OTTAWA — Firms making chips and other computing hardware in Canada are urging the federal government to include them in its $2-billion plan to build out the country’s artificial intelligence infrastructure, instead of splurging on U.S. tech giants.

“We don’t want all of that money to go to foreign companies, whether it’s Nvidia, Google or Microsoft,” said Paul Slaby, managing director of Canada’s Semiconductor Council (CSC), a lobby group. Far better, he argues, to buy Canadian.

In response to frustrations from research organizations and tech executives about access to the processing power to train and run AI models, the Liberal government promised in April to spend $2 billion on “compute.” Ottawa is currently deciding how to use that money—and there are plenty of opinions about where it should go.

Many AI startups want federal subsidies to lower the cost of their existing compute providers, which they are keen to keep using. But hardware executives say Ottawa shouldn’t just write cheques to cloud giants like Amazon, Google and Microsoft, or to Nvidia, whose chips fill data centres. 

Instead, some are calling for the new federal programs to require that compute providers buy technology from homegrown firms. “You should have Canadian content in AI hardware infrastructure,” said Hamid Arabzadeh, CEO of Ranovus, an Ottawa-based firm that makes light-based photonics parts to connect the chips inside data-centre servers.

Typically, startups and researchers get compute by buying it from cloud service providers, or by setting up their own clusters of servers. Right now, most of that infrastructure exists outside Canada’s borders. 

The CSC is proposing an approach for the federal compute programs modelled on the offset system when the military purchases equipment. For example, a defence contractor selling fighter jets may have to buy parts or services from Canadian suppliers, or to help fund their R&D work. 

In the case of the AI compute program, a cloud firm that received federal funding to build new data centres could be required to use chips or other hardware from firms in Canada. “It is an opportunity for semiconductor companies to develop and sell their products,” said Slaby. 

Arabzadeh said a Canadian content requirement could also help homegrown firms land other clients and partners. For example, Nvidia typically sells its clients stacks of servers containing dozens of its in-demand chips, as well as other components. Ottawa could buy or subsidize machines from the firm on the condition it used photonics from Ranovus or switches from Evertz, a Burlington, Ont. firm, Arabzadeh suggested.

“You’re looking for those small advantages to get into supply chains,” he said. “Government could provide that.” Once a server-maker has tested and integrated Canadian components into its design, it may choose to keep them as part of its offering in other countries. (Ranovus is already in advanced discussions with Nvidia, according to Arabzadeh.)

The CSC has also recommended that Ottawa give some of the $2 billion in compute funding to a proposed consortium of government departments, chip firms and researchers. The group would jointly develop a new type of data centre that is energy-efficient and can be used to both train and run AI models. Any Canadian content requirement should include foreign firms that have large R&D operations and workforces here, said Slaby.

The new programs should “support the buildout of capacity in Canada generated by Canadian companies,” argued Chris Walker, CEO of Untether AI. The Toronto-headquartered startup sells chips designed to run AI models that have already been trained for tasks like machine vision, and claims its systems perform better than Nvidia products while using less power. 

Walker said the federal compute programs should fund companies and researchers that are developing and adopting AI tools, but offer them more money to use Canadian-made infrastructure. For example, government subsidies could encourage firms in Canada’s sizable agriculture and aerospace sectors to adopt machine-vision technology, which can help sprayers use less pesticide and make planes better at avoiding collisions. That in turn could create demand for hardware from firms like Untether, Walker said “That is a virtuous system.” 

Firms developing hardware in Canada are already involved in large AI infrastructure projects in other countries. In India, Untether is working with Ola Krutrim, a Bangalore-based firm with plans to significantly expand its data centre capacity for AI models that are conversant in several subcontinental languages. 

Cerebras and Groq—two Silicon Valley chip firms with growing teams in Toronto—are also pursuing sovereign compute opportunities, both inside and outside Canada. Each has signed major contracts in the Middle East. Both firms  told The Logic this summer that they’d like to be part of Canada’s compute strategy.