TORONTO — Mozilla is assembling an alliance to build viable AI alternatives to those controlled by U.S. tech giants, and its participants say Canada in particular must back open source technology if it hopes to compete on the global stage.

The San Francisco-based developer of the Firefox web browser is using its technical capabilities and cash to back AI projects and firms that it feels can fill critical gaps in the AI field. Last month, Mozilla announced an R&D collaboration with Montreal’s Mila AI institute. Its venture arm has invested in early-stage funding rounds for Kitchener, Ont.-based Transformer Lab and San Francisco-based Adaption, which are taking new approaches to generative AI.

Alliance members say they share goals. “It’s really about building a version of AI that would be more accessible to everyone,” said Mila CEO Valérie Pisano, which she said is necessary to avoid people and governments being “locked into a very small set of AI providers.” 

Mozilla president Mark Surman said the firm wants to ensure there’s “an independent, truly open source stack for AI.”

Mozilla and Mila will work together in the coming months to analyze what’s freely available right now, and where there are gaps to fill. For businesses, governments and consumers to truly consider open source AI as an alternative to services like ChatGPT and Claude, they need to be just as easy to use, Surman said. “That’s our goal in the next year or two.”

Some of the technology is already out in the open. Firms like Alibaba, DeepSeek, Meta and Mistral release the “weights” that determine how their AI models function, although not the datasets on which they’re trained. Some critics argue that approach is not truly open source. By contrast, market leaders Anthropic, Google and OpenAI maintain tight control over their systems. 

Mozilla itself has built an open source tool called Any-LLM that lets developers swap between models. Still, there are gaps. Mozilla and Mila are working to fill one of them with their first project, to develop portable memory for AI. Generative tools like ChatGPT improve over time by gathering data about peoples’ preferences, the tasks they take on and the settings in which they operate. All of that context is lost when somebody swaps one chatbot, agent or model for another. 

“That’s a big cost,” according to Pisano. So Mozilla and Mila are trying to develop memory that moves with the user, which should create “greater human agency in our relationship with the technology and its capacities.”

Transformer Lab is trying to ensure users will have more models from which to choose. The firm’s open source tools make it easier for AI researchers to train and run new systems by connecting their experiments to available computing capacity. It also automatically runs the models through common and custom evaluations so they can measure and improve performance. 

“We want everybody in the world to be able to do the quality of research that only the big labs can do,” said Transformer Lab co-founder Ali Asaria. Customers include university research groups and so-called neolabs, new commercial upstarts looking to rival the tech giants for business.

That may seem like an impossible competition. OpenAI and Anthropic, for example, have raised about US$186 billion and US$61 billion to date, per PitchBook, while Google and Meta plan to spend hundreds of billions annually on AI infrastructure. Through fierce talent wars, U.S. tech giants have amassed huge teams of star researchers and engineers, and their products are already used by millions of individuals and lots of businesses. Large tech firms are also publishing less about their breakthroughs, as they look to capitalize on them.

Members of the open source alliance are small by comparison. The Mozilla Foundation had US$1.59 billion in assets at the end of 2024. Mila has over 1,500 affiliated researchers, but produced just 0.67 per cent of the top papers submitted to the 2025 edition of the Conference on Neural Information Processing Systems, according to an analysis from AI World, an offshoot of the Centre for European Policy Studies think tank. Google staff accounted for 2.82 per cent of papers, with Meta at 1.84 per cent and Anthropic at 0.2 per cent.

Still, the allies insist open source AI can offer a true alternative. “No matter how unbeatable these models look, the community as a whole is able to compete,” said Asaria, citing the thousands of individual developers and small teams working on local and collaborative versions of the technology. 

The best open-weight models lag behind the most advanced proprietary ones on key tests by about three months, according to Epoch AI, a widely-cited research non-profit. That’s a quicker catch-up than before.

Mozilla’s Surman cites precedent. Microsoft initially dominated personal computing and the web. Over time, though, Linux became the ubiquitous operating system for servers and supercomputers, if not for desktops. And Firefox took market share from Internet Explorer. A similar “switch to open source” will take place in AI, Surman predicted, arguing that it’s a more accessible, transparent and secure form of the technology. 

Open source AI could also boost adoption of the technology, Pisano said, particularly in economies that aren’t technologically advanced, because it’s cheaper and easier to use. She added that it could also make the field more multilingual and multicultural, since anyone can adapt and build on top of open source tech.

The allies also argue that there are geopolitical and strategic advantages to open source AI, as nations look to secure their digital sovereignty and avoid becoming beholden to the U.S. or China. “As country leader, can you even compete with those folks if you just try to go head-to-head?” Asaria said, calling open source “the only viable way” to create a counterbalance. 

More well-funded and technically-advanced projects could help Canada retain talented researchers, many of whom currently end up in the U.S., according to Asaria. Transformer Lab itself plans to eventually develop its own AI models, he added.

Surman said middle powers in particular should embrace open source AI. While Mozilla is headquartered in the U.S., Surman said about a fifth of the company’s 1,200 staff are based in Canada. He called for Canadian governments, pension funds and venture capitalists to invest in open source AI, and for Canadian firms big and small to build and use the technology. 

Europe is already leaning in, Surman said, citing the backing that French firm Mistral has received from government agencies and industrial giants like ASML. There’s a similar opportunity here to seize commercial and sovereign opportunities, he said. “I want Canada to be a leader in open source AI, because it will be a defining part of the economy.”