Mark Surman has a warning. The president of the Mozilla Foundation sees AI sliding toward control by a handful of powerful players. Yet he spots an opening. Nations that sit between superpowers are stepping up. Canada. The EU. The UK. They push a different model. One rooted in openness. One that gives developers and governments real choices.
Surman laid out the case in a piece for Fortune. Open source emerges as the consensus middle path. It promises transparency. Accountability. A chance for AI to reflect more voices than those from Silicon Valley or Beijing. The choices made now matter. They decide whether AI stays locked inside private systems or becomes a resource shaped by broader interests.
Canada moved first. Prime Minister Mark Carney used his stage at Davos this year to challenge fellow middle powers. Build a world that honors human rights, sustainable development, and sovereignty, he said. The country’s updated national AI strategy puts open-source technology at its heart. No longer content to watch from the sidelines, Ottawa bets on collaborative models that avoid full dependence on foreign closed systems.
The European Union followed suit. Its latest technology sovereignty package centers open source across the AI stack. Officials promise funding for startups. They rewrite procurement rules to tilt toward open alternatives. Germany and Japan study similar steps. Britain went further with its Open Source Builder’s Fund. The goal? Turn the UK into the global hub for open-source AI talent.
But words alone don’t build infrastructure. Surman points to four concrete dimensions that determine real control. Physical infrastructure — the data centers. Legal directives — who sets the licenses. Administrative controls — updates and remote access. And adaptive capacity — the freedom to modify and extend systems. Run a proprietary model on Canadian servers and you may check the first box. The other three? Often still owned elsewhere. “Anything can be run on Canadian servers, but it doesn’t mean you don’t have a foreign company leveraging remote controls,” Surman told the Centre for International Governance Innovation.
Canada already holds pieces of the puzzle. Cohere develops strong language models with open elements. Mila in Montreal pushes research that increasingly opens up. Transformer Labs and Ollama release tools and models freely. The Mozilla branch there adds decades of open-internet experience. These assets give the country a foundation few middle powers match.
Mozilla’s Counteroffensive Takes Shape
Mozilla doesn’t wait for governments. The organization released its open-source AI strategy early this year. The title says it all: “Owners, not renters.” Closed models turn users into tenants of intelligence they cannot inspect, cannot fully shape, cannot truly own. That path repeats the mistakes of earlier centralized platforms. Mozilla aims to flip the script.
Its 2026 plans focus on four pressure points. First, make open AI simpler than closed alternatives. The Mozilla.ai team builds a framework called any-s. It stitches together model routing, evaluation, guardrails, memory, and orchestration. Developers should need nothing more than a single call to get started. No infrastructure PhD required.
Second, fix the data problem. The Mozilla Data Collective creates a marketplace for licensed, high-quality datasets with clear provenance. Users gain agency over what they share. Third, accelerate open model development and evaluation. Fourth, push for more open compute options so the stack doesn’t collapse at the hardware layer.
Partnerships multiply the effort. In March Mozilla committed an initial $1 million to work with Mila. The project targets portable memory for AI agents. Users could take their conversation history from one model and move it to another without loss. Valérie Pisano, Mila’s president and CEO, captured the stakes. “We think this is something that could open up a different way of building and thinking about AI.” Surman added the broader vision: “We are working to build a future where AI development is rooted in openness, privacy, and humanity.” The BetaKit report on the deal shows how research meets real product needs.
Market signals encourage them. An a16z and OpenRouter analysis found open-source models captured roughly 30 percent of token volume by mid-2025. That number started near 1 or 2 percent late the previous year. Demand-side estimates for open AI reach $8.8 trillion. The private sector smells opportunity. Builders want alternatives that reflect local values and reduce vendor lock-in.
Yet gaps remain large. Developer tooling still lags. Many layers of the stack need filling before open options match the polish of frontier labs. Surman acknowledges the long game. By 2028 he wants Mozilla funding a mature enough ecosystem that open becomes the default choice for many developers and enterprises. That requires sustained investment, clear standards, and government buyers willing to choose open when it meets their needs.
The U.S. picture adds urgency. When Anthropic suspended access to its Mythos preview earlier this year, global observers took note. Dependence on single providers creates sudden vulnerabilities. Middle powers watched and accelerated their own plans. They see open source as insurance against both commercial whims and geopolitical pressure.
Surman frames it as a rebel alliance. Not a formal treaty. More a loose but determined network of governments, nonprofits, researchers, and companies. They share code. They pool talent. They align on principles that put human agency first. Canada and the EU supply policy momentum. Mozilla supplies technical direction and community trust. Startups like Transformer Lab and Oumi, both backed by Mozilla money, supply fresh tools.
Success is not guaranteed. Closed models still lead on raw capability in many benchmarks. Compute remains expensive and concentrated. Talent flows toward the highest salaries, often in closed labs. But the momentum builds. Recent X discussions echo the Fortune article, highlighting Mozilla’s coalition as a genuine challenge to established powers.
The CIGI analysis puts it sharply. Sovereignty through openness beats solitude or subordination. Middle powers cannot match U.S. or Chinese scale alone. They can, however, shape global standards, contribute to shared data commons, and maintain the ability to fork and adapt when needed. That flexibility beats any fortress built in isolation.
So the pieces move. National strategies align. Technical foundations expand. Funding flows to open projects. Developers experiment with the growing stack. The question is no longer whether an alternative exists. It is how quickly the alternative can mature into a serious competitor that governments, companies, and citizens actually prefer.
Surman believes the window remains open. But not for long. The next two years will decide whether open systems reach the coherence and performance needed to win on merit. If they do, AI could evolve into something governed more broadly. Something accountable to more than quarterly earnings. Something closer to the original promise of the open internet that Mozilla helped defend for decades.
The middle powers have declared their direction. The coalition assembles. The code writes itself in public. Now comes the harder part — execution at speed and scale.
WebProNews is an iEntry Publication