The Product Is Time-to-Power

Start with the scale. Meta’s Sturgeon County campus is its first data centre in Canada and its largest anywhere outside the United States. The announced capital spend exceeds CA$13 billion, plus CA$60 million for local roads and water infrastructure. Construction peaks near 3,000 workers; long-term operations support roughly 300 jobs. That works out to about CA$43 million of capital per permanent position.

That ratio is not an argument against the project. It tells us what the project actually is: a piece of power and digital infrastructure, not a labour-intensive factory. At full build-out, its one-gigawatt load is roughly three-quarters of Edmonton’s electricity demand. The scarce input is neither land nor cooling water—the campus uses closed-loop liquid and dry cooling—but a firm electrical connection on a known date.

That is Alberta’s product. In PJM and other congested markets, a multi-year grid queue can outlast the commercial life of the chips being installed. Meta expects first load in the second half of 2028, roughly two years after groundbreaking. For an AI developer, two years of additional compute revenue can matter more than shaving a few dollars from each megawatt-hour.

Alberta is not primarily selling the cheapest electricity. It is selling a date you can finance against.

When a Grid Slot Becomes Property

The speed did not come from spare capacity. By mid-2026, Alberta’s queue had grown to 42 large-load projects requesting 21.1 GW, while provincial peak demand was only about 12 GW. AESO responded by limiting near-term large-load access to 1,200 MW. Greenlight received 970 MW and Keephills 230 MW; the first phase was full.

Scarcity became visible during an approved transfer window. A company sold a 180 MW interconnection allocation for $18 million—roughly $100,000 per megawatt for the place in line, before buying a single electron. Grid access had acquired a market price.

Alberta then translated that scarcity into policy. Bill 8 created a bring-your-own-generation route and strengthened cost causation: the customer that creates the incremental system cost is expected to pay it. A related levy reinforced the signal—2% on computing equipment for a grid-dependent data centre, 1% when the project brings generation but stays connected, and zero when fully off-grid.

Two Supply Layers, One Financeable Contract

Meta’s power arrives in two layers. The first is a 250 MW, ten-plus-year supply agreement with Capital Power using existing plants, enough to energize the initial phase in late 2028. The second is Greenlight: a new 932 MW combined-cycle gas plant planned for the second half of 2030. The site is approved for 1,864 MW, so phase one uses only half its permitted capacity.

Meta does not own Greenlight. Morgan Stanley Infrastructure Partners and Pembina each hold 47.5%; developer Kineticor retains 5%. Roughly 60% of the CA$4.6 billion project is expected to be debt-financed. Meta signs a long-term tolling agreement: it pays for the plant’s available output much as a shipper reserves pipeline capacity.

The elegance is in the allocation. The owners carry construction and cost-overrun exposure. Meta supplies the long-duration payment obligation and bears the risk that it reserves more capacity than it ultimately needs. Its double-A credit quality turns that obligation into something lenders can underwrite. The technology is conventional; the contract is what makes the scale financeable.

Bankability is not the absence of risk. It is putting each risk with the party designed to hold it.

The Load Is Big—and Violent

AI data centres create a second problem beyond sheer volume. During training, thousands of GPUs enter and leave compute phases together. Tesla has published examples of load swings reaching 90% at frequencies as high as 30 hertz. Meta’s own engineers once used a software flag named pytorch_no_powerplant_blowup to burn dummy compute during gaps and keep the electrical draw flatter.

AESO’s June 2026 guidance classifies such facilities as non-conforming loads and indicates that data centres may have to limit ramping at the connection point to roughly 10 MW per minute. A one-gigawatt campus cannot simply pass every GPU fluctuation through to the transmission system.

This is where batteries move beyond backup. A battery energy storage system can respond in milliseconds, absorbing a sudden drop in server demand or filling a sudden rise. xAI’s Memphis campus uses 168 Tesla Megapacks with more than 650 MWh of storage for this job. Capital Power says batteries at its Genesee campus could unlock 500 MW of near-term data-centre capacity. No public filing yet confirms a BESS inside Meta’s Sturgeon County fence, but the system need is clear: bulk energy from firm generation, fast compliance from storage.

Cheap Fuel, Expensive Steel

Alberta’s gas advantage is real. AECO gas around CA$1.90/GJ can translate to approximately CA$13–15/MWh of fuel cost in an efficient combined-cycle plant. At full output, Greenlight could consume roughly 150 million cubic feet per day, creating a large, steady new customer for the basin.

But cheap gas no longer means a cheap gas plant. CA$4.6 billion divided by 932 MW is almost CA$4,900/kW—roughly two to three times the cost of comparable projects several years ago. Turbine shortages, construction inflation and shared infrastructure for future expansion all matter. The molecule is cheap; the steel, equipment and financing are not.

Meta is also expected to pay about CA$100 million per year in transmission charges and fund related grid upgrades. Alberta estimates the added load could reduce the transmission component of household bills by around 6% as fixed costs spread over a larger base. That is a government estimate, not an independently audited outcome, and gas demand or pool prices can move in the other direction. Still, the structural principle is sounder than putting the expansion on captive ratepayers: the anchor customer pays the freight.

What “100% Matched” Does—and Does Not—Mean

Meta says the campus’s electricity use will be matched with 100% clean and renewable energy. The word doing the work is matched. In practice, the campus draws from an Alberta grid that is roughly 60% gas-fired, backed by existing thermal plants and a new 932 MW gas facility. Renewable-energy certificates or contracts account for an equivalent quantity of renewable generation over a period of time. They do not identify the electrons serving the GPUs at 3 a.m.

Meta’s record is stronger than simple paper purchasing: it typically supports new wind and solar projects, and corporate buyers contracted about 3.3 GW of Alberta renewables between 2019 and 2025. The missing fact is project-specific. Meta has not yet announced the Alberta renewable projects that will match this campus.

Meanwhile, Alberta paused renewable approvals for seven months, tightened siting rules and saw new renewable construction fall to its lowest level since 2018. Even TD Economics has pointed to the tension between recruiting gigawatt-scale loads and slowing the new supply meant to match them. The gas plant has a contract, approval and final investment decision. The renewable side remains future tense.

The Calendar Behind the Carbon Risk

Regulation mattered to financing. Under the original federal Clean Electricity Regulations, a gas plant entering service in 2030 could have faced a near-zero emissions constraint from 2035—a five-year compliance cliff for a multi-decade asset. In November 2025, Ottawa and Alberta signed a memorandum that suspended the federal rule in Alberta while a constitutional dispute proceeds. The same agreement called for a provincial data-centre investment framework by July 1, 2026.

Then the sequence tightened: Bill 8 in December; a carbon-pricing agreement in April; implementation agreement in May; Greenlight approval on June 25; final investment decision on July 2; Meta groundbreaking on July 8. A calendar does not prove favouritism. It does show that the most serious federal compliance uncertainty moved out of the financing path before capital committed.

Greenlight advertises carbon-capture optionality, but optionality is not a funded scope. Alberta’s industrial carbon price remains CA$95 per tonne and reaches a minimum effective credit price of CA$130 only in 2040. A new best-in-class combined-cycle plant may also sit close to the power-sector performance benchmark, limiting its payable emissions difference. The bridge therefore has no legislated far bank today.

The Split Verdict

Speed-to-power: 9/10. Roughly two years to first electricity is the proposition. Firmness: 10/10. Existing fleet plus new combined cycle is dispatchable, and AESO’s rules force the load itself to behave. Economics: 8/10. Gas is extremely cheap and the risk structure is clean, offset by unusually expensive plant construction. Scalability: 8/10. The site can double and the queue is deep, though turbine supply, policy and social licence impose limits. Truly clean: 4/10. Meta has a credible history of renewable additionality, but today’s physical foundation is new unabated gas under a relaxed regulatory path.

So the transaction earns the label bridge. That is praise: it adds supply, assigns incremental cost to the user and parks risks with parties equipped to manage them. The environmental label earns greenwash. Not fraud—accounting. But accounting that lets a gas-fired gigawatt wear a renewable badge while hourly physical supply tells another story.

Three developments could change the judgment. First, Meta could sign Alberta wind and solar at the scale required to match the campus with genuine additionality. Second, carbon capture could move from optional language to funded engineering, budget and schedule. Third, the AI demand curve could validate the single-anchor exposure built into a very large plant. Until then, the correct way to read the deal is simple: follow the electrons, then the contracts, then the risk.

The deal is a bridge. The label is greenwash. Both can be true at the same time.