Major cloud computing providers are seeing demand in Canada soar on rocketing interest in artificial intelligence, as startups launch new applications and large firms race to adopt the technology.

Amazon, Microsoft and Google collectively command a near-supermajority of the worldwide market for cloud infrastructure services, a major source of revenue growth for each firm. The hyperscalers, an industry term reflecting their size, are now tapping into and fuelling the AI boom. “The level of demand and interest has been unlike anything I’ve seen,” said Sam Sebastian, a longtime Google executive who now heads the Canadian branch of the firm’s cloud unit. AI is a major topic “in every C-suite conversation I have.”

The hyperscalers’ core AI offering is compute, an industry term for the processing power and other infrastructure necessary to run clients’ workloads. Between them, they’ve spent billions on data centres in Canada, with plans for hundreds of millions more in the next few years to expand capacity as demand for cloud and AI services ramps up.

The investments in compute are only part of the role the hyperscalers are seeking in an AI-powered future. The three firms also provide ready-made machine-learning applications that customers can deploy internally or build into their products. The hyperscalers all sell their own foundational models, the kind of multi-purpose systems that underpin generative tools like OpenAI’s ChatGPT. But each hyperscaler also hosts a marketplace with the offerings of startups like Anthropic and Toronto-based Cohere, as well as open-source ones from Meta and Mistral.

Cloud executives say larger firms in fields like health care, logistics and financial services, as well as telecommunications and technology, are using pre-trained models and the hyperscaler’s tools to build in-house assistants, improve search functionality or launch customer-service bots. “What ChatGPT did was put [AI] into the hands of non-technologist businesspeople who could start to appreciate the power of some of these capabilities,” said Eric Gales, managing director for Canada at Amazon Web Services (AWS). 

Corporate Canada has been notoriously slow to adopt new digital technologies. Economists and AI startup founders worry about similar reticence with AI. But demand in Canada for the AI services of Microsoft’s Azure cloud unit has so far kept pace with that in the U.S., said Henrik Gütle, product go-to-market leader for the Americas. “I’ve been surprised to see large, established firms really jumping in and experimenting.” 

While many clients are buying pre-packaged solutions, AI startups are paying for the hyperscalers’ infrastructure but building their own products from scratch. There’s “huge demand—it’s almost like the dot-com era all over again,” said Gütle.

Developers need a lot of processing power to train and deliver machine-learning models. “This is the dawn of a new era of compute in this generative AI space,” said Gales. Hyperscalers are well positioned to participate because of their networks of data centres, access to vital chips—Google and Amazon now make their own—and existing business. Some startups choose or add cloud providers because their customers already use them, Gütle noted. (The AI boom is also creating openings for specialized compute vendors like CoreWeave, backed by Investment Management Corporation of Ontario.)

Many AI firms have links to multiple hyperscalers. Take Cohere. Under a November 2021 deal, the firm uses Google Cloud’s infrastructure, including tensor processing units, which are specialized for training and running AI. Cohere’s models are in turn available in the Google Cloud marketplace, but also those of AWS and Azure. 

“The cost of compute is not trivial,” said Francesco Iorio, CEO of Toronto-based Augmenta, whose tools produce code-compliant construction designs. Prices for time on machine learning-specialized infrastructure have risen in recent months amid all the demand for generative AI, he noted. Augmenta is relatively insulated, because it doesn’t need to train an enormous foundation model and most of its technology can run on traditional servers. 

But some AI startups are spending up to four-fifths of the capital they raise on compute, partners at financier Andreessen Horowitz estimated in April. Tech giants have linked the two in their investment strategies, with Microsoft’s US$11-billion-plus investment in OpenAI and Amazon’s US$4-billion commitment to Anthropic tied to use of their cloud services. Amazon and Microsoft declined to say whether they would make similar deals in Canada. At Google, Sebastian said he internally flags opportunities for the tech giant to invest in interesting AI startups that are starting to take off. 

While many Canadian clients use the hyperscalers’ AI compute infrastructure around the world, the firms have built infrastructure here too. Each of the three providers now has two cloud regions—clusters of multiple data centres—in the country, and all are looking to expand. Some customers want their information hosted domestically, or near enough to ensure applications run fast. 

AWS has a $24.8-billion Canadian infrastructure plan through 2037, including the $2.57 billion it’s spent to date. It launched a new Calgary region last month, adding to one in Montreal. That makes it the first hyperscaler in the country’s West, Gales said, forecasting that proximity will allow “a whole new variety of use cases, including in this domain of generative AI.”

In November, Microsoft committed US$500 million to more than triple cloud capacity in its Quebec City-based region over three years. (Azure’s other Canadian location is in Toronto.) And Google has plans for a $735-million data centre in Montreal, adding to its existing region in the city and one in Toronto.

Some Canadian tech executives have expressed concern that U.S. cloud giants will make an outsized share of the profits from the AI boom, as startups spend the capital they raise and large firms spend their technology budgets on hyperscalers’ compute and services. A few in the sector have called for Ottawa to develop a national compute strategy to ensure Canadian companies and researchers have access to affordable infrastructure on which they can scale and sell. Innovation Minister François-Philippe Champagne recently told The Logic he’d discussed the subject with startups and cloud providers.

Amazon and Microsoft declined to say whether they’d been part of such talks. But Google is open to participating. “We know the demand is going to continue going forward,” Sebastian said. “It’s a shared responsibility—both public and private.” 

The hyperscalers see a role for themselves in Canadian AI. Sebastian cited the significant part  researchers here played in development of the technology, and the amount of startup activity in the field. “We can’t lose that,” he said. “We have to capitalize.”