Canadian Procurement Pulse: Special Edition - AI Strategy

Date: June 9 2026

Canada's new AI strategy has the usual policy language, but the procurement story is sitting right on the surface. The federal government wants AI adoption to move from pilots and research labs into operating budgets. To make that happen, it is putting money behind grants, compute, health missions, source lists, public-service automation, certification, and Canadian suppliers.

In other words, Canada's AI strategy is a procurement strategy in disguise.

The strategy makes more sense when you put it beside grants, recent federal tender notices, current opportunities, and historical awards. Government is already buying AI across software, R&D, compute, chatbots, policy frameworks, predictive maintenance, document extraction, public-sector automation, and grant-backed commercialization.

The biggest takeaway: the first real AI market in government will likely show up through smaller doors. Source lists, mission programs, compute funds, health data platforms, are where the buying is starting to take shape.

The strategy sets a clear adoption target. Only 12 percent of Canadian businesses are currently using AI, and SME adoption is closer to 8 percent. The target is 60 percent business adoption by 2034.

That gap explains most of the document. Canada has plenty of AI research. The harder problem is turning research into normal operating spend: tools that companies use, services that governments buy, infrastructure that Canadian firms can access, and public-sector deployments that prove the technology works.

The money attached to the strategy points in that direction.

The most important phrase is "anchor customer." In procurement terms, that is the signal. Government wants Canadian AI companies to scale, and grants alone will not do it. At some point, government has to buy.

The demand side is now visible.

For this cut, we are focusing on direct AI tenders and awards: software, models, R&D, compute, document intelligence, chatbots, standards, training and applied AI systems. We are not treating generic TBIPS staffing vehicles, Microsoft/Copilot resale, or broad IT work that happens to mention AI as the core market.

On that basis, the data shows 81 historical AI awards, 57 federal AI tender notices, 10 currently live AI opportunities, and 26 AI grant or support programs.

That gives us three different views of the market:

The pattern is pretty clear. AI is already moving through government procurement as software, R&D, compute, chatbots, training, standards, automation, data tools, and public-sector workflow projects.

We also ran a wider analytics sweep because the Levio award proved the first search was too narrow. That broader pull found 794 analytics-flavoured awards across terms like analytics, predictive analytics, decision support, anomaly detection, data science, algorithms and artificial intelligence. The strict pass classified 106 as direct AI or applied predictive analytics, 41 as plausible but not safe to count from the record text alone, and 647 as ordinary analytics, BI, staffing, lab analytics, GIS, data subscriptions, generic IT or other false positives.

AI procurement is real, but the word "analytics" is a garbage market-sizing shortcut. It catches Palantir data analytics licensing, Equifax credit decision support, TBIPS data resources, lab testing companies with Analytics in their names, BI work, GIS tools and telematics. For this edition, the cleaner story is the smaller direct-AI layer plus a handful of large applied-predictive records, not every analytics contract with a shiny label.

The Palantir record is still useful, even though I would not count it as part of the focused AI award set. The Ontario Provincial Police previously had a $36.6 million Palantir data analytics software contract. Palantir's pullback from this market leaves a real opening: police, defence, emergency management and infrastructure buyers still need data fusion, threat anticipation, anomaly detection and operational decision-support tools. The question is who fills the space.

That is where Levio and ANVIL become more interesting than their line items alone. Levio's $10.35 million threat-anticipation award and ANVIL's AI/inference systems R&D are not replacing Palantir one-for-one, but they point to the same demand curve with a more Canadian supplier base. The opportunity is not just "AI strategy." It is the next generation of operational analytics platforms that public-safety and defence buyers will keep needing, especially if they want less dependence on controversial foreign platforms.

Across the historical records, we found 81 solution-focused AI awards worth $46.6 million. The value is modest compared with mainstream IT or infrastructure, but the buying pattern is the interesting part. Government AI procurement has already moved beyond experiments in a few places.

The 2025 spike is the one to watch. It is driven by larger applied-AI contracts rather than a big increase in small pilots.

Top known buyers by confirmed AI award value:

PSPC shows up because it is both a buyer and a procurement channel for other departments. Supply Ontario's $6 million "Artificial Intelligence Solutions" award points to provincial buying. The newer Cohere awards through Shared Services Canada and ISED are a different kind of signal: direct federal spend on Canadian foundation-model software.

The late add that changes the shape of the table is Levio's $10.35 million PSPC award for predictive analytics for threat anticipation. That is not a generic analytics project. It is a large applied-AI buy in defence/public-safety language, and it is exactly the kind of record that gets missed if the search only looks for "AI" or "machine learning."

That one award changes the read on 2025. Without it, 2025 looks like a stronger year for AI buying. With it, 2025 looks like the year applied AI started to show up as serious public-safety and defence analytics spend.

Some of the more useful award examples:

There are a few points worth pulling out.

First, AI buying has been practical. The awards are document extraction, chatbots, predictive maintenance, R&D, policy frameworks, GPU servers, transit, health hotlines, cybersecurity and municipal readiness.

Second, the federal buying pattern is more technical than the strategy language makes it sound. The tender archive includes sovereign LLM inference, generative and agentic automation, AI-enabled software development, AI security accreditation, geospatial AI, and GPU hardware. The work sits deep inside implementation.

Third, the services layer is already strong, but the more interesting split is now between direct AI solution buys and AI implementation capacity. Staffing and TBIPS-style vehicles may matter operationally, but they are a weaker market signal than awards for Cohere, ANVIL, Botpress, Diligen, MDA or other vendors where the AI product or system is the thing being bought.

One caveat on the vendor view: broad name searches can badly overstate vendor history. For example, ANVIL's real AI count in this dataset is 2 awards worth $1.4 million. A loose all-awards search for the word "Anvil" pulls in unrelated public-sector records and should not be read as AI procurement history.

The big point: Canadian firms are winning, especially in narrow AI products, R&D, LLM software, chatbots, document extraction, GPU infrastructure, digital health and defence-adjacent prototypes. The largest advisory and implementation work still favours firms with existing public-sector scale, even when those firms have Canadian partnership structures or Canadian operating arms.

That fragmentation is good for new entrants, provided they choose a lane. "We do AI" is too broad. "We help municipalities implement AI readiness and workflow automation," "we provide secure Canadian chatbot infrastructure for health service navigation," or "we run GPU systems for public-sector research" is closer to how the work is actually being bought.

The Palantir angle sharpens that point. A $36 million policing analytics platform is not a tiny pilot. When a controversial foreign platform pulls back, the demand does not disappear. Buyers still have the same operational problems. They still need to connect messy data, surface risks, explain recommendations and make analysts faster. The opening is for vendors that can bring the capability without forcing buyers into a black-box sovereignty fight.

For Levio, ANVIL and similar Canadian firms, that is the real market. Not "AI" as a branding exercise. Public-safety and defence buyers need usable, explainable, secure analytics systems. The next wave of opportunities will likely sit around threat anticipation, command-and-control decision support, anomaly detection, case triage, emergency management, infrastructure monitoring and secure AI deployment. That is where Canadian vendors have a chance to take the work that used to default to large foreign platforms.

If you sell AI software: get specific. The work already being bought is tied to use cases: document extraction, chatbots, translation, predictive maintenance, defence R&D, synthetic data and workflow automation.

If you sell services: there is already room in this market. Strategy, readiness, implementation, policy frameworks, ML engineering, data governance and training are showing up in awards.

If you sell compute or cloud: the strategy is directly relevant, but generic cloud will not be treated as AI. The stronger position is secure, sovereign, AI-ready compute with a clear story on data location, access, privacy and performance.

If you sell into health: expect the first wave to be governance-heavy. Health buyers need data spaces, privacy, integration and evaluation before they can scale patient-facing AI.

If you are a startup: the grant side matters. The Regional AI Initiatives, Compute Access Fund, Scale AI and BDC Data to AI programs are part of the route from product to public-sector credibility.

If you are already selling to government: watch the vehicles. The AI Source List is the obvious one, and more buying channels are likely to follow as the strategy moves from document to implementation.

The procurement market is ahead of the rhetoric in some places and behind it in others.

Governments have already bought AI, but the awards are uneven and often practical: chatbots, R&D, document extraction, predictive maintenance, readiness, policy frameworks, compute and automation. That matters more than a generic AI boom story because it shows where adoption actually starts.

The strategy's job is to make that buying less random. The government wants Canadian AI firms to scale, Canadian institutions to adopt, and Canadian infrastructure to carry more of the load. Grants help with the supply side. Source lists and public-sector missions help with demand. Compute sits underneath both.

The vendor lesson is straightforward: do not sell AI as a concept. Sell the use case, the channel, the proof, and the governance model.

Canada's AI strategy gives public buyers permission to start buying. The next year will show which suppliers are ready for that to become normal.

In a few years, you're going to see Publicus' name at the top of that list because we are on a mission to transform messy government data and make it useful for procurement departments and government contractors.