- Bell Canada plans to build a total of 1.2 GW of sovereign AI data center capacity in Saskatchewan
- TD Cowen analyst Vince Valentini said Bell has a first-mover edge in Canada’s sovereign AI market
- Bell is avoiding the risky GPU-as-a-service model, focusing instead on providing data center shells
Bell Canada is ramping its sovereign AI compute ambitions, announcing plans to quadruple its data center footprint in Saskatchewan over the next decade.
The operator said it will add an additional 900 megawatts (MW) of data center capacity in the province, building on an initial 300 MW footprint it announced in March. At that time, it said the first stage of the project would come online in the first half of 2027.
Bell expects to have the 900 MW of incremental compute operational “well within 10 years,” TD Cowen’s team wrote in a note to investors.
The operator stressed that its new development will adhere to Saskatchewan’s Responsible Data Centre Development Principles, with the 900 MW add-on fueled by natural gas power generation set to be developed by one of Bell Canada’s partners. The facilities will also use closed-loop cooling technologies to minimize water usage, Bell said.
"Canada's next generation of economic growth will be built on Canadian infrastructure,” Bell Canada CEO Mirko Bibic said in a statement. “By creating a path to 1.2 GW of AI capacity in Saskatchewan, Bell AI Fabric will give governments, businesses, researchers and innovators the secure, sovereign foundation they need to deploy AI at scale and compete globally.”
Building a sovereign AI empire
Its expansion in Saskatchewan marks the continued growth of Bell’s AI Fabric, a sovereign AI offering the operator launched in May 2025. Bell also has AI Fabric data centers in British Columbia and Manitoba.
TD Cowen Managing Director Vince Valentini told Fierce Bell appears to be one of the most organized Canadian players moving into sovereign AI infrastructure, with a dedicated division, construction partners and links to large language model players already in place.
Though rival operator Telus is also eyeing the market, Bell has been “fast out of the gates,” Valentini said. While Telus’ new CEO has articulated plans to organize and do more in this space, Bell has already secured something of a first-mover advantage.
Valentini also praised Bell’s AI infrastructure business model. Notably, he said, the operator is not trying to be a neocloud.
“They are not building the AI compute themselves, they’re not buying the chips, they’re not putting the electronics inside the building. They’re just sort of doing the shell and having it all packaged up and ready, and then they lease out this space inside to various tenants,” Valentini said. “But they’re reserving the right to say, ‘hey, if you need help selling some of that compute…we can open doors for you.”
Spending money to make money
Despite this relatively capex-light approach, the investment is significant for Bell. Valentini noted that its original project in Saskatchewan was expected to cost around CAD 1.7 billion for 300 MW of capacity. Extrapolating from there, the additional 900 MW would raise the capex figure by another CAD 5.1 billion.
Given Bell’s annual capex is in the CAD 3 billion range, “that’s pretty material for them,” even if the money is spent over several years, Valentini said.
“It gives you a sense that this is a large chunk of their capital spending commitment over the next few years,” he added. “Any by our math, it starts to become material in 2028.”
According to the guidance Bell has provided, Valentini said the AI infrastructure business is expected to grow to roughly CAD 525 million of EBITDA over that timeframe. While it won’t be 30% or 40% of their revenue “it starts to become more meaningful.”
“I think most people view this as a pretty big new venture for BC,” he concluded.