Canada’s highly anticipated AI for All strategy has been published for just over a month now. Documents like these often launch with huge fanfare and instantly become relics, but a few events in the wider industry have reinforced or amplified the pressure for Canada to deliver on its AI plans.
Sovereignty Gaps Create Immediate Consequences
The strategy contains an honest and optimistic view of Canada’s ambition for AI adoption. It’s clear that Canada has an enviable list of tailwinds as the world grapples with the possibilities and challenges of AI. From the vast natural resources and potential for renewable energy sources, to the AI heritage and talent that exists across the country, Canada should be bullish about its future in an increasingly AI-enabled world.
Eight days following the strategy’s launch, the US government “issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.“ This sent shockwaves through the technology ecosystem as it made what was previously a theoretical risk a real risk. While national security concerns were part of the rationale, a decision like this is always layered with political, economic, and technical choices. And it sends a strong signal that frontier models are being seen as a significant differentiator, which can alter competitive dynamics on a global scale.
Many organizations and governments end up relying on data residency as the answer to the sovereign question, however this real event shows that AI exposes how flimsy a defence that was in the first place.
Removing access to the models didn’t require any change to where data sources were however it still removed the Intelligence layer which uses data to create value.
Why AI Cost Models are also a Sovereignty Issue
Sovereignty isn’t just about who controls access, it’s also impacted by who can price you out of using it.
Over the last few months, cost models for AI tools have been changing with GitHub’s move to consumption for CoPilot as one of the highest profile. For years, the cost of inference has been heavily subsidized by continual investment in the AI ecosystem, but there are increasing signs that this is changing quickly. This hasn’t necessarily been in pure cost per token (a fixed unit price per unit of text processed). Consumption-based pricing is tied to total usage, which is harder to predict. Token consumption massively increases as teams shift from individual model conversations to agentic platform-style tools. As more work is delegated to teams of agents, each making their own model calls, costs multiply with them. This unpredictability creates a significant risk for Canadian businesses, citizens, and government. Even a small uptick in token pricing ends up getting multiplied significantly by token consumption, and so a use case which previously was affordable now becomes too expensive to run. What happens when that use case was a key part of a service?
This will really test where AI was being used to solve the right problem versus expensive frontier models being used to solve a problem that is either too trivial or doesn’t require generative AI to solve.
What Needs to Happen to Make Canada’s AI Strategy Effective?
While disruptive, these events couldn’t have come at a better time. They reinforce that Canada needs a proactive approach to control over the intelligence layer, not just the data layer.
The work is hard. And it requires tough decisions on where investment should be made, as the quick choices to get AI solutions up and running can end up being more dependencies on foreign-controlled, financially-subsidized (for now) frontier models.
This may be the right choice for many use cases, but it will increasingly end up turning into outsourcing intelligence itself. This isn’t the same as outsourcing your cloud hosting because that’s not your line of business, this is ensuring you understand how your organization makes decisions and ensures appropriate control.
The good news is that the technology is out there. Open-source tools such as LiteLLM can act as AI gateways and abstract away from individual providers. Companies like Cohere provide a Canadian frontier model which can be leveraged on your own environments to retain control over cost and execution. The key here is that choices are consciously made.
AI technology is exciting and evolving at previously unseen speeds, which only serves to amplify the importance of organizations understanding their users, understanding where they differentiate, and understanding where they can commoditize.
There might be no better example of Canada leading the way today than Alberta’s AI Academy and Velocity White Papers on how they are adopting an “AI Maximalist” approach. These resources can be freely adopted to upskill people of all backgrounds to leverage AI alongside practical examples on how ways of working are already changing. These are amazing tools to build the foundations Canadians will need to turn the AI strategy into deliverable benefits across our public and private sector.
Canada’s AI for All strategy is a great start, but it’s still just words on a page, the work is in the everyday choices on where we retain or cede control.
Kyle Thompson is the Chief Technology Officer for Kainos where he leads technology strategy, engineering excellence, and AI innovation for our clients in Canada and the United States. With extensive experience delivering large-scale modernization programs, Kyle specializes in helping organizations apply emerging technologies, platforms, data, and artificial intelligence to improve user outcomes and accelerate service transformation.
