Canada’s AI Story: From Hinton’s Lab to the Sovereign Compute Push

Layla LevesqueOctober 7, 20267 Mins read0

Canada's AI Story: From Hinton's Lab to the Sovereign Compute Push

Photo: Toronto Skyline September 2014 by https://www.flickr.com/photos/christine-wagner/ via Wikimedia Commons, CC BY 2.0

On September 30, 2012, a small team from the University of Toronto entered an image-recognition contest called ImageNet. Their neural network, later known as AlexNet, beat the runner-up by more than 10 percentage points on the key error measure, according to its documented history. The team was Alex Krizhevsky, Ilya Sutskever and their supervisor, Geoffrey Hinton. Within months, Google had bought the company they formed. Within a decade, the technique they proved out, deep learning, was running almost every AI system you’ve heard of.

That story is often told as Canada’s great AI triumph, and it is one. It’s also a cautionary tale. Sutskever went on to co-found OpenAI in San Francisco. Hinton spent ten years splitting his time with Google. The breakthrough was Canadian. A lot of the money it generated wasn’t.

Canada has spent the years since trying to hold on to more of what it helped create. Here’s how the country built its AI reputation, what it has done right, where it keeps falling short, and what the latest round of federal spending is meant to fix.

How a British researcher ended up in Toronto

Hinton didn’t come to Canada for the weather. He moved to the University of Toronto in 1987 and became a fellow of the Canadian Institute for Advanced Research (CIFAR). His biography notes that he left the United States partly out of disillusionment with Reagan-era politics and his disapproval of military funding for AI research.

What he found in Canada was a research home willing to back long-term ideas. Neural networks were a minority interest in computer science for years, and Hinton, along with Yoshua Bengio at the Université de Montréal, kept working on them. That persistence paid off spectacularly. In 2018 Hinton and Bengio shared the Turing Award, computing’s highest honour, with Yann LeCun for their work on deep learning.

Hinton later added a Nobel Prize in Physics in 2024, shared with John Hopfield, for foundational work enabling machine learning with artificial neural networks. The Nobel committee listed his affiliation as the University of Toronto. By then he had left Google, in May 2023, saying he wanted to speak freely about AI’s risks.

Three institutes, three cities

In 2017 Canada became the first country to launch a national AI strategy, according to CIFAR, which administers it. The Pan-Canadian Artificial Intelligence Strategy started with $125 million in federal funding and built around three hubs, each anchored by a star researcher.

Vector Institute, Toronto

Vector launched in March 2017, co-founded by Hinton, Brendan Frey and Raquel Urtasun, with about $200 million in combined public and private money at the start, including support from companies such as Google and Shopify. It now counts more than 140 faculty members and affiliates and hundreds of students, according to published figures. Hinton remains its chief scientific advisor.

Mila, Montreal

Mila is the oldest of the three, tracing back to a lab Bengio founded in 1993. It became a joint effort of the Université de Montréal, McGill, Polytechnique Montréal and HEC Montréal in 2017, and by 2022 had roughly 1,000 students and researchers, according to its profile. Bengio has become one of the world’s leading voices on AI safety. He chaired the International AI Safety Report, published in January 2025, and in June 2025 launched LawZero, a non-profit working on safeguards for autonomous AI agents.

Amii, Edmonton

The Alberta Machine Intelligence Institute grew out of a University of Alberta machine-learning centre created in 2002 and took its current name in 2017. Its strength is reinforcement learning, the technique of training AI through trial and reward. Richard Sutton, a University of Alberta professor and Amii’s chief scientific advisor, shared the 2024 Turing Award with Andrew Barto for laying the foundations of that field, as his biography records.

The research results have been strong. CIFAR’s impact page says the Canada CIFAR AI Chairs program has drawn more than 130 top researchers, and cites rankings putting Canada fourth in the world for AI research and development.

The commercialization problem

Research excellence has never been Canada’s weak spot. Turning research into large, Canadian-owned companies is. The country’s AI history is dotted with promising firms that ended up foreign-owned.

The clearest example is Montreal’s Element AI, co-founded in 2016 with Bengio’s involvement. It raised more than US$100 million from investors and was heralded as a national champion. By late 2020 it was running short of cash, and U.S. software company ServiceNow bought it for a reported US$230 million, according to its history.

When Canada’s AI strategy task force, appointed by AI Minister Evan Solomon, gathered recommendations in late 2025, the diagnosis was almost unanimous. BetaKit’s review of the 28-member group’s roughly 348 pages of submissions found the same themes again and again: too little domestic compute, trouble keeping talent, a gap between research and commercial products, slow business adoption, and worries about sovereignty.

Brain drain, then and now

Talent leaving for the United States has been the background hum of Canadian tech for decades. A 2020 Innovation Economy Council report, covered by BetaKit, found that nearly a third of Canadian computer science graduates left the country. U.S. salaries, deeper venture capital pools and the sheer gravity of Silicon Valley do the rest.

There are signs the current is shifting a little. In August 2026, Ottawa announced the first 64 recruits under a $1.7 billion research-talent initiative, with 48 of them coming from U.S. institutions such as Harvard, MIT and Cornell, BetaKit reported. AI was one of the targeted fields. Our view: recruiting researchers is the easier half. Keeping their startups headquartered here once they need a $500 million funding round is the harder one.

The compute gap

Modern AI runs on enormous clusters of specialized chips, mostly from Nvidia, and Canada has had far less of that capacity than its research reputation would suggest. An RBC report from March 2025, Bridging the Imagination Gap, said Canada had roughly one-eighth to one-tenth the compute performance per capita of G7 peers like the U.S., and trailed every other G7 nation in AI computing infrastructure.

Without that hardware, Canadian researchers and startups rent capacity from American cloud providers or move to where the chips are. That’s both a cost problem and, increasingly, a sovereignty one.

Ottawa’s response: the Sovereign AI Compute Strategy

In April 2024 the federal government announced a $2.4 billion AI package in that year’s budget. The bulk, $2 billion, went to computing capacity, which became the Canadian Sovereign AI Compute Strategy. The rest included $200 million for regional AI startups and adoption, $100 million for helping small and medium-sized businesses adopt AI, $50 million for workers in disrupted sectors, and $50 million for a new Canadian AI Safety Institute.

The compute money has since been flowing through several channels:

Program	What it does	Key milestone
AI Compute Challenge	Up to $700 million to help build or expand commercial AI data centres	Cohere awarded up to $240 million toward a $725 million project, December 2024
AI Compute Access Fund	$300 million to subsidize compute costs for smaller firms	$66 million to 44 businesses announced May 2026
Sovereign public supercomputer	Up to $890 million over seven years for a national AI system	Applications opened April 2026

The Cohere grant was the strategy’s first investment, BetaKit reported, intended to let the company train its next models in Canada. The Access Fund covers 67 cents on the dollar for companies using Canadian compute and 50 cents for foreign compute, a deliberate nudge toward domestic providers, according to BetaKit’s coverage. The public supercomputer program requires applicants to be Canadian non-profits or post-secondary institutions, or consortiums led by them.

Budget 2025, the first of Prime Minister Mark Carney’s government, kept going. It set aside $925.6 million over five years for large-scale sovereign public AI infrastructure, $800 million of it drawn from the existing compute strategy, plus $25 million for Statistics Canada to measure AI and technology use, as BetaKit summarized.

Private players are building too. Telus opened what it calls a Sovereign AI Factory in Rimouski, Quebec, in September 2025, powered almost entirely by renewable energy, per BetaKit.

Cohere: Canada’s flagship, with a German twist

If Canada has an AI champion, it’s Cohere. The Toronto company was founded in 2019 by Aidan Gomez, Ivan Zhang and Nick Frosst. Gomez was a co-author of “Attention Is All You Need,” the 2017 Google paper that introduced the transformer architecture behind today’s chatbots. Rather than chase consumers, Cohere sells AI models and agent tools to enterprises and governments that want to keep their data under their own control. Its annual recurring revenue passed a reported US$240 million during 2025, according to its company profile.

In April 2026 Cohere agreed to acquire German AI firm Aleph Alpha, with Germany’s Schwarz Group pledging €500 million to lead its next funding round. Reports put the combined company’s value around US$20 billion. Gomez will lead it from a global headquarters in Toronto. Minister Solomon called it “a big moment for Canadian AI,” BetaKit reported.

That’s an interesting reversal of the Element AI story. Instead of a Canadian firm being absorbed by a foreign buyer, a Canadian firm is doing the absorbing, betting that governments and companies on both sides of the Atlantic want an alternative to U.S. AI giants.

What it adds up to

Canada’s AI story has a familiar shape: world-class science, patient public funding and an outsized share of the field’s founding ideas, followed by a long struggle to keep the talent, companies and economic gains at home. The current push is the most serious attempt yet to fix the last part. Billions for compute, a dedicated minister and a homegrown company with real scale are all new ingredients.

Whether it works will depend less on announcements than on execution: whether data centres actually get built, whether Canadian startups can get the chips they need, and whether the next Hinton-trained founder decides Toronto, Montreal or Edmonton is a good place to build a big company. The ingredients are better than they’ve ever been. The test is the next five years.