Canadian AI: Cohere's opensource voice model tops the leaderboard, and governments open the chequebook.
And more on a $16M seed extension in Vancouver, multiple defence AI deals and an "AI Scientist."
Good morning! Welcome to the Canadian AI Newsletter, a weekly rundown for founders, operators and investors.
Was this forwarded to you? Subscribe below.
I am Raif Barbaros, Partner at Mistral Venture Partners. Views are my own.
This was the week both levels of government showed the receipts. Ottawa’s Budget 2025 received royal assent ($1.75B for VC, $81.8B for defence) and Ontario dropped a $4B investment fund targeting AI and defence the very next day. Meanwhile, Cohere shipped an open-source voice model that topped leaderboards on arrival, a UBC researcher’s “AI Scientist” published in Nature, and defence AI deals kept landing in Ottawa and Calgary. The money is moving. The infrastructure is getting built. And the policy machinery, for once, is actually keeping pace. Let’s get into it.
💰 Deals & Milestones
Cohere (Toronto) launched Transcribe, its first-ever voice model. Open-source, and immediately #1 on the leaderboard.
2-billion-parameter automatic speech recognition model, released under Apache 2.0. Ranked #1 on Hugging Face’s Open ASR Leaderboard with a 5.42 word error rate, beating OpenAI Whisper, IBM Granite, ElevenLabs Scribe, and NVIDIA Parakeet.
Supports 14 languages including English and French. Processes 525 minutes of audio per minute of compute. Runs on consumer-grade GPUs.
Will be integrated into Cohere’s North enterprise agent platform. Open-source strategy continues: Tiny Aya (Issue 2) for multilingual text, now Transcribe for voice. Cohere is building in the open faster than anyone expected.
(Disclosure: Cohere is a portfolio company.)
Miraterra (Vancouver) secured a $16M CAD seed extension for its AI-powered soil intelligence platform.
Oversubscribed (originally structured at $13.9M). Led by At One Ventures with Farm Credit Canada, S2G Investments, Sitka Foundation, and iSelect. Total funding now ~$40M.
Independent Terramera subsidiary uses AI to decode satellite and LIDAR imaging for soil chemistry and biology. Recently acquired Trace Genomics’ IP and labs.
50-person team, recently began generating revenue. CEO Nate Kelly: “Three years ago, we just had a Digitizer. Today, it actually has a full working stack.”
AgTech meets AI. The kind of deep-tech, real-economy startup Canada should be building more of.
Denvr Dataworks (Calgary) signed two defence AI partnerships in a single week.
With Ottawa’s Dominion Dynamics: building Canada’s first sovereign AI simulation environment for autonomous drones (Autonomous Collaborative Platforms) supporting the Royal Canadian Air Force. Simulation ready in ~3 months, operational drone capability in 24–30 months, targeting Arctic operations and NORAD modernization.
With Ottawa’s Sapper Labs: AI-enabled intelligence and cyber defence for the Canadian Armed Forces.
Both leverage Denvr’s Canada AI Platform (CAIP), a fully sovereign AI platform under Canadian jurisdiction. Calgary company, Ottawa defence customers, Arctic mission. The geography of Canadian defence AI is taking shape.
Larus Technologies (Ottawa) awarded an $8.3M IDEaS Test Drive contract by the Department of National Defence.
AI/ML for tactical planning and decision intelligence. The challenge: “Staying Four Steps Ahead: Understanding and Predicting the Behaviour of Adversaries.”
Larus will configure its MAABI platform for automated analysis, predictive insights, pattern identification in troop movements, and automated wargaming simulations.
DND is clearly in buying mode for AI. Between Denvr’s two deals and Larus’s contract, that’s three defence AI agreements announced in a single week.
Opendoor (San Francisco/Toronto) is opening an AI builder hub in Toronto with plans for up to 100 roles.
Led by former Shopify exec and CEO Kaz Nejatian. Office at King and Spadina, hiring across operations, finance, and engineering.
No immediate plans to enter the Canadian market. This is a talent play. Nejatian called Toronto “the single greatest source of raw, high-talent people in the world.”
I agree, 1000%.
🔬 Research
UBC’s “AI Scientist” published in Nature: a system that can autonomously conduct the entire scientific research process.
Built by UBC Prof. Jeff Clune and PhD student Shengran Hu (with Vector Institute, Sakana AI, University of Oxford). The system generates hypotheses, checks literature, writes code, runs experiments, analyzes data, writes papers, and performs peer review.
An AI-generated paper passed peer review at an ICLR workshop with a score of 6.33. The automated reviewer achieves 69% balanced accuracy, surpassing human-to-human agreement.
Clune noted this represents “the dawn of a new chapter.” Key finding: paper quality improves as foundation models improve, suggesting a scaling law for AI-driven science.
Ethical precautions included watermarking and withdrawing the accepted paper. A concurrent Nature editorial addressed the implications. This is the kind of research that makes you sit with it for a minute.
Mila and Mozilla announced a strategic partnership to build open-source AI tools. This is Mozilla’s first-ever partnership with a major AI research lab.
Initial $1M CAD investment for the first research project, plus extensive engineering collaboration. Expected to be multi-year.
First project: private memory architectures for AI agents, allowing users to switch between language models without losing conversation data. Reducing dependence on closed AI systems.
Mila CEO Valérie Pisano called it a “landmark.” Mozilla President Mark Surman: “Canada can lead on AI sovereignty; we’re joining with Mila to make it happen.”
Vector Institute extended its collaboration with Roche through 2030, continuing a partnership across all three national AI institutes (Amii, Mila, Vector) that began in 2020.
Vector provides talent, AI engineering capabilities, and specialized training.
Pharma is one of the clearest enterprise AI use cases with real budget behind it. Roche recommitting through 2030 signals this isn’t a pilot.
🏛 Policy
A capital injection unlike anything we’ve seen. In the span of 48 hours, two governments put an order of magnitude more money behind Canadian AI.
Federal: Budget 2025 received royal assent. The $1B Venture and Growth Capital Catalyst Initiative was detailed: $700M funds-of-funds stream, $200M life sciences stream, $100M emerging fund managers stream. Plus $750M for early-stage funding gaps. That $1.75B is roughly 4x the previous VCCI allocation.
Ontario: The provincial budget launched the $4B “Protect Ontario Account Investment Fund” targeting AI, defence, advanced manufacturing, life sciences, and critical minerals. A private-sector asset manager will be selected to run it.
Both are explicitly pushing pension funds and private capital to match. We’re talking about a step change in capital going into the Canadian innovation economy over the next few years. It’s an extraordinary moment. Good time to be building in Canada.
FCAC released the final report from the second Financial Industry Forum on Artificial Intelligence (FIFAI 2). Four workshops, 170+ financial ecosystem members, covering consumer protection, cybersecurity, financial crime, and stability.
The report emphasizes that consumers benefit from AI only when financial well-being and consumer protection are embedded into governance frameworks, with institutions remaining accountable for AI-driven outcomes.
My take: fintech doesn’t get the attention it used to, drowned out by the AI wave. But I think AI represents an enormous opportunity for fintechs specifically. Fintechs have real moats: payment rails, regulatory licenses, domain expertise, proprietary data. These are structurally hard to replicate and took years to build. With the right team, it’s easier for an established fintech to adopt AI than for an AI-native outsider to replicate a fintech’s regulatory and infrastructure moats. The rails are the hard part. The AI is increasingly commoditized. Watch this space.
Evidence for Democracy published a review of the federal government’s AI Register, documenting over 400 instances of AI usage across 42 federal agencies.
The Register (launched November 2025) documents AI most commonly in governance/public services, industry/innovation, and immigration/borders/security. Flagged uses include CBSA facial recognition, AI for immigration decisions, RCMP’s “Draft One” AI for police reports, and Global Affairs’ AI-generated briefing notes.
Commendable transparency. And for any AI startup looking to sell into the federal government, this is a goldmine. A public map of exactly where every department is using AI, and by extension, where the gaps are.
📊 Data
MNP released “The Business of AI 2026”, a national Ipsos survey of 250 Canadian business leaders.
91% of organizations are satisfied with their AI progress, but only 4% consider their use “transformational.” 48% describe it as “operational,” meaning gen AI in production but mostly focused on individual productivity rather than deep business model integration.
Translation: nearly everyone is experimenting, almost nobody has rewired their business around it yet. That 4% → something much larger is where the next wave of enterprise AI value gets created.
In brief
CoolIT Systems follow-up: The Logic reports that CoolIT’s 650 Calgary employees will receive cash bonuses ranging from $35,000 to $490,000 when the $4.75B USD Ecolab acquisition closes. KKR will make 15x its 2023 investment. The company is on track to double headcount and quadruple revenue versus 2023, all driven by AI data centre demand. This is what the AI boom looks like when it reaches the people who actually build things.
BMO at its Investor Day: expects AI to add more than $1B in pre-tax earnings by 2030. 96% of employees are already using AI company-wide. An internal chatbot searches 8,000 policy documents, delivering $4M in annual savings and a 60% reduction in help desk calls. BMO becomes the third Big Five bank to set a $1B AI target, joining RBC (Issue 2) and TD (Issue 3). CIBC and Scotiabank, your move.
Bell and BUZZ HPC partnered on sovereign AI infrastructure at the Bell AI Fabric facility in Merritt, BC. BUZZ HPC secured 6.5 MW of immediate GPU capacity with NVIDIA GB200 NVL72 and HGX systems. Part of Bell’s six-site B.C. supercluster. HIVE Digital Technologies (BUZZ’s parent) continues its pivot from Bitcoin mining to AI compute.
Shopify publicly launched Tinker, a free mobile app consolidating 100+ AI creative tools for images, video, logos, product photography, and 360° product views. Uses models from OpenAI, Google, and Anthropic. Available to anyone, not just merchants. Shopify’s AI drip campaign continues. We’ve now covered them in six out of seven issues.
SCALE AI announced the Canadian delegation for VivaTech 2026 (June 17–20, Paris): 74 AI/tech organizations selected from 170 applications. Canada sending 100+ organizations total for a third consecutive year. Partners include Calgary Economic Development, Desjardins, Global Affairs Canada, and Invest Ontario.
📅 Upcoming Events
N³ Summit (Toronto) — March 31 – April 1, 2026
ALL IN Vancouver — April 15, 2026
The AIA (Montreal) — May 5, 2026
Upper Bound (Edmonton)— May 19–22, 2026
ALL IN Toronto — May 28, 2026
Canadian AI Conference 2026 (Vancouver) — May 25-29, 2026
ALL IN Main Event (Montreal) — September 16-17, 2026
Thank you for reading! Subscribe, and please share any feedback 🙏
— Raif Barbaros