NextFin News - BCE Inc. is betting C$1.7 billion that Canada's biggest telephone company can reinvent itself as a landlord to the artificial-intelligence boom. Bell Canada's new 300-megawatt AI data centre in the Rural Municipality of Sherwood, Saskatchewan - the largest purpose-built AI data centre development in Canada once complete - is already fully contracted to U.S. tenants Cerebras and CoreWeave under long-term, non-cancellable agreements. But the bill for that transformation arrives immediately: BCE cut its 2026 free cash flow guidance by as much as 34% and warned that capital intensity will jump to roughly 20% of revenue, up from a sub-15% target, as the build-out consumes cash years before the first data hall powers on in early 2027.
The announcement, made March 16, 2026 by Bell Canada and the Government of Saskatchewan, is the clearest signal yet that a legacy telecom operator is attempting to escape the slow-growth trap that has kept its valuation near historic lows. BCE shares trade at 4.8 times trailing earnings and yield 5.4% - pricing that says the market sees a business in structural decline, not an AI infrastructure platform. The question investors now face is whether the Sherwood facility is the first brick in a higher-multiple future, or an expensive distraction that will bleed free cash flow while the core wireline and media businesses keep shrinking.
The Deal: C$1.7 Billion for the Shell, C$10 Billion of Chips Paid by Tenants
The structure of the transaction is more revealing than the headline size. Bell's capital investment is approximately C$1.7 billion for the physical facility - the building, the cooling systems, the power infrastructure. The compute hardware, which BCE's chief executive has estimated at roughly C$10 billion in combined value, will be funded and supplied entirely by the two anchor tenants. Cerebras will install its wafer-scale AI inference and training systems; CoreWeave will deploy scalable NVIDIA GPU compute.
"This project is Bell's largest-ever investment in Saskatchewan - an unprecedented commitment to Saskatchewan's technology ecosystem that places the province at the centre of Canada's AI-driven economic growth. Once complete, it will be the largest purpose-built AI data centre development anywhere in Canada."
Bell said in its announcement. The provincial government, through Premier Scott Moe, called the investment "great news for Saskatchewan's economy," adding that it would create jobs, strengthen research capacity, and enable new businesses built on advanced AI capabilities.
The economics, as laid out in the company's investor materials, resemble toll-road infrastructure more than telecommunications: at full run rate the facility is projected to contribute approximately C$500 million of revenue, C$400 million of adjusted EBITDA - an 80% EBITDA margin - and more than C$250 million of free cash flow, with a project-level internal rate of return of about 20%. The project is expected to be leverage-neutral on a run-rate basis and to support accelerated deleveraging, with no change to BCE's 2027 or 2028 leverage targets. Over time, Bell projects economic value of up to C$12 billion for the province, including job creation, tax revenues, and broader spillovers. Construction will support at least 800 trades and engineering jobs, with a minimum of 80 full-time roles once the facility is fully operational; industry research cited by Bell suggests as many as 750 additional community jobs could follow.
The timeline is tight. Construction on the first phase is expected to begin in the second quarter of 2026, with the facility coming online in stages: two data halls in the first half of 2027, two more in the second half, and all four halls at full run rate by the end of 2027. Revenue is expected to begin in the second half of 2027 as the initial phases come online. The 160-acre site sits a significant distance from residential areas, will launch with a Power Usage Effectiveness of approximately 1.3, and uses a closed-loop cooling system that draws no municipal water. SaskPower will supply electricity through a dedicated industrial feed separate from the residential grid, SaskEnergy will provide natural gas for backup generation, and SaskTel will link the campus to Bell's national fibre backbone. Bell's C$1.7 billion capital investment is approximately C$1.3 billion in 2026, partially offset by C$400 million in setup fees and prepayments, and funded through debt and cash on hand.
The Immediate Cost: Free Cash Flow Guidance Cut by a Third
Because the Saskatchewan project is material to BCE's financial profile, the company updated its 2026 guidance targets on the day of the announcement. The revenue and adjusted EBITDA ranges are unchanged - 1% to 5% revenue growth and 0% to 4% adjusted EBITDA growth. What moved is the cash line, and it moved sharply.
Free cash flow growth for 2026 is now expected to fall between negative 34% and negative 28%, to C$2.1 billion to C$2.3 billion, down from a prior target of 4% to 10% growth, or C$3.3 billion to C$3.5 billion. Capital intensity is expected to rise to roughly 20% of revenue, reflecting approximately C$1.3 billion of incremental capital expenditures, up from a sub-15% target. Adjusted EPS growth guidance of negative 11% to negative 5% was left unchanged.
The build-out is already visible in BCE's quarterly results. Second-quarter 2026 capital expenditures rose 41.5% year over year to C$1.08 billion from C$763 million, driven by greater investment in Bell AI Fabric data centres in Canada and US$163 million of American spending supporting Ziply Fiber's fibre-to-the-premises expansion. Free cash flow fell 9.5% to C$1.042 billion, even as cash flow from operating activities increased 11% to C$2.162 billion. Revenue was C$6.176 billion, up 1.5%, and adjusted EBITDA rose 1% to C$2.702 billion. Net debt leverage stood at approximately 3.7 times.
For a stock that trades on free cash flow and dividends, the arithmetic matters. BCE's annualized common share dividend is C$1.75 per share; with roughly 932.5 million common shares outstanding, that is about C$1.63 billion a year. The pre-announcement free cash flow guidance of C$3.3 billion to C$3.5 billion covered the payout roughly twice. The revised C$2.1 billion to C$2.3 billion still covers it, but the cushion narrows to about 1.3 to 1.4 times - adequate, though far less comfortable for a company that has marketed itself as a defensive income holding.
Market Reaction: Shares Edge Higher, But the Re-Rating Has Not Started
As of midday trading on September 14, 2026, BCE shares on the Toronto Stock Exchange were changing hands at C$32.76, up C$0.37, or 1.14%, from a previous close of C$32.39, with a 52-week range of C$29.66 to C$36.25. The stock's market capitalization stood at roughly C$30.5 billion, its price-to-earnings ratio at 4.8 times trailing earnings, and its dividend yield at 5.4%, with an ex-dividend date of September 15, 2026. The modest gain suggests investors are treating the Saskatchewan announcement as priced in - the market is rewarding the 20% project-level return but withholding judgment on whether a single 300-megawatt facility can re-rate a C$30 billion telecom.
Why Saskatchewan: Power Is the Scarce Resource in the AI Race
Power availability was the primary factor in choosing Saskatchewan, BCE's chief executive said. That answer captures the central bottleneck of the AI build-out. A 300-megawatt facility draws roughly the same electricity as the average consumption of a city of 300,000 people. In 2026, the scarce resource in AI infrastructure is not land, not capital, and increasingly not even chips - it is reliable, affordable power with a grid connection that can be permitted and built on a usable timeline.
The deal structure reflects this reality. Bell is not buying GPUs and hoping demand shows up. It is monetizing the two assets with the longest moats in AI infrastructure - the power connection and the building - while passing the technology-obsolescence risk to its tenants. If NVIDIA's next architecture renders today's accelerators uncompetitive, that is CoreWeave's problem, not BCE's. Bell collects rent on the shell.
This is the model that digital-infrastructure REITs such as Equinix and Digital Realty pioneered in the cloud era, transplanted into AI. The difference is that hyperscale cloud providers are now both the biggest customers and the fiercest competitors of the colocation landlords, squeezing margins in the traditional business. Bell's answer is to anchor on sovereign demand - a significant portion of the facility's power is dedicated to sovereign AI compute, ensuring that Canadian government agencies, researchers, and enterprises can access top-tier AI power while keeping data within Canada under Canadian law, meeting strict chain-of-custody and residency requirements.
The sovereign angle is not merely rhetorical. Bell has entered into an agreement with the George Gordon First Nation on Indigenous procurement participation and workforce development, and is exploring waste-heat reuse in a development project led by George Gordon Developments. Academic agreements with Saskatchewan Polytechnic and the University of Regina will support strategic AI use cases and explore a district energy system that could route waste heat to nearby university campuses. SaskTel is not just a utility counterparty; it is a go-to-market partner for AI-powered products and solutions in Saskatchewan, connecting the campus to provincial enterprise, public-sector, research, agriculture, energy, financial services, and critical-infrastructure use cases.
The Saskatchewan facility is not the whole of Bell's AI infrastructure push. By the second quarter of 2026, Bell AI Fabric had approximately 335 megawatts of contracted capacity, including the 300-megawatt Sherwood project, and the company said it had line of sight to roughly 800 megawatts of AI data centre capacity across Canada when its British Columbia facilities are counted. The Sherwood campus is being built under a long-term strategic partnership with Bird Construction, named lead construction partner in May 2026, under which Bird becomes Bell's preferred construction partner for a multi-year, Canada-wide AI data centre buildout. Bird will issue BCE warrants to acquire up to 2,625,000 common shares, with 750,000 vesting on delivery of the Sherwood facility. Satellite imagery analysis confirmed that earthworks were underway at the site as of May 9, 2026, next to an existing SaskPower substation.
Cyclical Capex Spike or Structural Re-rating? The Case for Structural
The central question for BCE investors is whether this is a cyclical capital-expenditure spike or a structural repositioning of the company. The distinction is not academic: cyclical capex destroys value when the cycle turns, while structural capex re-rates the multiple. Three pieces of evidence point toward structural.
First, the demand is pre-sold. The full 300 megawatts is contracted under long-term, non-cancellable agreements, giving the facility what Bell describes as infrastructure-like cash flows with 100% capacity utilization from day one. Second, the returns are contractually defined - roughly 20% at the project level, with an 80% EBITDA margin at run rate - economics that look nothing like BCE's low-single-digit top-line growth in its legacy wireline business. Third, the binding constraint is physical rather than commercial: power connections take years to permit and build, so the sites that secure grid capacity early gain a durable advantage.
Management has raised its ambition accordingly. BCE lifted its AI-powered solutions revenue target to C$2.0 billion by 2028, from C$1.5 billion, while leaving baseline capital intensity unchanged outside the Saskatchewan project. Its 2025-2028 outlook calls for revenue compound annual growth of 2% to 4.5%, adjusted EBITDA CAGR of 2% to 3%, and free cash flow CAGR of up to 16.5%.
The Counter-Thesis: Concentration, Power Risk, and the Value-Trap Reading
The strongest argument against BCE's pivot is concentration. Two private U.S. companies - Cerebras and CoreWeave - account for 100% of the facility's contracted capacity. Cerebras is betting on wafer-scale computing, an architecture that has promised revolutionary performance for years but has yet to prove itself at hyperscale. CoreWeave's GPU-rental model faces intensifying competition as Amazon, Microsoft, and Google bring more AI compute in-house and as chip supply loosens. If either tenant stumbles, BCE's "infrastructure-like" cash flows could prove less infrastructure-like than advertised. Long-term, non-cancellable contracts are only as good as the credit standing of the obligor, and neither tenant is a rated investment-grade counterparty.
There is also a power-price risk specific to the Saskatchewan structure. The province's industrial electricity rates are low today, but a 300-megawatt load gives the crown-owned utility a single, concentrated point of leverage. If SaskPower renegotiates rates at the next contract reset, the 20% project IRR could compress quickly, given that electricity is likely the largest operating cost of an AI data centre.
And then there is the valuation question that underlies the entire trade. BCE trades at 4.8 times trailing earnings not because the market is irrational, but because investors believe the legacy wireline, wireless, and media businesses are in structural decline - wireless pricing competition with Rogers Communications and Telus has compressed margins for years, and cord-cutting continues to erode the media segment. At C$500 million of run-rate revenue against BCE's roughly C$24.5 billion 2025 revenue base, the Saskatchewan facility represents about 2% of the company. It is a start, not a salvation. For the stock to re-rate, Bell AI Fabric must prove it can replicate Sherwood several times over, fast enough to offset the decay in the core.
History is not on the company's side. Canadian telecommunications operators have a long record of grand infrastructure bets that took decades to earn their cost of capital. The fibre build-out of the 2010s produced modest returns. Investors in this sector have been burned before by "transformational" capital-expenditure stories, and skepticism is earned rather than reflexive.
The Falsifying Signal: What Would Prove the Thesis Wrong
The thesis that BCE is structurally re-rating rather than chasing a cyclical boom rests on two observable conditions. First, capital intensity must fall back toward or below 15% of revenue once the facility is substantially complete. If capital expenditures remain above 18% of revenue through 2027 after the Sherwood build ends, the "disciplined growth" narrative is broken and the market will be right to treat this as a capex spiral. Second, both anchor tenants must remain creditworthy and fully utilizing capacity through 2027 and 2028. A tenant default, a material renegotiation of terms, or a delay in the first-half-of-2027 in-service date would confirm the value-trap reading.
Who Benefits, Who Is Exposed, and What to Watch
If the pivot works, the beneficiaries are clear. BCE shareholders would see the stock re-rate from a dying-telecom multiple toward a partial infrastructure multiple - the gap between a 4.8 times earnings multiple and a 20% project IRR is the opportunity. Saskatchewan wins up to C$12 billion in projected economic value, more than 800 construction jobs, and a foothold in the AI economy. Canadian AI sovereignty gains a meaningful chunk of domestic compute capacity. And the construction supply chain - Bird Construction, Alton Tangedal Architect, George Gordon Developments, and the Saskatchewan-based contractors already on site - gains a multi-year pipeline.
The exposed are equally clear. Dividend-focused investors face a narrowing free cash flow cushion; if capital intensity stays elevated or the core business weakens further, coverage of the C$1.75 dividend could compress toward one times. The legacy telecom business must continue funding the AI build while absorbing wireless pricing pressure and media cord-cutting. And the province of Saskatchewan is exposed to the reputational and fiscal consequences of a deal that was sold as a C$12 billion windfall.
Three scenarios frame the path. In the base case, the facility comes online on schedule in 2027, both tenants perform, capital intensity normalizes toward 15% by 2028, and BCE's free cash flow CAGR of up to 16.5% through 2028 becomes achievable - the stock re-rates modestly. In the upside case, AI demand accelerates, Bell AI Fabric announces additional pre-leased campuses under the Canada-wide Bird partnership, and the market assigns a partial infrastructure multiple; the C$2.0 billion AI-powered solutions target for 2028 is exceeded. In the downside case, construction delays push the in-service date past 2027, one anchor tenant renegotiates, or power costs rise; capex stays elevated, dividend coverage falls toward one times, and the stock remains a value trap.
What to watch, in order: the third- and fourth-quarter 2026 capital expenditure prints, which should show the C$1.3 billion incremental spend ramping; the first-half-of-2027 in-service milestone for the first two data halls; any update on tenant creditworthiness or additional pre-lease announcements; and, above all, the 2027 capital intensity guidance - the single best read on whether this is disciplined growth or a capex spiral.
BCE is not merely building a data centre; it is buying an option on Canada's AI future, paid for in today's free cash flow. That option is cheap only if the legacy business does not bleed faster than the new one grows.
Explore more exclusive insights at nextfin.ai.
