Summary

  • BCE reported Q2 capital expenditure of C$1.080bn, C$317m above the prior year and equivalent to 17.5% of revenue.
  • The increase reflected Bell AI Fabric construction in Canada and C$163m of Ziply Fiber investment in the United States; BCE does not publish a full project-level split.
  • Operating cash flow rose 11% to C$2.162bn, while free cash flow fell C$110m to C$1.042bn, mainly because of higher capital spending.
  • BCE cited important construction milestones in Saskatchewan and progress in Merritt, British Columbia, but did not define the physical stage or percentage complete.
  • The planned 300 MW Saskatchewan programme and C$1.3bn full-year capex increase versus 2025 are not Q2 commissioned capacity or Q2 project expenditure.
  • Combined Ateko and Bell Cyber revenue grew 29%; that measure is not Bell AI Fabric data-centre revenue, occupancy or utilisation.

The cash-flow divergence is the current infrastructure milestone

BCE’s operations produced C$2.162bn of cash, C$214m more than a year earlier. After capital investment and the other adjustments captured in free cash flow, C$1.042bn remained—C$110m less than in the comparable quarter. The direction of the two figures matters more than an isolated capex headline.

It shows that the group’s operating engine improved while the investment programme absorbed the gain and more. That is not evidence that Bell AI Fabric is underperforming. New infrastructure is expected to consume cash before it produces a return. It is evidence that the timing and quality of conversion now matter to BCE’s financial flexibility.

Free cash flow is the bridge between construction ambition and the capacity to fund dividends, debt reduction or further investment. The project’s strategic claim will ultimately be tested on how quickly today’s outflow becomes contracted and productive infrastructure.

C$1.080bn is a blended quarter, not an AI project invoice

Capital expenditure rose from C$763m to C$1.080bn, an increase of C$317m or 41.5%. BCE says the rise reflected higher Bell AI Fabric data-centre construction in Canada and C$163m of US spending on Ziply Fiber’s FTTP expansion.

The C$163m is the only project amount isolated in that explanation. BCE does not allocate the remaining capex between Saskatchewan, Merritt, ordinary Bell network investment and other programmes. It is therefore invalid to assign the whole C$1.080bn—or even the C$317m increase—to AI data centres.

Capital intensity moved from 12.5% to 17.5%. That is a useful group-level signal of a heavier build cycle, but it remains a ratio of total capex to revenue, not a measure of data-centre completion.

Full-year guidance sits on a different rung from quarterly actuals

BCE reconfirmed that 2026 capex is expected to increase by C$1.3bn from 2025, primarily because of Saskatchewan construction. It also retained guidance for capital intensity of about 20% and free cash flow of C$2.1bn to C$2.3bn, down 28% to 34%.

Those numbers describe management’s current full-year expectation. C$1.3bn is not the amount spent in Q2, not a complete Saskatchewan project budget and not a statement that every incremental dollar goes to Bell AI Fabric.

Keeping forecast and actual separate makes the next quarters auditable. Cumulative capex, year-end capital intensity and free cash flow can be compared with guidance without rewriting the guidance itself as achieved construction.

“Construction milestones” need a physical denominator

Management says Bell AI Fabric achieved important construction milestones in Saskatchewan and progressed the Merritt expansion. It does not specify whether those milestones concern earthworks, foundations, shells, substations, utility interconnection, internal electrical systems, cooling, network rooms, server installation or acceptance.

Each stage carries a different schedule and risk. A completed building shell may still wait for power. An energised electrical system may not have customer equipment. A technically ready hall may have no contracted load.

Without stage, percentage and site-level expenditure, the phrase confirms activity but cannot measure readiness. The useful next disclosure is not another adjective; it is a count of MW or halls at defined states.

Planned 300 MW must stay separate from live capacity

The current secondary report links the quarter to a previously announced 300 MW Saskatchewan programme. That figure represents intended project scale. BCE’s Q2 release does not say that 300 MW has been constructed, energised, accepted, reserved by customers or placed in service.

Capacity passes through planning, permitted, contracted power, construction, energisation, technical acceptance, customer fit-out and loaded operation. A planned number sits at the first end of that ladder. Moving it to the last end would erase most execution risk.

The most decision-useful update would state how many MW have secured utility supply, how many are under construction, how many have passed acceptance and how many support billed customers.

Enterprise revenue growth does not close the data-centre return question

Combined Ateko and Bell Cyber revenue increased 29% year on year. That is evidence of momentum in two enterprise technology businesses. It is not a data-centre utilisation figure and does not reveal whether Saskatchewan capacity has generated any revenue.

The distinction matters because services growth can support demand for a sovereign Canadian technology platform without measuring the economics of a specific physical site. Revenue may arise from consulting, managed services or cybersecurity rather than colocation or compute infrastructure.

BCE also notes that a prior-year delivery of the first Kamloops data centre did not recur in Q2 2026. Year-on-year comparisons therefore include a delivery-timing effect and cannot be read as a clean same-project growth curve.

Sovereignty value still requires customers and operating proof

Bell positions AI Fabric within a Canadian sovereign-infrastructure strategy. Domestic location can matter for data control, procurement, latency and regulatory preferences. But locality does not automatically produce demand or a return.

The release gives no customer commitments, power price, construction cost by site, occupancy, contracted recurring revenue or expected yield. It also does not state how the capital burden will be shared among Bell, utilities, partners or tenants.

Proof will come when physical readiness and commercial loading can be reconciled: capacity accepted, customer space fitted, contracted power consumed and revenue recognised. Until then, sovereignty is a strategic proposition supported by construction, not a measured operating outcome.

The next quarter should show conversion rather than another scale claim

BCE’s C$6.176bn of Q2 revenue rose 1.5%, and adjusted EBITDA increased 1% to C$2.702bn. Those group figures provide a funding base, while the capex programme raises the burden on free cash flow.

The next useful disclosure would connect cash to physical progress: site-level or programme-level capex, defined construction stages, energised MW, accepted halls, customer commitments and the date each asset begins generating revenue. It should also preserve the Ziply allocation so Canadian AI spending is not overstated.

Bell’s build is already economically visible because cash is being retained in the project instead of remaining free. It will become operationally visible when the same reporting identifies capacity that is powered, accepted and paid for.

Sources