On March 16, 2026, Canadian telecom BCE announced a $1.7 billion investment to build a 300-megawatt, purpose-built AI data centre in the Rural Municipality of Sherwood, just outside Regina, Saskatchewan. Reuters reports it will be the largest facility of its kind in Canada, with capital spending concentrated across 2026 and 2027 and all four data halls expected online by the end of 2027.
The capacity is already fully allocated to two anchor tenants — GPU cloud provider CoreWeave and AI chipmaker Cerebras, both bringing their own compute hardware. Include that equipment and the total footprint approaches $12 billion. In a year when compute buildouts are the running theme (our opening outlook flagged this early), the fact that a telecom-funded facility in the Canadian prairies can land tenants of this caliber says something about how far AI compute demand has spilled beyond the usual hubs.
The Deal at a Glance: $1.7 Billion, 300 MW, Two Anchor Tenants
Structurally, this is a “landlord builds the shell, tenants bring the silicon” split. BCE’s $1.7 billion buys land, buildings, and power infrastructure; CoreWeave and Cerebras supply the GPUs and wafer-scale systems that push the combined figure toward $12 billion.
On jobs: roughly 800 construction positions during the build, about 80 permanent roles across Bell, CoreWeave, and Cerebras once operational, and — per industry research cited in the release — as many as 750 additional community jobs through suppliers and knock-on activity. For a province of roughly 1.2 million people, this is a rare single-project scale.
Power, Fibre, and Cooling: Why Saskatchewan
The site logic revolves around three pieces of infrastructure. Power: the facility connects to the industrial grid rather than the residential one, drawing on an existing SaskPower allocation. President and CEO Mirko Bibic stressed that neither residents nor local businesses will see higher costs, and that “there’s no public money allocated to this” — the province’s contribution is the 300 MW power allocation, not a subsidy. Connectivity: Crown-owned SaskTel will run fibre to the site and tie it into its national backbone under an MOU.
Cooling is the other selling point. Bibic said the data centre “will not draw on local water sources at all,” relying on a closed-loop technical system, and Bell is exploring a district energy setup to repurpose waste heat. In a water-sensitive prairie province, that design is politically essential — and increasingly the default for large AI campuses everywhere.
Bell AI Fabric: A Telecom’s AI Pivot
The campus anchors BCE’s “Bell AI Fabric” strategy. Some capacity is reserved for sovereign AI demand from Canadian Bell clients — for governments and enterprises whose data must stay in Canada, this is the first domestic option at real scale. 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.”
On the financial side, Bibic tied the move to balance-sheet strengthening through 2025 and the four-pillar strategic plan unveiled at BCE’s October investor day, one pillar of which is leading the enterprise market with AI-powered solutions. The announcement “allows us to increase those growth targets,” he said, while guidance for the underlying business is unchanged. One caveat worth noting: BNN Bloomberg is owned by Bell Media, and most of these quotes come from Bibic’s own interview — read it alongside Reuters’ independent reporting.
What It Means for Developers and Product Teams
Three observations. First, sovereign AI is turning from slogan into a procurement category: there is finally at-scale Canadian-resident training and inference capacity, which matters for any product handling data with residency requirements. Second, telecoms are now formal AI infrastructure players. The combination of fibre rights-of-way, power access, and land is their natural leverage over pure cloud providers, and the “telecom builds the shell, compute company brings the chips” template should replicate in other markets. Third, the anchor-tenant model lowers expansion barriers: when a CoreWeave can scale without constructing buildings, capex splits into a shell layer and a silicon layer, and supply grows faster. When the four halls come online at the end of 2027, North America’s rentable large-scale training capacity gains one more source.
Sources
- Canada’s BCE to invest $1.7 billion in Saskatchewan AI data center — Reuters
- Bell plans $1.7B AI data centre in Saskatchewan as demand for computing power surges — BNN Bloomberg
- Media Backgrounder: Bell AI Fabric Saskatchewan Data Centre — BCE
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.