BankablePowering AI · The Alberta Model · one-page verdict

Meta's $13B Alberta bet: a bridge with no far bank

A gigawatt of AI compute, contracted in about two years in a market where everyone else waits four to seven. The transaction is the most honest firm-power structure in AI right now. The label on it is not. Companion to the video episode; every number below traces to the cited research file.

The deal — Bridge

Real cost-causation, real risk transfer, a genuine answer to the queue problem. If you want to know how AI power gets built, read this deal.

The label — Greenwash

"Matched with 100% clean and renewable energy" describes an accounting arrangement, not the electrons. The physical product is new unabated gas for 30 years.

Alberta didn't out-subsidize anyone. It out-structured them — and then quietly removed the deadline that would have forced the bridge to land somewhere.

The five gates, scored

Swipe the table to see every column →

Gate	The evidence	Score
Speed-to-Power how fast do electrons arrive	First load H2 2028, roughly two years from announcement, via a 250 MW agreement against an existing fleet. Full gigawatt rides the new plant in H2 2030. Elsewhere the interconnection queue is 4–7 years.	9/10
Firmness is it there at 3am	Combined-cycle gas plus a fleet-backed agreement — dispatchable, no intermittency exposure. The load side is the harder half: AI training swings up to 90% at frequencies to 30 Hz, and the grid operator now caps ramping at 10 MW/min.	10/10
Economics does it pencil	Fuel is genuinely cheap. But at C$4,935/kW the capital line is roughly 4× the fuel line — see below. Tolling puts construction risk on the owners; the offtaker is AA-.	8/10
Truly Clean does the label survive the physics	New unabated gas, ~150 MMcf/d, on a grid already ~5× the national carbon intensity. Matching is annual and contractual, not hourly and physical. Carbon capture is marketing optionality, not funded scope.	4/10
Scalability does it repeat	Site permitted for 1,864 MW — Phase 1 is exactly half. A 21.1 GW queue waits behind a 1,200 MW cap. Docked because the intake is gas-only and Phase 2 is undefined.	8/10

Scores derived from the sourced claims in the research file, not from the working hypothesis. Full [S#] mapping in the workbook's Sources sheet.

The finding that surprised me: this is not a cheap-gas story

Everyone quotes Alberta's gas. At AECO ~C$1.92/GJ and a modern heat rate, fuel is about C$13/MWh — genuinely cheap, and genuinely not the point.
Build the delivered cost properly and it lands near C$89/MWh: capital recovery ~C$57, transmission ~C$14, fuel ~C$13, fixed O&M ~C$5, carbon ~C$0.
That is capital outweighing fuel about 4 to 1, and capital is roughly two thirds of the delivered cost. The turbine shortage — not the gas price — is what sets the price of AI power.
In the workbook's sensitivity sweep, CAPEX moves delivered cost ~C$35/MWh across its range; gas price moves it ~C$13. Utilization moves it most of all at ~C$40. Change one number and watch the ranking hold.

The carbon cost is zero — and the reason matters more than the number

Alberta's industrial system charges a large emitter only on the intensity above a benchmark. The electricity benchmark is 0.37 t CO₂e/MWh — explicitly set to "the performance of the best combined-cycle natural gas generator."
A best-in-class new plant runs near 0.35. It therefore pays carbon cost on approximately nothing. That is not a loophole anyone snuck through; it is the benchmark working exactly as designed.
Here is the part worth sitting with: in the model, raising the carbon price from C$95 to C$170/t changes delivered cost by exactly zero. Below the benchmark, the price is irrelevant. Only the benchmark bites — tighten it to 0.30 and C$4.75/MWh appears out of nothing.
The public argument is about the carbon price. The number that actually decides this plant's carbon bill is the one almost nobody is arguing about.

The calendar nobody has connected

Under the federal Clean Electricity Regulations, a unit commissioned after 2025 faced a performance standard from January 1, 2035 — a five-year cliff after this plant's 2030 start, and a hard question for anyone financing a C$4.6B unabated asset.
Then: a Canada–Alberta memorandum (2025-11-27) suspends those regulations in Alberta. The Implementation Agreement (2026-05-15) holds them "in abeyance" pending a court reference. Alberta froze its carbon price at C$95/t; the committed floor reaches C$130 only in 2040.
The same memorandum commits Alberta to a data-centre policy framework by 2026-07-01. The carbon détente and the data-centre strategy are the same document.
Then the calendar runs: regulatory approval June 25 → final investment decision July 2 → groundbreaking July 8. The cliff disappeared roughly ten weeks before the money committed.
That is what "a bridge with no far bank" means. Bridges are defensible when something forces you to reach the other side. Here, the forcing function was removed — and the renewables that were supposed to be the far bank had their worst build year since 2018.

What would change the verdict

Truly Clean moves up if Meta signs new Alberta wind or solar — the corporate market contracted 3.3 GW here between 2019 and 2025, so the machine demonstrably works. None announced yet. That gap is the whole asterisk.
Truly Clean moves up further on hourly matching rather than annual, or on carbon capture reaching a funded investment decision rather than a project page.
Economics moves down if the fuel and carbon pass-through terms land on the offtaker in a way the public filings do not yet show — the single biggest unknown in the structure.
The whole thing re-rates if AI capex cracks. Strip the anchor tenant out and Alberta owns a 932 MW merchant plant. The counterparty's own risk disclosures name exactly that.

Bankable — which hard-tech bets actually pencil out. The editable workbook came with this download: change the gas price, the capex, the utilization or the carbon benchmark and watch delivered C$/MWh move — including the sweep that shows the carbon price doing nothing. The video is at bankable.show.

Prepared from public data by David Zheng, P.Eng. — over a decade commercializing first-of-a-kind energy technology, much of it in the industrial region this plant is being built in. The delivered-cost build-up, the capital-to-fuel ratio and the carbon-benchmark cliff are derived figures, computed in the accompanying workbook from cited inputs; financing terms are assumptions, clearly marked. The C$4.6B is Phase-1 all-in on a site permitted for twice that, so the headline C$/kW slightly overstates steady-state unit economics. Illustrative, not project-grade engineering. We judge deal structures and technologies, never stocks; nothing here is investment advice. Advisory: bankable.show/advisory · hello@bankable.show