CoreWeave + Cerebras + BCE data center
CoreWeave, Cerebras and BCE announced plans to build a major AI data center in Saskatchewan, expanding multi‑vendor capacity outside NVIDIA's direct control reported. The facility underscores growing third‑party capacity bets to serve inference demand beyond the hyperscalers' own fleets reported.
BCE/Bell said the build will cost C$1.7 billion ([investing.com)] for a 300‑megawatt campus in the Rural Municipality of Sherwood, just outside Regina, Saskatchewan ([bce.ca)]. The campus capacity is split with Cerebras securing 160 megawatts ([investing.com)] and CoreWeave taking 140 megawatts ([investing.com)], and Bell’s project notes CoreWeave’s racks will run on NVIDIA GPUs ([investing.com)]. Construction is slated to begin this spring 2026, with the first data halls expected online in the first half of 2027 and full build‑out targeted by the end of 2027 ([bce.ca)], and Bell projects the build phase will support about 800 construction jobs plus roughly 80 permanent roles afterward ([bce.ca)]. Bell describes the site as Canada’s largest purpose‑built AI data centre and projects it could generate up to C$12 billion in economic value for Saskatchewan; the campus will connect to Bell’s national fibre backbone via a SaskTel partnership ([bce.ca)]. Industry coverage reports the campus is effectively pre‑leased to the two tenants ([datacenterdynamics.com)]; CoreWeave ended 2025 with roughly 850 MW of active power across about 43 data centres and has a recent $2 billion equity arrangement with NVIDIA to accelerate a >5‑GW buildout by 2030 ([datacenterdynamics.com)]. Cerebras’s 160‑MW slot aligns with its wafer‑scale approach—Cerebras has been rolling out CS‑3 wafer‑scale systems and expanded a network of inference datacentres in 2025—signalling the space will likely host large CS‑class clusters rather than conventional single‑socket GPU pods ([cerebras.ai)].
