A proposed Canadian sovereign AI infrastructure concept
AI is dying of thirst.
Canada isn’t.
Desert data centres were sited for cheap land — now they queue for water. We propose building AI infrastructure where the power is hydro, the water is abundant, and the air is cold enough to cool the machines for free.
The advantage, in numbers
Design targets against published benchmarks — and we label which is which. Nothing here is an operating result: no facility exists yet.
Design-target water use, Canadian Vault sites
0.2 L/kWh
US typical 1.8 L/kWh
Grid intensity, Québec — the lead proposed region
1.2 gCO₂e/kWh
US East reference ~270 gCO₂/kWh
Design-target PUE
1.20
US fleet average 2024 1.55
Free-air cooling days at northern sites (design target)
280+ days/yr
Three structural advantages — siting is the strategy
1. Hydro power, stable price
Provincial hydro grids in Québec, Manitoba, British Columbia and Newfoundland sell among the most stable-priced clean power in North America. No gas-price exposure, no heat-derated transformers, gigawatts of headroom.
2. Water is not the enemy here
Our proposed sites sit beside continental-scale freshwater: the St. Lawrence estuary, the Mackenzie, the Churchill–Nelson system. Cooling draw that causes a crisis in a desert watershed is a rounding error beside a river system of this scale — and our design targets closed-loop anyway.
3. Cold air does the work
Northern sites target free-air cooling most of the year. Cold climates also make high-density GPU halls tractable: the harder problem in Arizona — rejecting heat — is the easy one here.
Water per kilowatt-hour: the comparison that matters
Litres of on-site water per kWh delivered — our design target versus typical US on-site use. This is the figure the industry is being forced to disclose.
On-site water use (design target vs published practice)
Table: Canadian Vault target 0.2 L/kWh (design target, not measured); US typical 1.5–1.8 L/kWh (published benchmark).
Thirteen proposed sites — nine on named river systems
Every site is conceptual: land, permits, power agreements and funding are all still to be earned. We publish them because the geography is the thesis.
Baie‑Comeau, QC
Proposed- Power
- Grid (Hydro‑Québec)
- Capacity
- 3→25 MW
- PUE target
- 1.25
- Water
- St. Lawrence Estuary
Churchill, MB
Proposed- Power
- Grid (Manitoba Hydro)
- Capacity
- 3→20 MW
- PUE target
- 1.25
- Water
- Churchill River estuary
Labrador City / Wabush, NL
Proposed- Power
- Grid (Churchill Falls)
- Capacity
- 2→15 MW
- PUE target
- 1.25
- Water
- Smallwood Reservoir nearby
Sovereignty, for what it’s worth
Canadian land, Canadian law, Canadian power
Compute that never leaves Canadian jurisdiction: PIPEDA, Québec’s Law 25, OSFI B-13 and Bill C-27 shaped into the design from day one. Water, power and sovereignty are the same argument: the resources AI needs most are the ones Canada has most.
Help us prove the concept.
We are raising a $50K pre-seed to take one site from geography to feasibility. Every figure on this site is a target — fund the work that turns targets into measurements.