Infrastructure Research · Data vintage September 2026

AI Data Center Economics in Canada: Build Models, Power Costs & Site Selection

Updated September 8, 2026 14 min read Strategic Crypto Reserve Blockchain Infrastructure

An AI data center is not one asset class. Seven distinct build models compete for the same megawatts in Canada, and their capital structures differ by a factor of four per megawatt of IT load. This page sets out how those models price, what provincial power actually costs, how long interconnection takes in each province, and where the modelling assumptions behind the SCR AI Build Configurator come from.


Run the Configurator The Seven Build Models
AI data center economics in Canada — build models, power costs and interconnection timelines from Strategic Crypto Reserve

Screening Model

7 Builds · 9 Provinces · CAD

What Strategic Crypto Reserve Actually Operates

 Plain disclosure, up front

Strategic Crypto Reserve does not own or operate an AI data centre. Our physical footprint is a single low-power SHA-256 Bitcoin miner drawing roughly 40 watts and producing about 5 TH/s, housed in an 8×8 foot outbuilding serviced for approximately 3 kW on the BC Hydro grid in the Comox Valley, British Columbia. That is a research bench and a hydro-rate test case. It is not a facility, and we will not describe it as one.

What we do operate at commercial scale is research and software: a published body of infrastructure analysis covering Canadian power markets, mining economics and data centre build structures, and a licensed modelling product used by site selection, brokerage and development professionals. If you arrived here expecting a colocation vendor, you are in the wrong place. If you arrived here wanting to understand how these projects price before you commit capital or a client relationship, keep reading.

The reason for stating that so bluntly is that this sector is saturated with pages describing GPU clusters and campus substations that do not exist. Those pages waste the time of the people this research is written for, and they make the honest numbers harder to trust. The figures below are indicative planning constants, dated by vintage, sourced from published utility tariffs, interconnection guidance and observed large-load precedent. Where we are uncertain, we say so.

Every number on this page is also the number inside the SCR AI Build Configurator. The page and the tool do not disagree, because they share one dataset maintained in one place.

Small-scale hydro-powered Bitcoin mining research bench operated by Strategic Crypto Reserve in the Comox Valley, British Columbia

Seven Build Models, Not One Asset Class

The single most common analytical error in Canadian AI infrastructure conversations is treating "AI data centre" as one thing. A powered shell developer, a neocloud operator and a mining company converting a hall are running three different businesses that happen to share a transformer. They carry different capital structures, different revenue mechanisms, different staffing loads, different build durations and completely different risk profiles.

Below are the seven configurations the SCR model prices. The distinctions matter because a per-megawatt figure quoted without a model attached is meaningless.

1. Hyperscale AI Factory

Purpose-built campus, dedicated substation, direct liquid cooling from day one, densities to 200 kW per rack. Power-first site selection — the interconnect date beats every other variable. Roughly 34 months to build.

100 MW – 1 GW+

2. Neocloud / GPU-as-a-Service

You own the silicon and sell the hours. Highest revenue per megawatt of any model and the sharpest depreciation risk attached to it. Utilisation is the entire business. Roughly 16 months to build.

5 – 300 MW

3. Powered Shell

Deliver energised, cooled, fibered space; the tenant brings the accelerators. Lowest technical risk, cleanest contracted cash flow, capped margin. Power is usually passed through to the tenant. Roughly 26 months.

10 – 500 MW

4. Bitcoin Miner Retrofit

Energised power is the head start, not the finish line. Mining halls tolerate dust, heat and single feeds; AI tenants require none of that. Capital lands in fit-out and networking. Roughly 18 months.

20 MW – 2.5 GW

5. Modular / Prefabricated

Factory-built power blocks and containerised white space, typically rear-door heat exchanger cooling. A per-megawatt premium bought in exchange for months instead of years. Roughly 9 months per phase.

0.5 – 50 MW per phase

6. Edge / Micro Deployment

Sited for latency, data sovereignty or operational control rather than scale. An inference profile, not a training one. Frequently justified by compliance rather than cost. Roughly 6 months.

20 kW – 2 MW

7. Hybrid Mining + AI

Hashing monetises megawatts while AI capacity is built, then yields power to the higher-value tenant. Two businesses sharing one interconnect and one balance sheet. Roughly 14 months.

5 – 500 MW

 Model any of the seven against your own region

The SCR AI Build Configurator lets you set IT load, rack density, cooling architecture and contract rate, then returns PUE-adjusted facility load, CapEx, annual energy cost, EBITDA and simple payback in Canadian dollars.

Open the Configurator

Where the Capital Actually Goes

The table below sets out indicative all-in build cost per megawatt of IT load, in Canadian dollars, excluding accelerator procurement. "Shell" covers structure, electrical distribution and mechanical plant. "Fit-out" covers white space, racks, cooling terminations and commissioning. Land, utility contribution and network are separate line items because they behave differently by site and by province.

Build model Shell
(CAD M / MW)		 Fit-out
(CAD M / MW)		 Land + utility + network
(CAD M / MW)		 Indicative total
(CAD M / MW)
Hyperscale AI factory20.04.51.3525.85
Edge / micro deployment22.03.00.8525.85
Modular / prefabricated11.02.00.8513.85
Neocloud / GPU-as-a-service9.04.01.1014.10
Powered shell8.51.00.9510.45
Bitcoin miner retrofit6.02.20.608.80
Hybrid mining + AI5.51.20.657.35

Three observations that matter more than the absolute figures.

Edge deployments are expensive per megawatt and that is not a defect. A 500 kW edge site carries the same design, permitting and commissioning overhead as a much larger one, spread across a fraction of the load. Edge projects are almost never justified on cost per megawatt; they are justified on latency, sovereignty or a compliance requirement that no central facility can satisfy. If an edge business case is being argued on unit economics, it is being argued wrongly.

Retrofits look cheap in the shell column and are not cheap overall. The shell figure is low because the structure and the interconnect already exist. The fit-out and network figures carry the real work, and in practice a retrofit's largest schedule risk is not construction but the discovery that the existing electrical topology cannot support redundant paths without substantial rework.

Accelerator procurement dwarfs all of this where the operator buys the silicon. In the neocloud and hyperscale models, the hardware line is typically the dominant capital item and depreciates on a schedule that does not care whether the building is finished. That timing mismatch — capital committed to silicon against a facility that cannot energise for two years — is the single most common way an otherwise sound AI infrastructure model fails.

Operating cost tracks capital intensity but not proportionally. Annual OpEx typically runs between 1.8% of CapEx for a powered shell and 4.0% for an edge deployment, with staffing ranging from roughly CAD 28,000 per megawatt annually for a shell to CAD 90,000 per megawatt for edge sites where headcount cannot be spread across load. For a line-item treatment of the mining-side equivalent, see our 300-unit crypto mining facility cost breakdown.

Provincial Power: Rate, Grid and Time to Energisation

Electricity price is the number everyone asks for first and the number that decides the fewest projects. The variable that actually decides sites in Canada right now is the interconnection queue — the indicative time from a signed service agreement to energisation. The spread across the country is roughly 22 months, which is long enough to invalidate an entire pro forma.

Province Utility / market Indicative blended rate
(CAD / kWh)		 Indicative queue
(months)		 Grid character
ManitobaManitoba Hydro0.05126Hydroelectric
QuebecHydro-Québec0.05536Hydroelectric
British ColumbiaBC Hydro0.06230Hydroelectric, >90% clean
Newfoundland & LabradorNL Hydro0.06828Hydroelectric
AlbertaDeregulated (AESO market)0.07814Gas-dominant, behind-the-meter permitted
New BrunswickNB Power0.08120Nuclear, hydro, thermal
SaskatchewanSaskPower0.08424Mixed thermal and renewable
OntarioIESO / local distributors0.09830Nuclear, hydro, gas
Nova ScotiaNova Scotia Power0.11524Coal transition, growing wind

The constraint behind each number

British Columbia. A competitive allocation process launched in January 2026. Data centre projects now compete for capacity rather than receiving it on request, and capacity is protected for mining, LNG and forestry loads. The rate is excellent and the power is genuinely clean; the question is whether a given project wins an allocation at all. Our BC Hydro electricity rate analysis covers the tariff structure in detail, and power grid strategies for Vancouver-area operators covers the practical service application path.

Alberta. The fastest interconnection path in Canada at roughly 14 months, and behind-the-meter generation is viable, which removes the queue from the critical path entirely for operators willing to build their own supply. Carbon intensity is the trade-off, and it is a real one for tenants with procurement standards tied to emissions reporting.

Quebec. Cheap, clean and heavily spoken for. New large-load blocks are allocated selectively through government-directed processes. A 36-month indicative queue combined with allocation risk makes Quebec a strong site for projects that can wait and a poor one for projects that cannot.

Manitoba. Among the lowest industrial rates in North America. The binding constraint is capacity headroom and transmission access rather than price, which is a materially different problem from an expensive grid and requires a different diligence approach.

Ontario. Global Adjustment mechanics dominate large-load economics to the point that the posted rate is close to uninformative. Class A participation materially changes the effective rate, and any Ontario model built on a headline number rather than a Class A analysis will be wrong by a wide margin.

Saskatchewan, New Brunswick, Nova Scotia and Newfoundland & Labrador. Smaller grids where an individual project can be material to system planning. New Brunswick is actively pursuing data centre load. Saskatchewan has limited large-load precedent, so expect bespoke negotiation and utility contribution requirements. Nova Scotia's higher rates and tighter capacity generally suit edge and modular scale rather than campus builds. Labrador holds substantial hydro capacity, with transmission and fibre as the practical limiters rather than generation.

Rates are indicative, blended and dated. They represent a planning-grade estimate of what a large continuous load pays all-in, not a posted tariff line, and they carry a September 2026 vintage. No figure on this page is a commitment from any utility. Confirm your applicable rate schedule and service timeline directly with the relevant utility before committing capital.

Density, Cooling and Why PUE Changes the Answer

Rack density sets the cooling architecture, the cooling architecture sets the PUE, and PUE sets how much grid power you must buy to deliver a given amount of IT load. That chain is the reason two projects with identical IT capacity can carry meaningfully different energy bills and different interconnection requirements.

Air cooling remains viable to roughly 40 kW per rack in a well-designed hot-aisle containment. Rear-door heat exchangers extend that envelope into the 40 to 80 kW range with a moderate capital premium and a modest PUE improvement. Direct liquid cooling is effectively mandatory above roughly 80 kW per rack, which is where current high-density accelerator deployments sit, and it delivers the best PUE of the three while adding fluid handling, leak detection and a maintenance discipline that many operators are meeting for the first time.

The practical consequence for site selection is that a density decision made in the design phase propagates directly into the megawatts requested from the utility. A facility designed at 60 kW per rack and one designed at 120 kW per rack supporting the same accelerator count request very different service capacity, and in a province where capacity is allocated rather than sold, that difference can determine whether the project is approved at all.

PUE also decides how much of your energy spend is doing useful work. At a PUE of 1.5, a third of every dollar of electricity is going to cooling and distribution losses rather than compute. At 1.15, that share falls below one seventh. Across a 50 MW facility over a fifteen-year hold, in a province at CAD 0.078 per kWh, that gap is a material line in the model rather than an engineering footnote.

Cooling architecture and rack density considerations for AI data centre design

The Miner-to-AI Conversion, Honestly Priced

Bitcoin mining companies hold something genuinely scarce: energised interconnects at scale, secured years before the current demand cycle. That is a real advantage and it is why the retrofit model exists. It is also routinely overstated.

A mining hall is engineered around an assumption that a failed hash costs almost nothing. Dust ingress is tolerated. Temperature swings are tolerated. Single utility feeds and minimal redundancy are normal. Networking is sized for a trickle of pool traffic rather than for east-west accelerator communication. None of that is a defect in a mining context; all of it is disqualifying in an AI hosting context.

Converting therefore means conditioned space, redundant power paths, carrier-grade fibre, structured cabling, and in most cases a cooling architecture the site was never designed to accommodate. Our model prices retrofit shell at CAD 6.0 million per megawatt against CAD 8.5 million for a purpose-built powered shell — a genuine saving — but adds CAD 2.2 million per megawatt of fit-out against the shell model's CAD 1.0 million, plus a substantially higher network allocation. The saving is real and it is smaller than the pitch decks suggest.

The hybrid model exists for operators unwilling to choose. Hashing continues to monetise megawatts while AI capacity is built out, and power yields progressively to the higher-value tenant as conversion completes. The model prices a blended revenue stream with an adjustable AI share, defaulting to 55%. It carries the lowest capital intensity of the seven at roughly CAD 7.35 million per megawatt, and the operational complexity of running two businesses on one interconnect. For background on the mining side of that equation, see Bitcoin mining on Vancouver Island and our crypto data centre overview.

Methodology and Limitations

Figures on this page and in the configurator are derived from published utility tariffs and rate schedules, publicly available interconnection guidance, and observed large-load precedent in Canadian markets, blended into indicative planning constants and stamped with a vintage date. The current vintage is September 2026.

What this model is

A screening tool. It exists to answer, in minutes rather than weeks, whether a given combination of province, build model, scale and contract rate is worth taking to a feasibility study. It is designed to be wrong in ways that are visible and directional rather than precise and misleading.

What this model is not

It is not a substitute for a utility service agreement, an engineering estimate, a geotechnical assessment or a tax and incentive analysis. Provincial and municipal incentives, water availability, fibre routes, seismic classification, local permitting posture and property tax treatment all move real project economics and none of them are in this model. A screening output that looks good should trigger diligence, not a commitment.

Known limitations we would rather state than hide

Blended rates compress a great deal of complexity — demand charges, time-of-use structures, Ontario's Global Adjustment, negotiated industrial arrangements — into a single figure. Queue estimates are indicative and vary substantially by specific interconnection point, not merely by province. Accelerator pricing and revenue-per-hour assumptions move faster than any published dataset can track and should always be overridden with your own figures where you have them. The Enterprise licence tier exists specifically so that operators can replace our cost constants with theirs.

If you find a figure that disagrees with what you know from direct experience, we would rather hear about it than defend it. Corrections go to support@strategiccryptoreserve.ca and are folded into the next dataset vintage, which propagates to every licensed deployment on the next page load.

 License the configurator for your own site

One script tag, isolated shadow DOM, zero dependencies. Standard CAD 4,000/year for three lease-model profiles; Pro CAD 10,000/year for all seven profiles with CSV export and white-label branding; Enterprise quoted with cost-constant overrides and multi-domain rights. Invoiced directly — no revenue share, no per-lead fee, no metering.

Licensing & Demo

Frequently Asked Questions: AI Data Center Economics

1. Does Strategic Crypto Reserve operate an AI data center?

No. Strategic Crypto Reserve operates a single low-power SHA-256 Bitcoin miner drawing roughly 40 watts, housed in an 8×8 foot outbuilding serviced for approximately 3 kW on the BC Hydro grid in the Comox Valley, British Columbia. That is a research bench, not a data centre, and we describe it that way deliberately. What Strategic Crypto Reserve does operate at commercial scale is a published body of infrastructure research and a licensed modelling product — the SCR AI Build Configurator — used by site selection, brokerage and development professionals to screen Canadian AI data centre projects. This page documents the methodology behind that model rather than marketing a facility we do not own.

2. What are the seven AI data center build models?

The seven models are: hyperscale AI factory (100 MW to 1 GW and above, purpose-built campus with a dedicated substation); neocloud or GPU-as-a-service (5 to 300 MW, operator owns the accelerators and sells compute hours); powered shell (10 to 500 MW, operator delivers energised cooled space and the tenant brings the hardware); Bitcoin miner retrofit (20 MW to 2.5 GW, converting existing energised mining halls); modular or prefabricated (0.5 to 50 MW per phase, factory-built power blocks); edge or micro deployment (20 kW to 2 MW, sited for latency or data sovereignty); and hybrid mining plus AI (5 to 500 MW, hashing monetises megawatts while AI capacity is built out). Each carries a distinct capital structure, revenue mechanism, staffing load and build duration, which is why treating them as one asset class produces misleading numbers.

3. How much does an AI data center cost per megawatt in Canada?

Excluding accelerators, indicative all-in build cost ranges from roughly CAD 7.4 million to CAD 25.9 million per megawatt of IT load depending on the model. A hybrid mining and AI conversion sits at the low end near CAD 7.35 million per MW, a Bitcoin miner retrofit near CAD 8.80 million, a powered shell near CAD 10.45 million, a modular deployment near CAD 13.85 million, a neocloud build near CAD 14.10 million, and both a hyperscale AI factory and an edge deployment near CAD 25.85 million per MW — the first because campus-grade electrical and mechanical plant is genuinely expensive, the second because fixed design and commissioning costs cannot be amortised across scale. Accelerator procurement, where the operator buys the silicon, is a separate and usually larger line item. These are screening figures, not engineering estimates.

4. Which Canadian province is cheapest for an AI data center?

On indicative blended large-load industrial rate alone, Manitoba is cheapest at approximately CAD 0.051 per kWh, followed by Quebec at 0.055 and British Columbia at 0.062. Nova Scotia is the most expensive at approximately 0.115. Rate alone is the wrong basis for a decision, however: Manitoba's constraint is capacity headroom and transmission rather than price, and both Quebec and British Columbia now allocate new large-load blocks through selective, government-directed or competitive processes rather than first-come service. Alberta carries a higher rate at approximately 0.078 per kWh but offers the fastest interconnection in the country at roughly 14 months and permits behind-the-meter generation, with carbon intensity as the trade-off for enterprise procurement.

5. How long does it take to get power for a data center in Canada?

Indicative time from a signed service agreement to energisation ranges from about 14 months in Alberta to about 36 months in Quebec, a spread of roughly 22 months. New Brunswick runs near 20 months, Saskatchewan and Nova Scotia near 24, Manitoba near 26, Newfoundland and Labrador near 28, and British Columbia and Ontario near 30. These are indicative planning figures for large loads, not commitments from any utility. The interconnection date frequently dominates every other variable in an AI data centre pro forma, because a project that cannot energise cannot earn, and accelerator depreciation runs whether or not the building has power.

6. Why are Bitcoin mining retrofits harder than they look?

An energised mining site is a genuine head start because the interconnect already exists, but energised power is not a finished AI facility. Mining halls tolerate dust, wide temperature swings, single utility feeds, minimal redundancy and low-grade networking, because a failed hash costs almost nothing. AI tenants tolerate none of that: they require conditioned space, redundant power paths, high-quality fibre, and increasingly direct liquid cooling to support rack densities above roughly 40 kW. Retrofit capital therefore lands mostly in fit-out, network and cooling rather than in shell construction, and the honest version of the conversation prices those requirements before a tenant is approached rather than after.

7. What is the SCR AI Build Configurator?

The SCR AI Build Configurator is a licensed JavaScript widget that embeds on a third-party website with a single script tag and models all seven build configurations against nine Canadian provincial power datasets, returning PUE-adjusted facility load, CapEx by line item, annual energy cost, EBITDA, simple payback and implied asset value in Canadian dollars. It runs in an isolated shadow DOM with no dependencies, so it cannot inherit or break the host site's styling. Licences are annual and invoiced directly: Standard at CAD 4,000 per year unlocks three lease-model profiles, Pro at CAD 10,000 per year unlocks all seven profiles with CSV export and white-label branding, and Enterprise is quoted and adds cost-constant overrides, attribution removal and multi-domain rights.