AI companies in Canada, grouped by what they actually do

Canadian AI is not one industry. Cohere, Mila, the Vector Institute and Amii all carry the label, and they answer to different owners and are paid in different ways. How each one is paid decides what it will do for a company that calls.

Derik Lawlis Founder, ThriveAI · builds AI systems inside manufacturers

Canadian AI organizations at a glance

The table lists 34 Canadian AI organizations in alphabetical order, and the order carries no ranking. Each name links to the organization's own site, and the city and description in each row come from its own pages. The Kind column uses the kinds described in the sections below, which explain what each kind will do for a manufacturer or distributor like yours.

If you run a plant, the rows most likely to matter are the applied products that work on a shop floor or a supply chain, such as Vention, Eigen Innovations, Avidbots and Kinaxis, and the consultancies that build on your own data. The institutes train people and fund projects, and Scale AI in particular funds AI projects in supply chains.

Five kinds of AI organization in Canada

Sort by who pays, and the sector comes apart cleanly. A research institute answers to a federal strategy and to its members. A model company answers to whoever buys access. A development shop answers to the client who wrote the scope. Each one gets paid for different work, so each one produces something different.

1. Public research institutes and clusters. Government and member money pays for research and for training people. The same money funds adoption projects that other companies carry out.
2. Foundation model and platform companies. Paid for access to a model, or to the platform built around it.
3. Applied and vertical companies. Paid for a product that does one job in one industry.
4. Consultancies and development shops. Paid by the hour or by the project.
5. Canadian arms and Canadian regions of foreign vendors. Paid for infrastructure or software on a foreign parent's price list.

The label AI company covers all five. A list that mixes them tells you nothing about who can help.

Public research institutes and clusters

These organizations run fundamental research and train people. They also run funding programs with matching money and eligibility rules, so a company can receive money through them without ever hiring them. What they do not do is build a system inside your plant and keep it running.

The line is not perfectly clean, and their own pages say so. Amii states that it delivers AI training and builds custom AI solutions for businesses alongside its fundamental work in reinforcement learning. Mila describes a community of more than 1,400 researchers in machine learning and also runs industry partnerships and Mila Ventures. Read the program page before assuming an institute is out of reach.

Geography follows the federal strategy rather than the market. Mila is in Montreal, and so is Scale AI, at 6795 Rue Marconi. In Toronto, the Vector Institute sits at the Schwartz Reisman Innovation Campus on College Street, and CIFAR is at the MaRS Centre on University Avenue. Amii is in Edmonton.

One federal item belongs beside them even though it is a policy rather than an organization. The Canadian Sovereign AI Compute Strategy carries $2 billion over five years from Budget 2024, starting in 2024 to 2025, across three pillars. Up to $700 million supports domestic AI compute capacity in the private sector, and up to $1 billion builds public supercomputing infrastructure. Up to $300 million sits in an AI Compute Access Fund so Canadian businesses can buy compute. No organization a buyer could call appears anywhere on that page.

AI companies in Toronto

Toronto holds the Vector Institute and CIFAR from this kind. Cohere was founded in Toronto in 2019 and lists an office there, and Waabi lists a Toronto team. Layer 6 works from MaRS as the AI centre of excellence of TD Bank Group, and Euristiq lists a Toronto address. BenchSci, Deep Genomics, Signal 1, BlueDot, Blue J and Voiceflow also give Toronto addresses on their own sites.

AI companies in Montreal

Montreal holds Mila, Scale AI and IVADO from this kind, and CRIM has an office there with its head office in Quebec City. Coveo Solutions Inc. lists its corporate address on Avenue des Canadiens-de-Montreal, and Osedea works from Rue Saint-Ambroise. Vention lists its Montreal office on Dagenais Street, and CGI has its head office on René-Lévesque Boulevard. Element AI was Montreal before ServiceNow bought it.

Foundation model and platform companies

Only a handful of Canadian companies build their own foundation models, and this is the kind most people picture when they type the question. A model company sells access to a model on its own terms, through an API or through a deployment it supports. You are buying the behaviour of a model, and nobody has yet connected it to your records.

Cohere says on its about page that it was founded in Toronto in 2019 and builds foundation models and AI solutions, including the Command models and North, an AI platform for workplace productivity that it launched in 2025. Its site lists private deployments on premises or in isolated VPCs, which is as far as that listing goes. Ask what a private deployment covers before reading it as an answer to a residency question.

Waabi belongs in the same kind with a different sort of model. It builds one model serving both autonomous trucks and robotaxis, which makes it physical AI rather than a language product. Its company page lists teams across Toronto, San Francisco, Dallas, Phoenix and Pittsburgh.

Neither one sells you a build. A model company hands over a capability, and somebody still has to connect that capability to the systems holding your parts, prices and orders.

Applied and vertical companies

A reader with a plant will find this kind the most useful. Each company sells a product that does one job, and the job is named on the label. Those jobs are planning a supply chain, searching an enterprise, detecting financial risk, reading a workforce, answering customers and moving parts on a line. The label carries a category of work rather than a category of technology.

Kinaxis sells Maestro, a supply chain orchestration platform, from Ottawa, and is one route into AI for supply chain. Coveo Solutions Inc. sells an AI relevance platform for search, personalization, generative answering and recommendations, with a corporate address on Avenue des Canadiens-de-Montreal. MindBridge sells a platform that surfaces errors, risk and fraud across financial flows, from Ottawa. Visier sells people analytics and workforce intelligence from Vancouver.

Sanctuary AI deploys physical AI for industrial automation on existing and future robotic hardware, and develops hydraulic robotic hands. That is industrial AI in the literal sense, working on hardware rather than on records. Ada is Canadian and sells an AI customer service platform whose agents automate customer conversations across messaging, voice and email.

The honest limit of the kind is easy to state. A vertical product does its one job well and does not read the rest of your systems. Suppose quoting, costing and purchasing each live in a different file. Buying a product for one of them leaves the other two where they were.

Consultancies and development shops

Consultancies outnumber every other kind, and the directories are mostly made of them. A consultancy is paid by the hour or by the project, so the output is whatever the scope said and nothing past it. That suits a buyer who already knows what to build.

Four questions are worth asking before anyone signs. Ask where the engineers sit, because that decides who can walk into your plant. Ask who owns the code once the invoices stop. Ask who fixes the system in March, after the project closed in November. Ask what the handover document contains.

One fact about the visible end of this kind belongs on the record. The Top 25 AI companies in Canada in 2026 list is published by Euristiq. Its own about page lists offices in six countries. The addresses given are 2 Vulcan St in Toronto, Twarda St 18 in Warsaw and Uhorska St 14 in Lviv. Euristiq appears first on that list.

Osedea works from 4000 Rue Saint-Ambroise in Montreal. It describes itself as an innovation firm covering product design, software, AI, robotics and data, following its acquisition of Ventriloc. TELUS Digital is much larger and sells customer experience solutions, AI and data solutions, digital transformation and platform integrity services. It was renamed from TELUS International in 2024. TELUS completed the privatization of TELUS Digital on October 31, 2025 and now owns all of it.

Canadian arms, and where the Canadian regions are

A residency question runs into this kind fast. Several Canadian AI names that still circulate on these lists now sit inside foreign or domestic parents. The infrastructure underneath any vendor may be operated from somewhere else. Knowing who owns what is the first half of the answer.

The Canadian regions are published by the providers themselves. AWS lists Canada (Central) and Canada West (Calgary). Azure lists Canada Central and Canada East. Google Cloud lists two, northamerica-northeast1 in Montreal and northamerica-northeast2 in Toronto, each with three zones. Its locations page adds that the Montreal zones sit in one or two physical data centres and are being expanded to at least three.

A region tells you where a disk is. It does not tell you what happens to a prompt after the answer comes back, and those two things are settled in different documents. Ask for both in writing, from any vendor in any of the five kinds.

Do not read a Canadian region as a statement about the model. The model that reads the data may run elsewhere, and the only control covering that is contractual. AWS gives a concrete case in its post on Amazon Bedrock in Canada. A request from the Canada (Central) Region for models such as Claude Sonnet 4.5 is routed to US Regions or to Regions worldwide, depending on the inference profile. AWS states that data at rest, including logs and knowledge bases, stays in Canada (Central), while the inference processing may happen in another Region.

Names on the usual lists that have changed hands

Four names on lists of this sort have changed owner, and each change is on the record with a date. Three of them still read as independent Canadian AI companies.

• ServiceNow announced the purchase of Montreal's Element AI and closed it on January 8, 2021 for about US$230 million. It said it would establish an AI innovation hub in Canada.
• Nasdaq completed its acquisition of Verafin on February 11, 2021. Nasdaq said at closing that Verafin would continue to operate from St. John's, Newfoundland and Labrador as a stand-alone organization with its existing leadership.
• TD Bank Group announced its acquisition of Layer 6 Inc. on January 9, 2018, describing it as an AI company based in Toronto. Layer 6's own site now says it was founded in 2011, forms the AI centre of excellence of TD Bank Group and is a trade name of The Toronto-Dominion Bank, with an address at the MaRS Discovery District, 661 University Avenue.
• TELUS completed the privatization of TELUS Digital on October 31, 2025, at US$4.50 per share and about US$539 million in total. TELUS now owns all of it.

Check the date on any list before you use it, then check who owns each name on it. The GitHub repository that circulates for this query lists roughly 200 companies by city and cites a 2017 ecosystem post as its source.

How to tell which kind you need

Four questions settle it, in this order.

1. Do you need a result inside your own operation this year, or do you need research and training? Research and training point to an institute: Mila, the Vector Institute, Amii or CIFAR. A result points to an applied product or to a firm that will build one.
2. Is the problem one job, or the fact that your systems disagree with each other? One job points to a vertical product: Kinaxis for supply chain, Coveo for search, MindBridge for financial risk, Visier for workforce, Ada for customer service. Disagreement points to a firm that builds on your own data, which is what AI for manufacturing work usually turns out to be.
3. Who has to be able to see where a number came from? When a customer, an auditor or an insurer asks, every value needs a line back to the record it came from. That is a build question rather than a licence question.
4. Where do the records have to sit, and what happens to a prompt after the answer comes back? Ask any vendor to write both down, and ask which model tier the retention statement applies to.

Answer those four before opening any directory. A directory of hundreds of listings becomes readable once you know which kind you are looking for.

Where ThriveAI sits

ThriveAI belongs to the applied and vertical kind. We build private AI systems for manufacturers and distributors. Each one is a data warehouse assembled from the systems a plant already runs, with modules for quoting, costing, scheduling, order entry, purchasing and forecasting. A chat answers from that warehouse, names the record behind each number, and a named person approves every action. We are a small firm. We are not comparable in size or in purpose to a research institute or to a foundation model company, and this page ranks nothing.