Bell’s new data-centre plan shows it has an AI strategy that can actually make money

BCE and Saskatchewan will build a 300-megawatt, $1.7B AI data centre in Regina (opening 2027) that BCE forecasts will add $500M in annual revenue and >$250M in free cash flow, implying a ~20% ROIC if on budget. BCE is using a capital-light model with Cerebras and CoreWeave as 10-year non-cancelable anchor tenants and has secured power contracts with SaskPower and TransGas, plus ~425 MW reserved for future projects. The structure de-risks BCE’s AI push and makes further expansion likely via third-party partnership capital, which is positive for BCE’s infrastructure-led growth prospects.

Analysis

Telcos are finally leveraging three durable, hard-to-replicate assets—fibre footprints, long-term power access, and regulated domestic trust—to convert AI demand into annuity-like infrastructure economics. That combination shifts value away from pure-software winners (who capture model/IP rents) toward operators who can offer sovereign, low-latency, contracted capacity; the re-rating mechanism will be visible when recurring revenues and FCF growth replace cyclical capex narratives.

A capital-light partner model (anchor tenants + third‑party capital) materially reduces build/obsolescence risk versus a 100% self-funded roll‑out, but it also creates new counterparty concentration and vendor‑ecosystem exposure—outsized dependence on GPU/accelerator roadmaps and on a handful of cloud/stack partners. Electricity contracts de‑risk margin forecasts but create a second-order arbitrage: owners can flex power into higher‑margin use cases (burst AI, reserved private clouds, or spot compute markets) or monetize excess capacity via third‑party capital providers.

Key catalysts are contract wins with sovereigns/financials, additional long-term anchors, and third‑party JV announcements that compress funding risk; the primary reversals would be major anchor withdrawals, a sharp jump in power prices, or an industry pivot toward more distributed (edge) compute that reduces demand for large centralized sites. Time horizon: expect public financial recognition over 12–36 months as projects scale and cash flows become visible; binary risks can materialize much faster (days–months) around specific contract or construction events.

Key Decisions for Investors

• Long BCE (ticker: BCE) — 12–24 month horizon. Buy into initial headline pullbacks and size as an overweight in core Canadian infra allocation. Hedge execution risk by selling 9–12 month out‑of‑the‑money covered calls to finance cost basis; upside is multiple expansion if recurring AI infra cash flows scale, downside is limited by existing dividend + regulated telecom cash generation.
• Long CoreWeave (ticker: CRWV) or equivalent GPU-cloud exposure — 6–18 month horizon. Use long-dated calls if illiquidity or optionality preferred. Rationale: anchor-tenant demand for specialized accelerators should drive utilization and premium pricing; tail risk is a sudden shift to in‑house hyperscaler capacity or a large-scale hardware refresh cycle that devalues current fleet.
• Long Brookfield (ticker: BAM) — 18–36 month horizon. Allocate to capture asset-management/third‑party capital flows into data‑centre and energy-linked infrastructure. Reward: fee/asset valuation uplift as JVs and institutional capital roll in; risk: mark-to-market compression of alternative-asset multiples in a macro weakening scenario.
• Pair trade — Long BCE / Short GOOGL or META (dollar-neutral) — 12–36 month horizon. Idea: capture secular reallocation of sovereign and sensitive workloads to domestic providers while hedging broad AI demand exposure. Keep the pair size modest (e.g., 0.5–1% NAV) and use a 10% stop on the pair; major downside is hyperscalers retaining or undercutting pricing for these workloads.