On January 20, 2026, Canadian Prime Minister Mark Carney delivered his Davos manifesto at the World Economic Forum annual meeting. It represented a historic rally cry for middle powers to unite and chart what he termed a "third path" between building higher walls and accepting subordination to hegemons. “Collective investments in resilience are cheaper than everyone building their own fortress,” he argued. “Shared standards reduce fragmentation.”
Coincidentally, that same day also marked one year since DeepSeek released R1, the model that catalyzed an explosion in open-source AI. These two moments are connected by more than just mere coincidence. In fact, the collaborative infrastructure Carney called for already exists and it's called open source.
Canadian politician and Minister of AI Evan Solomon has likened AI’s trajectory to the internet’s ascent. “Every technology has similar characteristics,” he told the Financial Post. “There's a tremendous amount of enthusiasm (and) anxiety…great opportunities to capitalize on [AI] and negative unanticipated consequences.” He pragmatically acknowledged that “we need to master this and do it on our own terms."
This same logic explains why Canada should be investing in open-source AI. The internet would not exist without US government investment in open innovation. The Defense Advanced Research Projects Agency’s (DARPA’s) open research model made the internet open by design, publishing protocols, prioritizing interoperability and requiring that research be shared. TCP/IP, HTTP and HTML were all open standards — the sets of shared rules and conventions that made the internet globally interoperable. That openness enabled the incredible levels of global innovation and commercialization that followed.
What if Canada could play a role in ensuring AI advances have the same transformative effect? What kind of policy choices and governance would we need?
The Sovereignty Paradox
Solomon has argued that sovereignty “doesn't mean Canada would be completely siloed.” Mark Surman, executive director of the Mozilla Foundation, agrees and offers a framework for what sovereignty-through-openness requires. In an interview, Surman mapped out four dimensions of control: physical infrastructure, for example, data centres where systems run; legal directives, referring to licensing terms and who controls those; administrative controls, meaning software updates and remote access; and adaptive measures, including the freedom to modify and extend. “Anything can be run on Canadian servers, but it doesn't mean you don't have a foreign company leveraging remote controls,” Surman said. Proprietary software hosted in Canada may address the physical requirement but fails on the other three dimensions of control. Open source, by contrast, addresses all four of these dimensions.
Sovereignty in AI, however, faces a deeper paradox, one that makes global collaboration essential, not optional. Richard Gold, chief policy and partnerships officer of Conscience, an open science drug discovery organization, puts it bluntly: “If it's not open, you've lost. The reason is not just the pace, but the need for large datasets, which by necessity have to be open.”
The 2024 Nobel Prize in Chemistry for AlphaFold, the AI system that predicts protein structures, has accelerated drug discovery worldwide. It was made possible because of an open database called the Protein Data Bank, a global open-access molecular database and the result of a decades-long open science collaboration. No single country could have assembled a data set like that on its own.
This is the sovereignty paradox: genuine autonomy in AI requires participation in global open ecosystems. The breakthroughs that will matter, from health care and climate change to quantum and scientific discovery, will be built on data and collaboration that transcend borders.
Even big tech agrees. As Kevin Chan, head of public policy at Meta Canada, noted, the Treasury Board’s own analysis found that “digital sovereignty in a literal sense, meaning we would make everything ourselves, is not feasible nor practical. In fact, open source means more independence: you can download the models, use secure and private information to refine the models, and run inference with that data locally.”
Sovereignty Through Collaboration
“We were first in AI,” Solomon has said. “The fact that the world has rushed to catch up doesn't mean we've lost our way.”
The good news is we have already quietly built more open-source infrastructure than we recognize. Canada has Cohere, as well as an important branch of the Mozilla Foundation. Transformer Labs, Ollama and many others are actively releasing open-source models and tooling. Even our research institutes seem to be stepping up: Mila, in Quebec, is “starting to develop that capability, opening their mind” to open-source approaches, Surman notes. Recently, the University of Waterloo announced Open Quantum Design, a nonprofit offering the world’s first open-source, full-stack quantum computer, describing itself as “a distinct third path alongside academic labs and startups, prioritizing shared progress and open access.” So in many ways, Canada is already building Carney’s third path.
And yet, examples of the contrary also exist. Although invented in Canada, Ethereum moved its community to Switzerland in large part because of Canada’s restrictive regulatory environment. We cannot repeat the same mistakes with AI. The talent and ideas here are highly collaborative, an essential Canadian trait. However, it remains to be seen whether the approach to growing our AI ecosystem will match the ambitious approach of Carney’s third path.
In Canada, Richard Gold's work at Conscience demonstrates how open approaches solve problems that closed models cannot. His organization actively supports open science and runs drug discovery competitions where proprietary algorithms compete, with the difference that winning molecules enter the public domain. He says that “by being upfront and saying ‘this is not proprietary,’ we don’t have to worry about IP. It makes it less expensive, faster and more productive.”
The approach addresses market failures, including the sad fact that, as Gold states, “95 percent of rare diseases don't even have a simple treatment. Ten million people die a year of anti-microbial resistance.” This is a problem that no single country or company will solve alone.
The same logic applies across other sectors as well. The RAISE program — launched by Human Feedback Foundation with The Dais at Toronto Metropolitan University and Creative Destruction Lab — is piloting responsible AI adoption for Canada’s nonprofit sector. Our country’s 170,000-plus nonprofits contribute CAD $192 billion to the economy, yet only 4.8 percent use AI. The RAISE AI Adoption Playbook is being released under an open-source licence, and learnings from the program are designed as shared infrastructure that any organization can adapt and extend.
Canada has a real opportunity to be a strong middle power in AI and carve a path different from the American or Chinese path.
Open Source Is Winning in Canada
Harvard and the Linux Foundation estimate that open-source software represents USD $8.8 trillion in value globally. It is no surprise that the United States, China and the European Union have all already outlined open-source AI as a strategic priority. The EU's Apply AI Strategy, launched in October 2025, explicitly promotes open source and a "buy European" approach for the public sector. France and Germany announced a partnership with Mistral AI, committing to support open-source solutions. Chinese organizations moved from marginal contributors to a dominant force, with Baidu releasing over 100 models in 2025. In the United States, Meta has led the way and changed history by releasing Llama.
Just as DARPA’s model created the internet's open standards, today's AI infrastructure is being built through global coalitions prioritizing interoperability.
The Agentic AI Foundation governs protocols such as the Model Context Protocol (MCP) with AWS, Anthropic, Google, Microsoft and OpenAI collaborating on shared standards. The AI Alliance, co-founded by IBM and Meta with over 50 members, including the University of Toronto, develops safety benchmarks and trust tools. Domestically, Shopify and Google just launched UCP, an open-source protocol to power agentic commerce. Now is a great time for governments to step up.
The G7, Surman suggests, "could be pooling resources" into truly open large language models. Canada "has a real opportunity to be a strong middle power in AI and carve a path different from the American or Chinese path, but the only way to get there is to leverage open source extensively."
We’re still in the early days of countries rolling out comprehensive national AI strategies, yet the United Arab Emirates offers proof of where the investment will go: its $300 million Falcon Foundation has produced world-class open models and "won friends in Global South countries."
Surman offers concrete priorities: "If we want sovereign assets like research labs, we could put that research funding into projects that leverage open source and build open source." He identifies developer tooling as the critical gap: "The biggest gap is developer tooling to make it easy to take open source and use it." A stronger open-source pillar in Canada's AI strategy would deliver on multiple fronts: it could accelerate AI adoption, lower costs, advance sovereignty, and build a foundation for public trust needed for AI to credibly serve the public interest.
Middle powers can build influence not through proprietary control, but through shared access, incentives and contributions to open science, data commons, protocol layers, maintenance and evolution of the open-source infrastructure.
What Leaning In Looks Like
If Canada wants genuine AI sovereignty, we do not need to “own the rails,” nor do we need to chase the “made in Canada” approach blindly, without thinking about interoperability and global competitiveness. We can lead and join global coalitions, grow domestic capacity and talent, and strategically invest in open source.
“The powerful have their power,” Carney said at Davos. “But we have something too: the capacity to stop pretending, to name realities, to build our strength at home, and to act together.” Open source is a path for how middle powers can achieve genuine sovereignty: through collective governance and collaborative shaping of emerging ecosystems, alternatives and protocols.
Canadian organizations are already building the third path in quantum computing, drug discovery, financial services and AI adoption. The infrastructure exists, and from this rapture, as Carney put it, we can build something better. Canada just needs to walk the path.
