OTTAWA — The Liberal government plans to spend $2 billion to increase the computing power available to Canadian artificial intelligence firms and researchers, and follow the U.S. and U.K. in setting up an AI safety institute to study the technology’s risks.

Prime Minister Justin Trudeau announced Sunday that this month’s federal budget will allocate the $2 billion to infrastructure for the technology, including a new AI Compute Access Fund that will pay for companies and scientists to use the processing power they need immediately. Longer term, the government is aiming to secure domestic capacity via a “sovereign compute strategy.” 

Ottawa will also spend $50 million to establish the Canadian AI Safety Institute, which will assess the negative outcomes that “advanced or nefarious” systems might produce. 

Trudeau made the announcements at the Montreal offices of the federally-backed Scale AI cluster. It’s the latest stop in a roadshow in which the Liberals have been previewing major proposals from the forthcoming federal budget, which Finance Minister Chrystia Freeland is scheduled to table on April 16. Previous announcements on the tour have focused on housing and affordability. 

The budget document is likely to include more details of how the government will implement the package of new AI measures, which have a total cost of $2.4 billion.

The moves on compute answer calls from executives in Canada’s AI ecosystem for the government to address availability and affordability of processing power, or risk falling behind other countries. 

Developers and researchers need processing power and other infrastructure to train and run machine-learning models. The largest, multi-use systems that enable generative tools like ChatGPT require vast quantities of those resources, known in the industry as “compute.” Startups and research institutions acquire specialized chips for in-house infrastructure or, more commonly, buy time from cloud service providers. Both are in high demand, and increasingly expensive.

Computer performance is measured in floating-point operations per second (flops), and supercomputers typically in petaflops. As of November 2023, Canada had 41 petaflops of compute capacity under academic or public-sector control, compared to 82 for the U.K. and almost 3,726 in the U.S., according to a study that think-tank The Dais published last month. 

Other countries have pledged significant sums to improve their AI infrastructure. The U.K. has promised £1.7 billion ($2.92 billion) in spending on new supercomputers for scientists working in AI safety, biotech and other fields; Canada’s innovation minister, François-Philippe Champagne, and his U.K. counterpart Michelle Donelan signed a compute cooperation agreement in January. In the U.S., federal agencies have launched a program to give AI researchers access to processing power and data. 

Champagne told The Logic at the U.K. AI Safety Summit in November that his office was exploring increasing domestic compute capacity and establishing an AI safety institute. At the Bletchley Park event, dozens of governments agreed to cooperate more closely on the technology’s governance.

Both the U.K. and U.S. announced plans for AI safety institutes at the summit. London has secured agreements with eight leading developers—including Google, Meta and OpenAI—to get access to their models. Last week, the two countries agreed that their institutes would jointly develop tests for those systems. 

Champagne signalled at the summit he was considering following suit, and swiftly received high-profile backing from deep learning pioneer Yoshua Bengio, who said the move could help “keep Canada in the lead in AI.”

The Liberal government is currently pushing new legislation to govern the technology. The budget will include $5.1 million to enforce the proposed Artificial Intelligence and Data Act via a new commissioner’s office, Trudeau also announced Sunday. The House of Commons industry committee is due to begin clause-by-clause review of Bill C-27 on Monday.

In addition to the compute and safety measures, Sunday’s announcement included plans to offer incentives for AI R&D and use. The budget will include $200 million for the federal regional development agencies to subsidize firms in sectors like agriculture, health care and manufacturing to adopt the technology. It will also give the National Research Council Canada’s flagship Industrial Research Assistance Program an extra $100 million to back startups developing AI products and solutions.

The proposal for new AI funding adds to the $568.8-million Pan-Canadian AI Strategy, launched in the 2017 budget and renewed in the 2021 edition, which primarily subsidized universities to recruit and retain top faculty in the field, working with Montreal’s Mila, Toronto’s Vector Institute and Edmonton’s Amii.   

The strategy included $40 million over five years starting in 2022–23 to pay for additional capacity for researchers at the three AI institutes. Ottawa also provided the non-profit Digital Research Alliance of Canada $31.5 million in funding that fiscal year. But the public compute provider is significantly oversubscribed—in 2022, it received applications for almost five times as much use of its AI-specialized graphics processing units (GPUs) as it had capacity to provide.

Provinces have also funded compute capacity. In last month’s budget, Ontario allocated $47.4 million to upgrade supercomputers for the University of Toronto and University of Waterloo, and another $18 million for research resources. Provincial agency Investissement Québec has invested in Lévis, Que.-based QScale, a data centre builder.

Canada already hosts data centre clusters from global market leaders Amazon Web Services, Microsoft Azure and Google Cloud, and all three have announced expansion plans here. In February, GPU giant Nvidia agreed to work with Canada on its compute capacity needs.