Summary: BCE announced on March 16, 2026 that it would spend $1.7 billion on a purpose-built 300-megawatt AI data centre near Regina, Saskatchewan, with all four data halls scheduled for the end of 2027. CoreWeave and Cerebras have committed to all capacity and will bring their own compute hardware, lifting the combined investment toward $12 billion. The project reserves part of the capacity for Canadian sovereign AI workloads and is designed with closed-loop cooling rather than local water use.
Canadian telecom BCE made the announcement on March 16, 2026: it will put $1.7 billion into a purpose-built, 300-megawatt AI data centre located in the Rural Municipality of Sherwood on the outskirts of Regina, Saskatchewan. Going by Reuters, the finished facility will stand as the largest of its kind in the country. Capital outlays are concentrated within 2026 and 2027, and all four data halls are slated to be live by the end of 2027.
Every unit of capacity is already spoken for by two anchor tenants: CoreWeave, a GPU cloud operator, and Cerebras, an AI chip company — and both will ship in their own compute hardware. Count that equipment and the combined investment climbs toward $12 billion. Compute buildouts are the defining story of 2026 (our opening outlook called this one early), and a telecom-financed site on the Canadian prairies drawing tenants of this standing is itself a measure of how far AI compute demand has spread past the established hubs.
Breaking Down the Deal: 300 MW, $1.7 Billion, and Two Anchor Tenants
The division of labor comes down to “landlord builds the shell, tenants bring the silicon”: BCE takes care of the building, the power, and the connectivity, while CoreWeave and Cerebras handle their own compute equipment. BCE’s $1.7 billion goes toward land, structures, and power infrastructure; the GPUs and wafer-scale systems owned by the tenants lift the all-in figure to nearly $12 billion.
The jobs picture: around 800 construction roles while the site goes up, some 80 permanent positions split among Bell, CoreWeave, and Cerebras once it is running, and — based on the industry research the release cites — up to 750 further community jobs coming from suppliers and knock-on activity. Saskatchewan has roughly 1.2 million residents, so a single project of this size is genuinely rare.
Why Saskatchewan: Power, Fibre, and Cooling
Three pieces of infrastructure explain the location. On power: the site draws from the industrial grid rather than the residential one, using an allocation from SaskPower that already exists. President and CEO Mirko Bibic made a point of saying households and local companies face no cost increase, and that “there’s no public money allocated to this” — what the province supplies is a 300 MW power allocation, not a subsidy. On connectivity: SaskTel, the Crown-owned carrier, will extend fibre all the way to the campus and hook it into its national backbone, with a memorandum of understanding setting out the cooperation.
The cooling story carries equal weight. In Bibic’s words, the data centre “will not draw on local water sources at all” — a closed-loop technical system does the work — and Bell is studying a district energy arrangement that would put the waste heat to use. For a prairie province that guards its water closely, the design is a political necessity, and it is fast becoming standard practice for large AI campuses everywhere.
Bell AI Fabric: BCE’s Wager on AI
The whole campus sits at the center of BCE’s “Bell AI Fabric” strategy. A slice of capacity is held back for sovereign AI needs from Canadian Bell clients — governments and companies bound by data residency in Canada finally get a domestic option with genuine scale. In Bibic’s framing: “Saskatchewan wants to be at the epicentre of AI industrialization in Canada,” and “Regina is going to be one of the epicenters of AI industrialization in Canada.”
Financially, Bibic connected the momentum to a balance sheet that strengthened through 2025 and to the four-pillar strategic plan laid out at BCE’s October investor day — a plan that counts leading the enterprise market with AI-powered solutions among its pillars. The announcement, he said, “allows us to increase those growth targets,” yet the financial guidance for the underlying business stands unchanged. One thing to flag as you read: Bell Media owns BNN Bloomberg, and most of the quotes above come from Bibic’s own interview — pair them with Reuters’ independent reporting.
The Read for Developers and Product Teams
Three takeaways. Sovereign AI has moved from talking point to procurement line item: training and inference capacity resident in Canada now actually exists at scale, giving any product that touches data under residency requirements another compliance route. Telecoms, meanwhile, are formally in the AI infrastructure business. Fibre rights-of-way, power access, and land add up to built-in leverage over pure cloud providers, and the “telecom builds the shell, compute company brings the chips” playbook should show up in other markets too. Finally, anchor tenancy trims the barriers to expansion: a CoreWeave can grow without ever erecting a building, capex divides into a shell layer and a silicon layer, and supply scales faster as a result. Once the four halls go live at the close of 2027, North America gains another source of rentable large-scale training capacity.