A credible data-center estimate starts with a declared basis, not a headline price per megawatt. Define the commissioned IT load, redundancy, utilization, location, schedule and inclusions; build capital costs from quantities; then model electricity and every recurring facility cost. Report both total cost and cost per commissioned IT MW, with the scope attached to each number.
The method below gives you a repeatable estimate for a new facility, expansion or leased shell and shows how PUE, tariffs, interconnection work, schedule risk and high-density computing change the result.
As an Amazon Associate I earn from qualifying purchases.
What a useful data-center cost estimate must contain
A single “cost per MW” figure is only meaningful when two projects use the same denominator and scope. One benchmark may include land, substations and fiber; another may cover only the building and mechanical-electrical fit-out. State the basis on the first page of the model and carry it into every comparison.
Free tools Windows power users keep installed
One-click scans. No signup required.
Declare the project basis
- IT load: planned and commissioned MW of server and storage demand, not the utility service rating.
- Area: gross building area, white-space area and any shell reserved for later phases.
- Rack profile: average and peak kW per rack, rack count and the share of high-density or liquid-cooled racks.
- Resilience: availability objective and topology, such as N, N+1 or 2N for power and cooling.
- Commissioning level: testing, integrated systems testing, seasonal testing and operational-readiness requirements.
- Phasing and utilization: energized capacity by phase and the expected ramp from empty halls to operating load.
- Inclusions: servers, tenant fit-out, land, utility interconnection, substations, roads, water works and carrier or fiber extensions.
- Commercial basis: currency, tax treatment, price date, escalation convention, financing assumptions and contingency policy.
Use “commissioned IT MW” as the denominator for a live facility. If a project has 24 MW of utility capacity but only 16 MW of commissioned IT load, its CAPEX per IT MW is based on 16 MW. Show utility capacity separately so the spare headroom is not mistaken for delivered computing capacity.
#1 Best Overall
Build CAPEX from the bottom up
Bottom-up estimating turns a conceptual design into quantities, unit rates and explicit allowances. The estimate should have a scope dictionary, quantity takeoffs, vendor quotations and a log of inclusions and exclusions. Published market comparisons identify planning, permitting, land, construction, substations and fiber as capital components; omitting any of them can make a per-MW benchmark look artificially low.
CAPEX work breakdown
| Cost block | Include in the estimate | Questions that change the value |
|---|---|---|
| Land and site development | Land purchase or lease premium, surveys, geotechnical work, grading, drainage, roads, retaining structures and security perimeter. | Is the parcel powered and serviced, or must the project fund major civil and utility extensions? |
| Planning, design and approvals | Architecture and engineering, environmental studies, planning, permits, legal, development charges and owner’s representative costs. | Are permitting and utility studies included, and how much redesign is likely as requirements change? |
| Building shell and interiors | Structure, envelope, roofs, loading areas, office and support spaces, data halls, floors, fire-rated partitions and interior finishes. | Is this a new shell, an adaptive reuse project or a fitted-out colocation hall? |
| Utility intake and high-voltage systems | Medium- and high-voltage service, substations, transformers, switchgear, protection, metering and interconnection works. | Who pays for the substation, feeder, easements and network upgrades, and when is energization available? |
| Critical power | UPS modules, batteries, generators, fuel systems, paralleling gear, automatic transfer, busway, distribution and grounding. | Does the topology require N, N+1 or 2N equipment, and what autonomy and testing regime is required? |
| Cooling and heat rejection | Chillers, cooling towers or dry coolers, pumps, heat exchangers, piping, controls, water treatment and discharge systems. | What climate, water availability, free-cooling hours and rack-density profile must the design support? |
| Fire, security and controls | Detection and suppression, access control, CCTV, intrusion systems, building-management systems and DCIM. | Are special suppression zones, dual control rooms or additional monitoring required by the operator or insurer? |
| Connectivity | Meet-me rooms, carrier entrances, diverse pathways, internal network rooms and carrier or fiber extensions to the site. | Does the site already have diverse carrier routes, or must the project fund a new extension? |
| Commissioning and owner’s costs | Factory and site acceptance tests, integrated systems testing, training, documentation, spares, insurance, project management and temporary facilities. | What evidence is required before a hall can accept production load? |
| Escalation, financing and contingency | Price escalation to the midpoint of construction, interest during construction, currency allowance and contingency matched to design maturity. | Is the estimate early concept, developed design or construction-ready, and how long can procurement take? |
AI-oriented and other high-density deployments can add electrical and cooling capacity, liquid-cooling distribution, heat exchangers, control changes and retrofit allowances. Turner & Townsend’s 2025–2026 index identifies AI demand, power and water constraints and alternative delivery models as major cost themes. Treat those items as explicit line items rather than hiding them in a generic mechanical allowance.
Calculate and disclose CAPEX per MW
Use CAPEX per IT MW = eligible project CAPEX ÷ commissioned IT MW. Define “eligible” in a footnote: for example, “includes land, utility interconnection, substations, fiber and commissioning; excludes servers and tenant equipment.” Report total CAPEX, the denominator and the inclusion list together.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →REOS’s 2025 Global Data Center Market Comparison reported construction costs exceeding $10 million per MW in its referenced markets. That is a market comparison, not a universal price. It cannot be applied to a different country, redundancy standard or scope without normalization. A shell-only project and a fully powered, commissioned campus should never share one unlabeled benchmark.
Account for schedule, escalation and delay
Schedule is a cost variable. Long-lead transformers, switchgear, generators, chillers and utility works can move the energization date even when the building is complete. Build a procurement and construction schedule that identifies design release, equipment orders, factory tests, site delivery, installation, commissioning and utility energization.
Rank #2
Use dated escalation assumptions
Apply an index from the estimate date to the expected spending date, not a single blanket percentage on the final total. Cushman & Wakefield reported average APAC construction-cost inflation of 3.8% in 2024–2025. Use that figure only as a dated APAC reference; local labor, equipment and currency movements may differ.
Cushman & Wakefield also reported a 21% average rise in U.S. and Canadian construction cost per MW versus its prior edition in 2026. The comparison signals market movement, but it does not replace a location-specific quantity estimate. Record the index, geography, publication date and currency in the workbook.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Model a delay case
At minimum, create a base schedule and a downside schedule with delayed utility energization or a long-lead equipment slip. Add the resulting interest during construction, extended project-management and security costs, storage or re-handling, temporary power, and lost operating revenue where relevant. Keep delay costs separate from physical construction cost so management can see which risk is driving the total.
Model annual operating expenditure (OPEX)
Electricity is usually the most sensitive operating variable because every unit of IT energy is multiplied by PUE and then priced at the delivered tariff. Cushman & Wakefield describes power as one of the most significant operating expenses of a data centre. The OPEX model must still include non-energy costs that continue when utilization is low.
Annual OPEX categories
| Category | What to model | Common driver |
|---|---|---|
| IT electricity | Server, storage and network energy by hour or operating state. | IT load, utilization, workload mix and growth ramp. |
| Facility electricity | Cooling, pumps, fans, lighting, controls and other overhead represented through PUE. | PUE, weather, redundancy mode and operating set points. |
| Utility charges | Delivered energy tariff, demand charges, taxes, wheeling, power-factor penalties and other adjustments. | Tariff schedule, monthly peak kW or kVA and jurisdiction. |
| Water and treatment | Make-up water, sewer, cooling chemicals, testing and discharge fees. | Cooling topology, climate, cycles of concentration and local rates. |
| People and security | Facilities engineering, operations, security, training, uniforms and shift coverage. | Staffing model, site hours, access controls and service-level obligations. |
| Maintenance and spares | Power, cooling, controls, fire and security contracts; inspections; consumables; critical spares. | Equipment count, warranty terms, redundancy and service response time. |
| Connectivity and software | Carrier circuits, cross-connects, DCIM, monitoring, licenses and support. | Carrier diversity, bandwidth, tenant services and software seats. |
| Property and compliance | Insurance, property taxes, leases, permits, audits, certifications and reporting. | Asset value, local taxes, contract obligations and regulatory regime. |
| Lifecycle reserves | UPS batteries, generators, chillers, pumps, controls and other planned replacements; eventual decommissioning. | Replacement interval, failure consequences and inflation. |
Express long-lived costs as annual reserves or scheduled cash outlays. A facility with low routine maintenance can still require a large battery or chiller replacement in a particular year; averaging that obligation prevents the operating plan from overstating steady-state cash flow.
Use PUE correctly in the power calculation
The European Union definition is PUE = total data-center energy (E_DC) ÷ IT-equipment energy (E_IT). A PUE of 1.5 means the facility uses 1.5 units of total energy for each unit delivered to IT equipment; it is not a percentage efficiency score.
Core formulas
Annual facility energy = annual IT energy × PUEAnnual electricity cost = annual facility energy × delivered tariff + demand charges + taxes and utility adjustmentsPresent value of OPEX = discounted annual OPEX + planned lifecycle replacements + decommissioning costs
Use an hourly or monthly load profile where tariffs or demand charges vary. A simple full-load estimate is useful for a sanity check, but it can overstate energy when the site ramps slowly or runs below nameplate.
Worked 20 MW illustration
Cushman & Wakefield’s 2025 illustration places annual power cost at $30.2 million for a 20 MW IT workload, PUE of 1.5 and an average tariff of 11.48 cents per kWh. The figure is an APAC-average example, not a universal price, and it excludes the project-specific effects of utilization, hourly load shape and demand charges.
The same illustration reports approximately $1.9 million in annual power-cost savings for a 0.1 reduction in PUE. Recalculate the saving with your own metered IT energy and tariff: saving = annual IT kWh × 0.1 × tariff. If the workload is not at full load all year, the result will differ from a nameplate-based calculation.
Compare sites and designs on lifetime cost
Rank alternatives using one normalized IT-load basis and a common study period. A cheaper shell can have a higher lifetime cost if it has an expensive tariff, slow interconnection, poor cooling efficiency or a heavy maintenance burden.
Recommended Free Tools
| Comparison axis | What to measure |
|---|---|
| Capital intensity | CAPEX per commissioned IT MW with identical inclusion and exclusion rules. |
| Time to power | Utility commitment, interconnection milestones, equipment lead times and commissioning duration. |
| Electricity price | Delivered energy tariff, demand charges, taxes, power-factor rules and renewable-power premiums. |
| Efficiency | Design PUE, expected seasonal performance and measured PUE after stabilization. |
| Water and resilience | Water use, drought exposure, cooling redundancy and ability to operate during supply restrictions. |
| Availability | Power and cooling topology, maintenance bypasses, fault tolerance and testing requirements. |
| Operating burden | Labor, security, maintenance contracts, spare parts and specialist availability. |
| Connectivity | Carrier count, route diversity, latency, meet-me-room capacity and extension cost. |
| Carbon obligations | Grid emissions, renewable procurement, reporting requirements and potential carbon charges. |
| Expansion | Reserved electrical, cooling, structural, land and fiber capacity, plus the cost of activating it. |
Put the alternatives in a discounted cash-flow model. Discount annual OPEX, planned replacements and residual or decommissioning costs, then add CAPEX and financing. This exposes cases where a higher upfront efficiency investment pays back through lower energy and demand charges.
Run base, downside and upside scenarios
A point estimate hides the assumptions that matter most. Create at least three cases and vary the inputs independently where they are not correlated.
Minimum sensitivity set
- Commissioned IT load and utilization ramp.
- PUE, including seasonal or degraded performance.
- Energy tariff, demand charge and power-factor penalties.
- Construction inflation, currency and financing rate.
- Commissioning date and utility-energization delay.
- Contingency percentage according to design maturity.
- Water price or availability and cooling technology.
- High-density rack share and liquid-cooling allowance.
Show the change in total CAPEX, annual OPEX, present value and break-even date for each sensitivity. For an efficiency project, compare its incremental CAPEX with the avoided energy and demand cost at the tariff scenarios; do not use a single average PUE or tariff if the project will operate across several seasons or price bands.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Measure the operating facility and reconcile the model
Replace assumptions with revenue-grade measurements after commissioning. Install meters that separately capture utility intake, major mechanical and electrical systems and IT distribution. A DCIM or energy-management system can aggregate the readings, but the meter hierarchy and data quality rules must be documented.
Monthly reconciliation procedure
- Record total facility energy, IT-equipment energy, peak demand, water, fuel tests and maintenance spend for the month.
- Calculate PUE from the same boundary used in the model:
E_DC ÷ E_IT. - Compare actual IT load and utilization with the forecast ramp; explain any difference before changing the baseline.
- Reconcile the utility bill to revenue-grade meter totals, including demand, taxes and adjustments.
- Allocate energy and maintenance costs by hall, phase or tenant where submetering supports it.
- Update the forecast with measured weather, tariff and equipment performance, and version the assumptions by date and geography.
Natural Resources Canada’s 2024 guidance recommends monitoring PUE, power density, carbon-usage effectiveness, water-usage effectiveness, cooling systems and subsystem energy. European Union rules require standardized energy measurement. These practices make the estimate auditable and reveal whether a design PUE is being achieved in operation.
A practical estimating workflow
- Write the basis sheet. State IT MW, rack density, area, redundancy, phase plan, utilization, commissioning standard, location, currency and every inclusion or exclusion.
- Map the site and utility scope. Confirm land, geotechnical conditions, water, fiber routes, utility capacity, substation responsibility and energization milestones.
- Create the CAPEX work breakdown. Use the cost blocks above, attach quantities and vendor quotations, and separate owner’s costs, escalation, financing and contingency.
- Normalize the benchmark. Divide eligible CAPEX by commissioned IT MW and publish the scope beside the result. Do not compare shell-only and fully commissioned values.
- Build the schedule. Link long-lead procurement, construction, utility works and commissioning; price a delay case.
- Build the OPEX workbook. Calculate IT energy, apply PUE, price the tariff and add demand, water, labor, maintenance, connectivity, taxes, insurance and lifecycle reserves.
- Run scenarios. Produce base, downside and upside cases for load, PUE, tariff, inflation, financing, contingency and commissioning date.
- Set measurement requirements. Specify meter accuracy, current-transformer range, voltage, logging interval, communications protocol and certification before purchasing equipment.
- Reconcile after go-live. Compare modeled and metered E_IT, E_DC, PUE, demand peaks, water and maintenance each month, then update the forecast with a controlled version history.
Common estimating failures and their fixes
Failure: quoting a universal price per MW
Fix: attach the geography, price date, redundancy, commissioning level and inclusion list to every benchmark. Use the $10-million-plus-per-MW REOS reference only as a dated market comparison, not a project quotation.
Failure: treating utility works as a minor add-on
Fix: identify substations, feeders, easements, protection, studies, fiber and utility deposits as separate scope. Confirm which party pays and what must be complete before energization.
Failure: using nameplate load as annual energy
Fix: model the hourly utilization ramp and calculate demand charges from the applicable billing rule. Keep a full-load sanity check, but do not present it as the forecast.
Failure: assuming design PUE equals operating PUE
Fix: model seasonal performance, partial-load behavior and commissioning settings, then verify with boundary-consistent meters after stabilization.
Failure: omitting lifecycle replacements
Fix: schedule batteries, generators, chillers, pumps and controls as dated cash outlays or annual reserves, with escalation and disposal costs.
Failure: buying meters that cannot support the analysis
Fix: match voltage, current-transformer range, logging interval, communications protocol and certification to the facility and the required measurement boundary. A meter that cannot capture the load range or export reliable intervals cannot produce a defensible PUE trend.
What the finished estimate should deliver
The final package should contain a basis sheet, site and utility scope, bottom-up CAPEX, dated schedule, escalation and contingency policy, annual OPEX, PUE and tariff calculations, lifecycle cash flows, scenario sensitivities and a measurement-and-reconciliation plan. Its headline numbers should be total CAPEX, CAPEX per commissioned IT MW, annual OPEX at stated utilization, and present value over the chosen study period—each with its assumptions visible.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
