Mila and Mozilla are pushing their open source AI partnership into a much bigger project.
The two organizations have announced a new initiative to build an open source AI foundation layer that businesses, governments and institutions can operate on their own infrastructure. The idea is straightforward: organizations should be able to use advanced AI without handing complete control of their data and technology stack to a proprietary provider.
The project was announced on September 17 at ALL IN, Canada’s AI and technology event in Montréal. Mila will lead technical delivery and coordination, while Mozilla will contribute technical expertise and an initial $5 million investment. Canadian technology company Hypertec is adding another $1 million in first-year funding to support early deployments on its hardware. The Government of Canada is also backing the initiative.
This is less about releasing another AI model and more about fixing what happens after an open model is downloaded.
Mila and Mozilla Want Organizations to Own Their AI
Open models are widely available, but running them reliably inside an actual organization remains far more complicated. Businesses need infrastructure, security controls, integrations, governance, maintenance and people who understand how those pieces work together. Mila and Mozilla want their new foundation layer to remove some of that friction by providing a ready-to-use system that organizations can install on infrastructure they control rather than building an entire AI stack from scratch. Local deployment is a major part of the idea, allowing organizations to potentially use AI with internal documents, procedures, code and proprietary information while keeping sensitive data within their own environment.
A manufacturer, for example, could use a private AI assistant to search equipment manuals, answer technical questions, draft operational reports or support developers working on internal software. Instead of automatically sending that information to an external cloud AI service, the organization could keep more of the processing within infrastructure it manages directly.
The Project Is Building an Open AI Foundation Layer
At the center of the initiative is an architecture designed to remain open and interchangeable rather than locking organizations into one model or technology provider. Mila and Mozilla plan to develop an open standard based on interface contracts, allowing individual components of the AI stack to be replaced as better technologies emerge. Alongside that standard, the organizations intend to create a working reference implementation that companies and institutions can install on their own infrastructure, connect to their own data and configure around the models they prefer.
Governance and access controls are also expected to form part of the system from the beginning. That modular approach could prove particularly useful in an AI market where models, inference systems and supporting technologies change quickly. Today’s preferred model or database may not remain the obvious choice a year from now, so an architecture built around interchangeable components could give organizations more room to adapt without rebuilding the entire system.
Open Models Are Popular, but Production Remains Difficult
Mozilla’s research highlights a gap between experimenting with open AI and actually putting it into everyday use. According to its State of Open Source AI research cited in the announcement, 79% of developers adding AI capabilities use open models, yet only 53% of teams reach production. Cost, security, integration and ongoing maintenance remain among the biggest obstacles, showing that access to an open model does not automatically translate into a usable enterprise AI system.
That is the problem Mila and Mozilla are trying to address. The model itself is increasingly only one piece of the puzzle. Organizations still need authentication, permissions, retrieval systems, monitoring, infrastructure and security around it. Large technology companies may have teams dedicated to those tasks, but smaller businesses, public institutions and nonprofits often do not. A standardized open foundation could reduce some of that technical burden.
Mila and Mozilla Have Already Spent Months on the Architecture
The initiative isn’t starting from a blank page. Mila and Mozilla said they have spent the past six months designing the architecture, selecting open source components for different layers and testing whether the overall stack works from end to end. Hypertec will also help move that work toward practical deployment among Canadian organizations by supporting early implementations on its hardware.
The new initiative builds on a strategic partnership Mila and Mozilla announced in March 2026. That collaboration centered on advancing open source and sovereign AI, with research covering areas such as trust, usability and private memory architectures for AI agents. Mozilla described the agreement at the time as its first strategic research partnership with a major AI research lab. The latest project moves that relationship beyond research and toward infrastructure that organizations could actually deploy.
First Reference Implementations Are Expected Within Six Months
Mila and Mozilla are setting an ambitious development schedule, with working reference implementations for enterprise, government and public-interest use cases expected within six months. The broader initiative will stretch across two years and aims to make open source AI considerably easier for organizations to deploy without requiring them to assemble every technical layer themselves.
Mila brings a research community of nearly 2,000 researchers and professionals to the effort, while Mozilla contributes decades of experience developing open technologies and supporting open source ecosystems. The organizations are also inviting companies, developers, governments, research institutions and funders to participate. That wider involvement could become an important part of the project because an open standard becomes more useful when hardware companies, software developers and AI providers begin building compatible tools around it.
The Bigger Question Is AI Control
The project also taps into a broader debate around who controls the infrastructure behind artificial intelligence. Many organizations currently access advanced AI through platforms operated by a relatively small number of technology companies. That approach offers convenience, but it can also create dependencies around pricing, data handling, model availability and platform rules. A practical open source stack gives organizations another option by allowing them to decide where their AI systems run, which models they use and where internal information is stored.
Mozilla has been pushing this idea for years through its focus on transparent and controllable technology. With the Mila partnership, that philosophy is becoming more practical. Rather than treating open AI simply as downloadable models, the organizations are focusing on the infrastructure required to turn those models into systems that companies and institutions can realistically operate themselves.
Canada Is Positioning Open Source AI as Strategic Infrastructure
Canada’s role in the initiative also carries strategic significance. The country has invested heavily in artificial intelligence research, and this project frames open source AI infrastructure as part of its broader ability to develop and operate advanced technology with greater independence. Instead of focusing only on building larger models, the initiative places attention on the tools and standards organizations need to deploy AI under their own control.
Mila President and CEO Valérie Pisano said the project is intended to give businesses, nonprofits and governments the ability to run, control and maintain AI models themselves. Mozilla President Mark Surman also positioned the initiative around creating an AI ecosystem that organizations can build upon instead of relying entirely on infrastructure controlled by a handful of providers.
There is still plenty to prove. Reference implementations must turn into reliable real-world deployments, organizations will still need technical support, and open source infrastructure does not eliminate the costs of compute, maintenance or cybersecurity.
Still, the initiative targets a real weakness in today’s open AI ecosystem. Downloading an open model has become relatively easy. Building everything around that model — securely, reliably and at scale — remains much harder.
Mila and Mozilla want to close that gap.
Sources
CNW / Newswire: “Mila and Mozilla Announce New Initiative To Build Trustworthy Open Source AI For Everyone”
Read the original announcement
Mozilla: “Mozilla and Mila announce strategic research partnership to advance open source and sovereign AI capabilities”
Read Mozilla’s announcement
