OpenAI’s head of countries says Canada has the energy and land for data centres

OpenAI is evaluating Canadian data-centre partnerships, citing the availability of energy and land, according to its head of countries, George Osborne. The move follows OpenAI's April pause of a UK project pending more favorable regulation and energy costs, underscoring how power availability and policy are shaping AI infrastructure investment.

Analysis

This is not yet a demand award; the investable signal is that hyperscale AI compute is increasingly constrained by deliverable power, permitting certainty, and interconnection queues rather than chip availability. Canadian provinces with surplus hydro or nuclear generation could command higher long-term contracted-power values, but utilities will capture upside only where regulators permit rate-base expansion without forcing existing customers to subsidize data-centre loads. The near-term beneficiary is more likely land-and-power aggregators than merchant generators.

Brookfield (BN) is the cleanest liquid proxy because its infrastructure, renewable-power and real-asset platforms can package sites, generation, transmission and financing; a credible AI-campus transaction would support fee-bearing capital and asset-rotation multiples rather than merely power-price exposure. Cameco (CCJ) has a more indirect 6-18 month read-through: sustained AI-load forecasts improve the political and economic case for nuclear life extensions and new-builds, but this does not change uranium demand materially without provincial procurement commitments. Fortis (FTS), Emera (EMA), and Algonquin Power (AQN) require identifiable transmission, distribution, or generation capex before the theme becomes earnings-relevant.

Consensus may overvalue cheap electricity while underweighting time-to-power. A nominally low-cost Canadian site is commercially inferior if transmission upgrades, environmental approvals, or provincial data-sovereignty conditions delay energization by 24-48 months; that creates an advantage for incumbent powered-land operators and established cloud campuses. The thesis is falsified if prospective projects lack firm PPAs/interconnection agreements, or if provincial regulators cap dedicated-load returns and impose materially punitive reliability obligations.

Key Decisions for Investors

• No directional position solely on this exploratory signal; set an event-driven alert for a named Canadian site, utility counterparty, MW capacity, interconnection date, and contracted power price. Treat a signed PPA plus a sub-24-month energization schedule as the threshold for underwriting earnings impact.
• Accumulate BN on market weakness over a 3-12 month horizon as the diversified Canadian AI-infrastructure proxy, but size modestly: upside requires a disclosed development/financing mandate, while downside is broad real-asset multiple compression if rates rise. Exit the thematic leg if no identifiable AI-campus transaction or capital commitment emerges within two quarters.
• Use CCJ as a 6-18 month watch-list beneficiary rather than a direct data-centre trade. Add only if provincial load forecasts translate into nuclear refurbishment, life-extension, or new-build procurement; absent that policy conversion, uranium pricing remains driven by the broader contracting cycle rather than incremental AI demand.
• Avoid treating FTS, EMA, or AQN as automatic winners until regulators specify cost recovery and load-service terms. A utility announcement is actionable only if incremental capex is rate-based, customer concentration is capped, and the project does not require equity issuance that dilutes per-share growth.