Arctic Edge

Cold-Climate,
Renewable-Anchored
Canadian Data Centers

Small, factory-made data centers in cold Canadian towns with cheap, clean power.

Press → to advance · Esc for overview

Why now?

  • AI demand for GPU compute is outrunning supply.
  • Buyers increasingly require Canadian data residency.
  • There is demand for a credible clean-energy story.
  • Canada's edge: free cooling, cheap hydro (≈US$0.04/kWh), and a 30% federal ITC on solar+BESS.
  • Waste heat reuse proven in Canada — Equinix TR5/Markham heats 14M sq ft.
  • Small + modular sidesteps multi-year permitting, grid queues, & new DC surcharges.

The Core Thesis

A repeatable module that is three businesses at once.

Compute

Colocation, high-density AI hosting, Canadian residency.

Energy & Heat

Demand response, storage, waste-heat to local thermal sinks.

Product

Once proven, sell, operate, or license the template itself.

Scale by adding identical modules, rather than building custom mega-facilities.

Business Model Layers

  1. Compute — Colocation / GPU hosting / Managed compute.
  2. Energy & Heat — Demand response, storage integration, heat offtake.
  3. The Module as a Product — Sell the box or license the design/orchestration software.
Request full brief →

The Module (Mark I, ≈1 MW)

  • ≈1 MW IT: Kept under 5 MW total to dodge large-load regulatory/rate thresholds.
  • Prefabricated skid/container blocks; commissioned rapidly on site.
  • High-density, liquid-cooled AI racks (GB300 NVL72 today, Vera Rubin NVL72 in H2 2026) + heat-recovery skid.
  • On-site solar + battery; tuned to each site's local energy profile.
Request module reference design →

What you're buying

Reference build: 5× GB300 NVL72 racks (360 GPUs), ≈810 kW IT

The raw numbers

  • 5.4 exaflops FP4 dense — more than the #1 supercomputer had in 2023
  • 100 TB unified HBM3e memory
  • Each rack is a single NVLink domain — 72 GPUs at 1.8 TB/s, no network bottleneck
  • Trillion-parameter models fit in one rack's memory

What it does

  • Serves ≈19,000 concurrent AI users (DeepSeek-R1, 2.9M tok/sec)
  • Runs Llama 405B at production scale (93K tok/sec)
  • Trains a 70B model in days, not weeks
  • Replaces US$13–19M/yr of neocloud GPU rental

Vera Rubin (H2 2026): 4 racks deliver 14.4 EFLOPS — 2.7× more compute in fewer cabinets.

Does it pencil?

Stabilized year. All figures USD (≈0.73× CAD). Includes founder salary & corporate overhead.

1 Module2 Modules3 Modules
Revenue (USD)$2.4M$4.8M$7.2M
All-in opex (USD)($1.2M)($2.1M)($2.9M)
EBITDA (USD)$1.2M$2.8M$4.3M
EBITDA margin52%57%60%
Net income (USD)≈$0.07M≈$0.4M≈$0.7M
Equity required (USD)$3.5M$7.0M$10.5M

Module 1 proves the model. Module 2 pays you and generates real returns. 30% federal ITC on solar+BESS (≈US$330K/module) not included — pure upside.

Request financials →

Where to build

Win condition = Lock three contracts: power, anchor customer, heat off-taker.

  • Portage la Prairie, MB — Safest beachhead: cheapest power, no DC surcharge, agri-food heat sinks.
  • Saint-Félicien, QC — Cleanest proof-of-model; stay ≤5 MW to avoid proposed ≈US$0.095/kWh DC rate.
  • Tier 2: Prince George & Kamloops (BC, competitive allocation); Medicine Hat (AB, self-supply build).

New since Dec 2025: AB mandates self-supply; ON requires ministerial approval >5 MW; QC proposes 2× power rate for DCs. Our ≤5 MW modular design avoids most of these.

Request town shortlist →

The competition

Incumbents own the metal; we win on siting + heat + orchestration.

Schneider

#1 prefab/modular (≈20% share). EcoStruxure pods.

Request profile →

Vertiv

#1 thermal & large UPS. Strong in AI liquid cooling.

Request profile →

Costs to start

All figures USD.

StageWhatCost (USD)Funding
1. ValidateFeasibility, design, LOIs≈US$110–290KPre-seed
2. Build #1First ≈1 MW module≈US$8.75M60% debt / 40% eq
3. ReplicateModules 2…N≈US$8.75M ea.Debt + cash flow
Request the full pitch →

The first move

A 4–6 month feasibility sprint on two towns.

Three yeses → a fundable ≈US$8.75M project. If not, we've spent a fraction to learn.

Detailed reports shared on request: Request brief · Request financials · Request towns

Start a conversation

Operators, tenants, towns — tell us what you want to explore.

Full financials and town briefs are shared on request after we talk.