key digital groups across the public service,” including Shared Services Canada and the Canadian Digital Service. It will lead the use of AI and other new technology across government. The budget promises a Digital Transformation Canada Act to go with it; that bill hasn’t been introduced yet.

Three dollar figures are attached to it.

The headline one is the Government Transformation Fund: “approximately $1 billion over five years, on a cash basis, starting in 2026-27” (about US$715 million), to speed up IT projects that improve service delivery.

Separately, a fiscal-table row labelled “Advancing artificial intelligence in government through Digital Transformation Canada” totals CA$606 million over five years (about US$433 million). It starts small, CA$12 million in 2026-27, and climbs to CA$192 million in 2029-30. The table flags it as a capital investment.

Then there’s a fellowship program run through Build Canada Exchange, meant to bring outside technical experts in for priority projects: CA$17.6 million over five years plus CA$4.2 million in 2031-32 (about US$12.6 million for the five-year piece).

Here’s the trap. The CA$1 billion fund is explicitly counted on a cash basis, while the CA$606 million row sits in the budget’s fiscal framework, which, as New Space Economy’s breakdown points out, is a different accounting basis. The budget doesn’t say whether one includes the other, and New Space Economy warns against adding them together. Our read: the fair description is “between CA$606 million and CA$1 billion for AI and IT transformation, depending on how you count.” It is not CA$1.6 billion.

AI for All, and a gap nobody explains

Chapter 5, “A safe and sovereign Canada,” holds the national strategy. The budget says AI for All, announced June 4, 2026, is “supported by an initial investment of $2.7 billion on a cash basis” (about US$1.93 billion), covering privacy, AI safety, compute access and capital for AI companies.

The fiscal-table row for the same strategy totals CA$2,038 million from 2026-27 to 2030-31 (about US$1.46 billion). That leaves roughly CA$660 million between the text and the table. The chapter doesn’t explain it. The likelier explanation, and it’s our inference, is a cash-versus-accrual split: capital costs get spread over the life of an asset, or some spending lands after 2030-31.

And AI for All launched five months ago. The budget records its cost; it doesn’t create it.

The genuinely new AI proposal in Chapter 5 is small. CA$6.2 million over four years starting in 2027-28 (about US$4.4 million) keeps Canada in the Global Partnership on Artificial Intelligence and supports its Montréal expertise centre.

Big numbers with no line in the table

Three of the most-quoted AI items don’t appear in the fiscal tables at all.

Cohere first. The budget says “Canada committed to an equity investment of up to $1.6 billion in Cohere” (about US$1.14 billion), listing it among actions taken “since September 2026,” and expects the company to create at least 1,000 high-paying technical jobs in Canada by 2030. There’s no table row for it, and no department or start year attached. Our read: an equity stake is an asset purchase, not program spending, which would explain why it isn’t costed like a program.

Mitacs gets CA$162 million over five years (about US$116 million) for 10,000 co-funded AI work placements for students, recent graduates and postdoctoral fellows. Same since-September list. Same missing table line.

Then the compute pledge. The budget says Ottawa will “enable Canadian AI companies to access 500 MW of sovereign AI compute” using tools “such as backstops and offtake agreements.” It gives no dollar value, no timeline and no department. The pledge follows the AI Sovereign Compute Infrastructure Program, which Innovation, Science and Economic Development Canada opened to applicants in April, though the budget doesn’t tie the two together.

For scale, BetaKit reports that Canada has about 300 MW of AI data-centre capacity today, citing a Boston Consulting Group report. The founders and investors BetaKit spoke with mostly welcomed the idea. Several said pricing and allocation would decide whether it works.

“Almost nobody has capacity available at any price,” Ali Asaria of Transformer Lab told BetaKit.

A backstop is, in effect, a contingent liability. It costs nothing on paper until a startup can’t pay for the compute it signed up for. That’s how the pledge can be the loudest number in the AI chapter without being a number at all.

Counting it up

Here is the tally from the budget’s own fiscal tables over five years, 2026-27 to 2030-31:

AI for All: CA$2,038 million (about US$1.46 billion), announced in June.

AI in government through Digital Transformation Canada: CA$606 million (about US$433 million).

Global Partnership on AI: CA$6 million in the table, CA$6.2 million in the text (about US$4.4 million).

That comes to CA$2.65 billion (about US$1.89 billion) in costed lines labelled as AI. Take out AI for All, and the AI spending this budget adds to its tables is CA$612 million, about US$437 million. Outside the tables sit the CA$1 billion cash-basis transformation fund, the Cohere stake of up to CA$1.6 billion, CA$162 million for Mitacs, and a 500 MW compute promise with no price on it.

Digital Transformation Canada also arrives as the public service shrinks. According to CP24, the government says it is “on track” to cut the public service by 10 percent by 2029; the budget itself points back to a Budget 2025 projection of about 40,000 positions, or 10 percent, by 2028-29. There were 332,000 federal public servants as of August 2026, down from 368,000 at the 2023-24 peak. The budget outlines no cuts beyond that projection. What it does offer is AI as the way a smaller workforce keeps up: CANChat, and GCtranslate, which the budget says has translated more than 400 million words since June 2025.

Back to that range. The government’s own estimate of what CANChat saves runs from 15,000 to 60,000 hours a week. For the 500 MW, there’s no estimate at all.

// Contributor, AI Explained
Catherine Crowe

Catherine Crowe covers AI explained for prompt/power: the plain-English guides that break down how the technology works, what the jargon means and what it changes for everyday people. Originally from Canada, she writes from New Zealand.

Latest from prompt/power

  1. Quebecor’s offer for TELUS says nothing about TELUS’s AI factories. Here’s what’s at stakeNov 16
  2. Can an AI chatbot be your therapist? Where the rules stand in the U.S., Canada and beyondNov 16
  3. ChatGPT, Claude, Gemini and Copilot: what every AI plan costs in CanadaNov 16
  4. Nvidia’s first earnings without Jensen Huang: the numbers to watch on TuesdayNov 16
  5. SoftBank wrote off 39% of its AI power equipment spendingNov 16

Leave a Reply

Your email address will not be published. Required fields are marked *