Accountability
Every number, and where it came from
This site mixes three kinds of numbers, and the label on each tile tells you which: our own design targets, published benchmarks about the comparator world, and published projections. Anything we cannot source, we cut. Figures marked to-verify are scheduled for source verification before our public launch.
| Figure | Value | Type | Source | Accessed | Caveats |
|---|---|---|---|---|---|
| Design-target water use, Canadian Vault sites | 0.2 L/kWh | Design target | Internal design target | — | An engineering goal for closed-loop cooling at hydro-adjacent sites — not a measurement. No facility has been built. |
| US average on-site water use per kWh | 1.8 L/kWh | Published benchmark | Li, Yang, Islam & Ren (UC Riverside) — Communications of the ACM / arXiv:2304.03271 | 2026-09-10 | Real facilities span 1–9 L/kWh with climate and technology (an Arizona commercial hall hits 9 in summer); hyperscaler fleet averages run 0.5–1.0 (Google ≈1.0, Microsoft ≈0.55), Meta reported 3.7 in 2023. |
| Projected global AI water withdrawal in 2027 | 4.2–6.6 billion m³ | Published projection | Li, Yang, Islam & Ren (UC Riverside) — Communications of the ACM / arXiv:2304.03271 | 2026-09-10 | 4.2–6.6 billion cubic metres — 4–6× the annual water withdrawal of Denmark. |
| Freshwater evaporated training GPT-3 in US data centres | 700,000 L | Published projection | Li, Yang, Islam & Ren (UC Riverside) — Communications of the ACM / arXiv:2304.03271 | 2026-09-10 | — |
| US data centres’ direct (on-site) water use in 2023 | ~17 billion gal | Published benchmark | Lawrence Berkeley National Laboratory, for the U.S. Department of Energy | 2026-09-10 | Hyperscale facilities alone are projected to reach 16–33 billion gallons per year by 2028. |
| Projected US hyperscale data-centre water use by 2028 | 16–33 billion gal/yr | Published projection | Lawrence Berkeley National Laboratory, for the U.S. Department of Energy | 2026-09-10 | — |
| US data centres drawing cooling water from stressed watersheds | ~1 in 5 | Published benchmark | Environmental Research Letters | 2026-09-10 | Nearly half of US servers rely on electricity from water-stressed regions. |
| Water consumed by Virginia data centres in 2023 (+86% since 2019) | 2.1 billion gal | Published benchmark | JLARC, Commonwealth of Virginia | 2026-09-10 | — |
| The Oregonian’s public-records suit forced Google to stop treating The Dalles water use as a trade secret | 2022 | Published benchmark | The Oregonian / Wikipedia | 2026-09-10 | — |
| US data-centre projects blocked or delayed by local opposition, 2024–2025 | $64B | Published benchmark | Data Center Watch | 2026-09-10 | — |
| Grid intensity, Québec — the lead proposed region | 1.2 gCO₂e/kWh | Published benchmark | Canada Energy Regulator | 2026-09-11 | Corrected 2026-09-11: our earlier materials used a conservative internal estimate of 38. Remote northern sites (NWT/NU) target hydro-adjacent or wind+storage hybrids — per-site design targets, higher than Québec’s grid. |
| Grid intensity, Virginia data-centre corridor | ~270 gCO₂/kWh | Published benchmark | US Environmental Protection Agency, eGRID | 2026-09-10 | Corrected 2026-09-10: our earlier materials cited an older ~420 figure that overstated the comparator; eGRID 2023 is the current federal dataset. |
| Hydro share of the proposed energy mix | 80 % | Design target | Internal design target | — | Design assumption: provincial hydro grids plus on-site renewables and storage where sites are remote. |
| Design-target PUE | 1.20 | Design target | Internal design target | — | Power Usage Effectiveness — the ratio of total facility energy to IT energy. 1.20 is an aggressive cold-climate goal, not an operating result. |
| US fleet-average PUE (2024) | 1.55 | Published benchmark | Lawrence Berkeley National Laboratory, for the U.S. Department of Energy | 2026-09-10 | Our 1.20 target is conservative against purpose-built AI halls (~1.15 today) but robust against the fleet average — and unlike the fleet average it is designed for a water-scarcity world. |
| Free-air cooling days at northern sites (design target) | 280+ days/yr | Design target | Environment and Climate Change Canada (normals table as published for Baie-Comeau) | 2026-09-11 | Derivation: outside-air cooling is available whenever daily mean sits below ~18 °C. At Baie-Comeau even July averages 16 °C, so the physical ceiling is near year-round; 280+ days is our deliberately conservative engineering target (equipment margins, heat-rejection days). |
| Proposed locations under evaluation | 13 | Design target | Internal design target | — | All sites are conceptual. None has land, permits, power agreements, or funding. |
| Proposed initial → target capacity per site | 1–3 → 25 MW | Design target | Internal design target | — | — |
| Pre-seed round sought (concept validation) | $50K | Design target | Internal design target | — | — |
What our targets exclude
- Embodied carbon of construction and equipment.
- Upstream energy generation impacts outside the operating boundary.
- Any customer-side efficiency claims: we measure our halls, not your models.
The illustrative US comparators (grid carbon, PUE, water) trace to public references compiled in our 2025 impact worksheet (EIA, NRCan provincial grid intensities) and are scheduled for re-verification against the latest LBNL, eGRID and Uptime Institute editions before public launch.