Abundant Intelligences links six locally rooted research pods across Canada, the United States, Hawai’i and Aotearoa. Its architecture starts with community authority over knowledge, not a centralized model or data lake.
Canada’s most interesting sovereign AI experiment may not be a national data centre, but a network of locally rooted Indigenous research pods.
In 2023, Canada’s New Frontiers in Research Fund awarded C$22,830,281 to Abundant Intelligences, a six-year, Indigenous-led research program headquartered at Concordia University’s Indigenous Futures Research Centre. The project runs from March 2023 through February 2029. It is not a single Indigenous large language model. It is not a conventional university consortium organized around one principal investigator, one data repository or one model. It is an international network designed to develop AI methods, datasets, models, algorithms and infrastructure from within Indigenous Knowledge systems.
Its basic unit is the pod. The dominant AI development model begins with a general-purpose system, gathers as much data as possible and then looks for users. It begins with relationships, local protocols and community-defined purposes. The technical form follows.
A network built to remain local
Abundant Intelligences is co-directed by Jason Edward Lewis, a Kanaka Maoli and Samoan researcher at Concordia, and Hēmi Whaanga, a Māori researcher at Massey University. Its current research and impact page lists 48 co-investigators and collaborators, 13 universities or research institutes, eight community-based organizations and 158 students.
The program’s foundational, peer-reviewed paper in AI & Society describes a team connected to 16 Indigenous communities across Aotearoa New Zealand, North America, the Pacific and Africa. Its pods are locally rooted research clusters, usually anchored in Indigenous-centred research or media labs. They bring together knowledge holders, language keepers, cultural practitioners, scientists, engineers, artists and community organizations.
The pods collaborate internationally, but their work begins in place.
| Pod | Local base | Publicly described work |
| Haudenosaunee | Onkwehonwe Research Environment, Western University | AI for Haudenosaunee speech, song, dance, storytelling and creative practice, including whether computational architectures need to be reshaped |
| Hiringa Te Mahara | Massey University and Te Hiku Media, Aotearoa | Māori-centred language, culture, wellbeing and whakapapa applications, grounded in Māori methodologies, epistemologies and protocols |
| Ka Hawai’i Pae ’Āina | University of Hawai’i West O’ahu and Create(x) Lab | Environmental stewardship, language revitalization, creative practice and Indigenous data stewardship |
| Niitsitapi | Centre for Indigenous Arts Research and Technology, University of Lethbridge | Niitsitapi stories expressed through speech, song and poetry, with research into language, art and computational design |
| Wíhaŋble S’a | Wíhaŋble S’a Center at Bard College | Indigenous protocols for AI, wearable and digital technologies, developed through performance, sound, visual art and dream research |
| T’Karonto | OCAD University and York University | The Dish With One Spoon Wampum Belt Covenant as a framework for AI development, plus technical work to identify shared themes across pods |
The network gives these communities a way to share research capacity while keeping the questions, protocols and priorities locally grounded. That is a different concept of scale. In conventional AI, scale usually means centralizing more compute, more data and more users. Here, scale means connecting more locally accountable centres without assuming that all knowledge should be pooled. The approach is unified, the instances are retained by each community.
Sovereignty is more than server location
The language of sovereign AI has become common in government and business. It often means that data and compute remain inside a national border, under domestic law. Abundant Intelligences asks a more demanding set of questions: Who decides what may be collected? Who can train on it? Who can inspect or reuse it? Who receives the benefit? Who has the authority to say no?
The program has committed to developing data frameworks that address sovereignty, colonialism and cultural property, including alignment with the First Nations principles of OCAP® where applicable. OCAP® stands for ownership, control, access and possession. The First Nations Information Governance Centre is explicit that these are First Nations principles, not a generic framework to be imposed on every Indigenous people. Each nation and community has its own laws, knowledge systems and protocols.
The same logic extends to infrastructure. Abundant Intelligences says it will collaborate with communities to build AI infrastructure aligned with community needs and with sovereignty as those communities define it.
The project is well past planning o to implementation. My collaborator Laszlo Lakatos-Hayward has consulted with the community initiative. In a June 2026 interview published by the network, Kānaka Maoli researcher K. Ka’ulawena Alipio described work in the orbit of the Hawai’i pod on decentralized, self-controlled data centres. She said Keolu Fox’s lab works with Indigenous communities so data can remain on their land and the community can retain final authority over access and use. The research also poses how small, decentralized facilities could reduce pollution through solar power, circular systems and bioremediation. That is a much more complete definition of sovereignty than putting the same hyperscale cloud architecture on Canadian soil.
Where open weights fit, and where the evidence stops
Open-weight models are a logical fit for this architecture. A model whose weights can be downloaded and run locally can be adapted without sending every prompt or dataset to a corporate API. It can operate under local access rules and remain available even if a vendor changes its prices, policies or product strategy.
But open weights do not create sovereignty by themselves. Communities still need compute, technical capacity, governance, security and the legal right to use and modify the model. They also need authority over the data used for retrieval, fine-tuning and evaluation.
Abundant Intelligences’ public materials say the network is developing AI models, tools and applications as well as community-aligned infrastructure. A member of its technical pod has described hands-on prototyping, technical workshops and open-source tool development to support Indigenous-led projects.
What the network has not publicly documented is equally interesting. There is no published network-wide inventory of foundation models, model weights, hardware or inference software. There is no public architecture showing a shared open-weight stack, nor that their local data is exchanged across a common technical network. It’s for community use, so if the model is built successfully, there may be experience structured to share with other decentralized communities who are using open weight AI as a way to connect, build knowledge, best practises and intelligence. It’s a beautiful form of collaboration and one of the most interesting ways I’ve seen the technology applied yet. Abundant Intelligences is building the research, governance and infrastructure conditions for Indigenous-controlled AI. The completed federated open-weight deployment across all its communities is underway.
Operating
By the standards of a research network, Abundant Intelligences is highly active. It has six operating pods, a peer-reviewed intellectual framework, a growing international membership, student training, technical workshops, community partnerships and research extending from language technology to environmental data and neuro-AI. Its 2026 activity includes an Indigenous AI Gathering in Montreal, open-source technical work, Quechua language research and decentralized data-centre development.
The project is a little more than halfway through a six-year mandate, and several of its intended outputs are institutional: methodologies, data governance, durable relationships, technical training and community research capacity. These are slower to build and harder to measure than a chatbot launch. The network, methodology and training layers are largely complete. It is emerging for prototypes and infrastructure. Production deployment and measurable community outcomes are underway.
Those distinctions will shape the next phase of the project. The sovereignty questions now are concrete ones: Which models are running? On whose hardware? Where is data stored? What moves between pods? What remains local? Who owns the resulting code and models? Which systems are in regular community use? How will communities judge whether the work has succeeded?
An alternative architecture for AI
Abundant Intelligences matters even before those answers are complete because it challenges two assumptions at the centre of the AI industry. The first is that intelligence can be separated from culture, language, land and relationships and converted into a universal technical resource. The second is that progress requires centralization.
The pod model proposes something else: local authority with shared capacity, collaboration without compulsory aggregation, and technical development that begins with obligations rather than extraction.
Most sovereign AI plans are about where the servers and the data sit. Abundant Intelligences is building something that allows cultures to determine themselves who has the right to decide what the system knows, how it learns and whom it serves. That may turn out to be the more important architecture.
Reporting current to August 2026.
