Newfoundland and Labrador generates about 43 TWh of green electricity a year and sends most of it across the Québec border. The Churchill Falls renegotiation now underway is the first chance in fifty-seven years to reserve a scheduled block of Labrador power for the highest-value electricity load on earth — AI compute. It is currently on track to miss it.
Newfoundland and Labrador owns one of the cleanest large power systems in North America. Churchill Falls alone — 5,428 MW, roughly 34 TWh a year — has been sold under contracts that captured a small fraction of what firm, carbon-free power commands today. In May 2026 the Independent Churchill River Review Committee found the December 2024 Memorandum of Understanding not in the long-term public interest as written, citing among other flaws limited provincial rights to use the power in-province. A definitive agreement is being renegotiated and remains unsigned. The province's stated priority — more power for Labrador's economy — has so far been framed almost entirely around critical minerals. Mining matters. But mining is not the buyer that has repriced electricity across this continent. Data centres are, and no one has yet written them into the plan.
| Asset | Capacity | Annual energy |
|---|---|---|
| Churchill FallsCFLCo — NL Hydro 65.8% / Hydro-Québec 34.2% | 5,428 MW | ~34–35 TWh |
| Muskrat FallsCommissioned 2020–21 | 824 MW | ~4.9 TWh |
| Island hydroBay d'Espoir and others | ~1,000+ MW | ~4 TWh |
| Provincial total (2023) | — | 44.3 TWh generated 43.1 TWh renewable |
That is 97.4% renewable generation, hydro-dominated, dispatchable, and concentrated at a handful of interconnection points. For scale: a 1,000 MW AI campus running continuously consumes about 8.76 TWh a year. Churchill Falls is on the order of four such campuses in energy terms — a genuinely significant resource, and a finite one. This is an argument for allocating it deliberately, not for pretending it is limitless.
Under the 1969-lineage contract structure, Churchill Falls base energy was priced on the order of 0.2¢/kWh as an export price. Media-reported figures put the cumulative split to 2019 at roughly $28 billion in profit to Hydro-Québec against $2 billion to Newfoundland and Labrador.
The December 2024 MOU headline was an average of 5.9¢/kWh in 2024 dollars. The Independent Review's technical briefing decomposes that average into its real path: about 3.8¢/kWh from 2024 to 2041, then about 16.7¢/kWh from 2042 to 2075. The value is back-loaded into decades when the plant will be approaching a century old.
Meanwhile, in February 2026, Hydro-Québec proposed a dedicated data-centre tariff of 13¢/kWh — roughly double its large-industrial rate — and still expects takers. Québec is deliberately rationing and repricing exactly the hydro-seeking compute demand that Labrador could serve. Under the MOU, Newfoundland and Labrador would be selling Québec the underlying power at 3.8¢.
The historical 0.2¢ figure is an export price under a specific contract. It is not a rate available to any customer today, and we never present it as one. Commercial power on the island runs roughly 11–13¢/kWh delivered. This brief is a value-retention argument, not a grievance argument.
| When | What happened |
|---|---|
| Dec 2024 | Province and Hydro-Québec announce an MOU to replace Churchill Falls terms and enable expansions — framed as a 2025–2075 arrangement. |
| May 19, 2026 | Independent Churchill River Review Committee reports that the MOU as configured is not in the overall best long-term interest of the province — flagging limits on using the power for in-province economic development, transmission access, pricing, and governance. |
| May–Jun 2026 | Renegotiation team and oversight committee appointed. The Premier names more power for Labrador's economy as a priority — framed around critical minerals and industry. AI and compute are not named. |
| Jul 2026 | The definitive agreement is still unsigned. The April target was missed. The window to write a compute-eligible block into the architecture is open, and it will not stay open. |
The Review also recommended new transmission from Churchill Falls to Labrador West specifically to enable energy-intensive industry. The institutional appetite for in-province industrial load already exists. What is missing is a named, eligible use class that includes compute.
The International Energy Agency projects global data-centre electricity consumption to grow from 415 TWh in 2024 to about 945 TWh by 2030. Power availability — not latency, not tax incentives — has become the primary site-selection criterion. Interconnection queues in the established hubs now run for years.
| Jurisdiction | 2025–26 posture |
|---|---|
| Québec | Ministerial approval required for large new loads; proposed dedicated data-centre tariff of 13¢/kWh (Feb 2026, before the regulator). |
| British Columbia | Crypto loads barred; competitive AI power call capped in the range of 300 MW. |
| Ontario | Economic screening introduced for large load connections. |
| Alberta | An interim connection cap on the order of 1,200 MW against multi-gigawatt request queues. |
| Newfoundland & Labrador | No policy. No tariff. No allocation process. No project. |
Remote, cold and hydro-rich is precisely the profile of the current build-out. Stargate Norway, announced August 2025, placed a 230 MW campus scaling to 520 MW and roughly 100,000 GPUs in Narvik — above the Arctic Circle — as a joint venture between Nscale and Aker, running on hydropower, chosen for surplus clean power and cold climate. Meta has operated a hydro-powered campus in Luleå, Sweden since 2013. Google is expanding in Hamina, Finland.
Two technical facts make this work. AI training is latency-insensitive — training runs tolerate around 100 ms between regions, because no end user is waiting. Latency-sensitive inference stays near cities; capital-intensive training goes to the power. And cold helps, but power decides: free cooling for eight to ten months a year is a real efficiency edge, but liquid cooling is narrowing it. The durable advantage is firm, clean, uncommitted power. On that measure Labrador's position is close to unique in eastern North America.
Canada's Sovereign AI Compute Strategy commits more than $2 billion, with further Budget 2025 funding, and explicitly seeks Canadian-owned, Canadian-controlled projects above 100 MW with Indigenous participation. A Labrador consortium — Crown power, Innu equity, an NL developer, Canadian anchor demand — scores on every one of those criteria. The first intake window closed in 2026 with no Newfoundland and Labrador entrant, because there was no NL project to enter.
The province is not being asked to spend, build, or subsidize. It is being asked to write five instruments into the post-agreement policy architecture.
A scheduled allocation of Labrador capacity eligible for AI and compute loads. An initial tranche in the 100–150 MW range as new generation phases in, with a roadmap to 500+ MW as expansion proceeds and consent allows.
A published firm industrial compute rate — indicatively in the 4–6¢/kWh range plus full network-addition cost causation. High enough to be unimpeachable after Muskrat Falls; low enough to beat Québec's proposed 13¢.
Score for NL majority ownership, Indigenous equity participation, creditworthy offtake, local employment, and balance-sheet substance. These criteria — not a ban — are the durable answer to the Great North Data experience.
Direct NL Hydro to publish Labrador dark-fibre capacity and recall-block headroom, and commission a diverse-route fibre feasibility study — co-fundable with federal connectivity programs.
A joint NL–Canada expression of interest under the Sovereign AI Compute Strategy, so the next intake window opens with a Labrador consortium already on file.
Item 1 belongs in the definitive agreement's architecture or its enabling provincial framework. Items 2 through 5 are ordinary policy instruments the province can move this year.
The structure that works — demonstrated at Narvik, consistent with the Independent Review's governance critique, and politically viable after Muskrat Falls — puts each risk with the party built to carry it.
Capital does not create power; power attracts capital. Québec, British Columbia, Ontario and Alberta have each just demonstrated which direction that queue runs.
A case built on cheerleading will fail in this province, and it should. Five constraints are real. None of them is a reason to leave the block unwritten — because a block that is never reserved is a permanent decision in favour of exporting raw electrons.
Labrador's terrestrial backbone is essentially a single route of roughly 48 fibres built in 2012. Region-wide outages from single events are on record, and a subsidized Labrador broadband build was abandoned in May 2025 after a large cost overrun. Hyperscale campuses want multiple physically diverse paths; Labrador has approximately one.
The Churchill expansion program will itself absorb the province's construction trades at camp premiums. Muskrat Falls remains the controlling cost precedent for remote Labrador construction.
Ratification of the Innu Nation–Hydro-Québec settlement remains unresolved. Geotechnical work at Gull Island was halted by blockade in July 2025. Consultation at Happy Valley-Goose Bay may involve three overlapping claims.
The province's one prior data-centre experience was a Labrador crypto operator that went bankrupt in 2019 owing money to the utility and public agencies. Municipal skepticism toward container mining is earned.
Lease cancellations in 2025 and a credible skeptic corpus warn of overbuild. If capex contracts later this decade, marginal sites die first.
Open People is a Newfoundland and Labrador AI company developing sovereign, self-hosted computing — software built to run without resupply, on infrastructure its users actually control. Our role in this campaign is deliberately narrow and capital-light: we assemble demand, we build the sovereignty software layer, and we make the public case. We are not asking to own the steel.
Tom Lane is an entrepreneur based in St. John's. He is the founder of Open People and co-founder and owner of snōw white laundry, a seasonal restaurant on Water Street that opened in 2026. He took that project from lease negotiation through design, permitting and construction — acting as his own project manager with a local general contractor — and built its operating stack himself: supplier price intelligence, demand forecasting, financial automation, and a custom staff platform.
It is a small building with a large point behind it. A regulated, multi-permit, multi-trade physical build delivered on a compressed schedule is the same job shape as a first small compute node — site control, power and mechanical, inspectors, trades, and a licence gate at the end. He runs a 70-seat restaurant with the instrumentation of a data centre. This campaign is that continuity, applied to a bigger load.
This document is meant to be argued with. The following are open questions, not settled claims, and we say so in every room:
Every figure in this brief maps to a public primary or reported source. Where a number is an estimate, a range, or an illustration rather than a forecast, it is labelled as one. If something here is wrong, we want to know — corrections change the document.
Prepared independently. All figures as publicly reported. This document is intended to be falsifiable — a full claim-by-claim evidence table with confidence ratings sits behind it and is available on request.
If you work on energy policy, Labrador development, Indigenous partnership, compute demand, or infrastructure capital in this province — we would like to talk.
tom@openpeople.ai