iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
The reliable way to estimate a data center is to define the project boundary first, then model capital cost, operating cost, financing, replacement cycles, and schedule risk on a common capacity and service-level basis. A single “cost per megawatt” figure is only a screening input: shell-and-core construction, greenfield development, tenant fit-out, IT hardware, utility work, and land can sit in different scopes.
For an initial 2026 screen, JLL forecasts $11.3 million per MW of IT capacity for average global shell-and-core construction of a single-tenant, 50 MW, air-cooled facility. That figure excludes land and active IT equipment. A separate Cushman & Wakefield benchmark puts modern U.S. and Canadian greenfield development at $17.6 million per MW excluding chips and GPUs. Neither number is a project budget. Use the workflow below to reconcile scope, local conditions, workload, and lifecycle cost before seeking quotes.
What a data center cost estimate must define
Start by writing a one-page estimate basis. Every number in the model should carry these assumptions:
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallOutdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware match- Geography and currency: identify country, metro, currency, tax treatment, and the price date.
- Capacity basis: state whether capacity is IT load in MW, critical power, white-space area, rack count, or another measure. Do not mix facility megawatts with utility-service megawatts.
- Workload: distinguish enterprise applications, cloud services, high-performance computing, and AI. Rack density and cooling requirements can change the fit-out substantially.
- Availability target: document redundancy, maintenance philosophy, uptime objective, and any certification or regulatory requirement.
- Cooling and power design: specify air, liquid, or hybrid cooling; voltage; utility connection; generators; batteries; and fuel arrangements.
- Schedule: include land control, permitting, utility energization, construction, commissioning, and the date the first useful compute is available.
- Ownership model: decide whether the business needs a wholly owned facility, leased capacity, or computing services.
- Useful lives and refresh policy: set separate replacement periods for facility assets, power and cooling equipment, and IT hardware.
These definitions prevent a common error: applying a benchmark to a project with a different boundary and treating the result as precise.
#1 Best Overall
Choose a benchmark with the same scope
Published figures are useful for screening, but their boundaries differ. Put the scope beside the number in your spreadsheet and in every presentation slide.
| Benchmark | Reported value | Included or excluded scope | Appropriate use |
|---|---|---|---|
| JLL global average, 2026 | $11.3 million per MW | Shell-and-core construction for a single-tenant, 50 MW, air-cooled facility; excludes land acquisition and active IT equipment. | Early facilities-cost screen before local design and quotes. |
| JLL global average, 2025 | $10.7 million per MW | Construction benchmark on the same general basis as the JLL series. | Trend comparison, not an all-in ownership cost. |
| JLL global average, 2020 | $7.7 million per MW | Construction benchmark in the JLL series. | Historical context only. |
| Cushman & Wakefield, 2026 | $17.6 million per MW | Average greenfield development cost for the most modern U.S. and Canadian facilities; excludes chips and GPUs. | Broader greenfield screen after reconciling land, utility, and owner scopes. |
| KPMG sample, United Kingdom | $8.5 million per MW | Base-build construction; excludes tenant white-space fit-out, substations, fibre connections, and work outside the builder’s scope. | Base-build comparison only. |
| KPMG sample, Spain | $6.7 million per MW | Same stated base-build exclusions as the UK figure. | Regional base-build comparison, not a complete project price. |
| Epoch AI stylized U.S. 1 GW model | $38 billion upfront CapEx; $0.9 billion annual OpEx | AI data-center model with its own assumptions; not an estimate for a particular facility. | Illustration of lifecycle modeling, not a vendor quote. |
JLL’s series rose from $7.7 million per MW in 2020 to $10.7 million in 2025, described as a 7% compound annual growth rate, and forecasts $11.3 million in 2026, a 6% increase. These are construction figures, so they should not be presented as total cost of ownership.
JLL’s 2026 market examples
For its 50 MW air-cooled shell-and-core case, JLL’s 2026 chart shows these market-level figures:
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errors| Market | Shell-and-core benchmark |
|---|---|
| Chicago | $14 million per MW |
| Northern Virginia | $12 million per MW |
| London | $14 million per MW |
| Frankfurt | $13 million per MW |
| Tokyo | $18 million per MW |
| Singapore | $15 million per MW |
| Mumbai | $7 million per MW |
These values retain JLL’s chart scope; they are not local bids or universal prices for every project in those markets.
Adjustments that can change the benchmark
JLL notes that liquid-cooled facilities carry a 10% cost premium in its assumptions and that multistory facilities in the Americas require a 20% construction-cost addition. It also says AI infrastructure technology fit-out can cost as much as $25 million per MW. Apply such adjustments only when the project matches the stated condition, and keep them separate from the base benchmark so they are not counted twice.
Build the capital-cost model line by line
Create a cost breakdown that allows an estimator to assign an owner, quote, or assumption to every line. Typical lines are:
- Land and site control: purchase or lease cost, options, surveys, geotechnical work, environmental remediation, and site preparation.
- Civil and shell works: grading, roads, foundations, structure, envelope, roofs, loading areas, and drainage.
- Electrical plant: utility service, transformers, switchgear, generators, UPS systems, batteries, distribution, grounding, and controls.
- Mechanical plant: chillers, cooling towers, pumps, heat rejection, air handlers, piping, liquid-cooling distribution, and water treatment.
- Power infrastructure and interconnection: substations, transmission or distribution extensions, utility studies, protection equipment, easements, and connection fees.
- Security and life safety: access control, monitoring, fire detection and suppression, physical barriers, and emergency systems.
- Fit-out: white-space construction, racks, containment, cabling, overhead systems, staging, and tenant-specific controls.
- IT and network equipment: servers, storage, switches, optics, accelerators, orchestration systems, and spare capacity.
- Professional and owner costs: architecture, engineering, program management, legal, permits, testing, commissioning, insurance, and taxes.
- Financing and contingency: interest during construction, commitment fees, escalation allowances, design development, and risk reserves.
Check each benchmark for owner-furnished equipment and utility work before adding those lines. KPMG’s base-build comparison specifically excludes tenant white-space fit-out, substations, fibre connections, and work outside the builder’s scope. Adding those items to a benchmark that already contains them will double-count; omitting them will understate the project.
Use a transparent calculation
For a first screen, calculate:
screening construction cost = benchmark ($/MW) × planned IT load (MW)
For example, applying JLL’s $11.3 million per MW shell-and-core benchmark to a hypothetical 10 MW IT load gives $113 million of shell-and-core construction. That arithmetic excludes the land, active IT, and other items JLL excludes; it is not an all-in price. A second column should add project-specific lines and show which are quoted, estimated, or not yet known.
Estimate annual operating cost
Operating expense starts with the actual tariff and load profile for the candidate site, not a generic industry average. Model energy in separate components:
- IT electricity: expected load by hour, season, and growth stage.
- Facility energy: cooling, pumps, fans, lighting, controls, and losses.
- PUE assumption: state the assumed power usage effectiveness and test a range rather than hiding it in a single total.
- Demand and connection charges: include tariff structures, contracted capacity, peak demand, and power-quality services where applicable.
- Maintenance: electrical, mechanical, controls, generators, batteries, fire systems, elevators, and building fabric.
- People and services: operations staff, security, cleaning, compliance, training, software, monitoring, and specialist contracts.
- Other utilities and risk costs: water, fuel, waste handling, insurance, property taxes, network transport, and spare-parts inventory.
The available benchmarks do not establish a universal electricity tariff or generic business operating cost. Obtain a utility tariff and a site-specific energy model, then document escalation, renewable-energy contracts, taxes, and pass-through charges.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Annualize assets and model refresh cycles
A build-cost comparison is incomplete without timing. Separate cash spending from annualized economic cost. Set useful lives for the facility, electrical and mechanical plant, building systems, and IT equipment; then model replacements, financing, discounting, and residual value.
Epoch AI’s stylized U.S. 1 GW AI model illustrates the sensitivity. It uses $38 billion of upfront CapEx and $0.9 billion of annual OpEx, and reports an annualized total of about $8.5 billion under assumptions of five years for IT equipment and 14 years for the facility. Changing only the IT life to three years raises the annualized result to $12 billion; extending it to seven years lowers it to $7 billion. Epoch explicitly says the model is not an estimate for any particular facility.
Build a schedule that shows when each asset is paid for and refreshed. A server refresh may require new power distribution, liquid-cooling capacity, network optics, or floor loading even when the building remains useful. Conversely, a facility can be available before IT equipment arrives, creating financing and idle-capacity costs.
Compare build, colocation, and cloud on equivalent service
Do not compare a construction estimate with a monthly cloud bill until the alternatives provide the same workload, geography, performance, availability, growth, and term. Normalize these dimensions:
Recommended Free Tools
| Comparison axis | Questions to answer |
|---|---|
| Workload and capacity | Are compute, storage, network throughput, accelerator type, and growth assumptions equivalent? |
| Geography and latency | Do data-residency rules, users, and latency requirements require the same region? |
| Service level | Is redundancy, maintenance access, recovery time, and availability genuinely comparable? |
| Capital versus recurring fees | Which option requires upfront investment, deposits, minimum commitments, or take-or-pay capacity? |
| Refresh responsibility | Who replaces servers, batteries, chillers, network equipment, and software? |
| Energy and network charges | Are power, bandwidth, cross-connects, egress, and fuel included or passed through? |
| Implementation time | How long do design, migration, testing, and commissioning take before production? |
| Risk and contract term | Who bears construction, utility, price, capacity, and obsolescence risk, and for how long? |
Build
Building can provide control over design, security, and long-term capacity, but the business carries land, permitting, construction, utility, financing, staffing, and refresh risk. It is most sensitive to schedule and utilization: a delayed or under-filled facility can create substantial costs before revenue or productivity benefits arrive.
Colocation
Colocation can transfer much of the facility construction and operations responsibility to a provider while preserving control of owned IT equipment. The estimate must identify what the contract includes: power, cooling, racks, remote hands, cross-connects, security, expansion rights, escalation, minimum commitments, and equipment-refresh responsibility.
Cloud
Cloud converts much of the infrastructure investment into usage or committed-service charges. Compare the complete workload cost, including storage, data transfer, network connectivity, licensing, support, migration, resilience, and egress. A short-term usage price is not automatically comparable with a 14-year facility life and a five-year hardware cycle.
The cited benchmarks do not establish a universal winner or current price for a particular workload. Obtain current, workload-specific calculations or quotes for each option.
Make location, power, and schedule first-class variables
Location affects more than land price. KPMG identifies labor cost, contractor depth and pricing power, planning complexity, utility-interface requirements, supply-chain concentration, and compliance as construction-cost drivers. It also finds owner-supplied mechanical and electrical equipment less regionally variable than construction in its sample, while fit-out becomes more important as rack density and compute intensity rise.
Cushman & Wakefield’s September 2026 release says its guide’s per-MW construction costs rose 21% from the prior edition in the fourth quarter of 2024, citing supply-chain constraints, skilled wages, land, and critical components. It reports powered land in primary U.S. markets averaging $584,000 per MW in 2026 year to date, up 51% year over year. That is a specific powered-land measure, not a general land-price assumption.
Power availability can dominate an otherwise attractive site. JLL reports that average grid-connection wait time in primary data-center markets exceeds four years and identifies speed to power as a primary site-selection criterion. Treat that as a market-level outlook; verify the utility’s study, interconnection queue, deposits, energization date, firm capacity, curtailment terms, and upgrade responsibility for the actual site. JLL says operators are exploring on-site power and colocated battery storage amid constraints, but those options require their own permitting, fuel, safety, emissions, maintenance, and capital models.
“The data center industry is entering a new phase where access to power, land, and skilled labor are increasingly determining where growth can occur,” said John McWilliams, Head of Data Center Insights at Cushman & Wakefield, in the September 3, 2026 release.
Free tools Windows power users keep installed
One-click scans. No signup required.
Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.
Stress-test the estimate before approving it
Create a sensitivity table rather than one “best guess.” At minimum, vary:
- IT load ramp and ultimate capacity;
- air versus liquid cooling and resulting fit-out;
- utility energization date and interim capacity;
- electricity tariff, demand charges, and load factor;
- construction escalation and long-lead equipment;
- labor availability, contractor pricing, and permitting duration;
- contingency, financing rate, and construction draw profile;
- IT refresh life and facility life;
- utilization, migration timing, and stranded capacity.
Show at least three cases—base, delayed or higher-cost, and accelerated or lower-cost—with the assumptions visible. Do not insert a universal overrun percentage: the available sources do not establish one. Use supplier quotes, utility correspondence, design quantities, and signed commercial terms as uncertainty narrows.
Common estimation mistakes and corrections
- Using one generic $/MW number: label shell-and-core, base-build, greenfield, fit-out, and IT scopes separately.
- Mixing IT MW with facility or utility MW: define the capacity basis on every worksheet.
- Ignoring workload density: model rack power, liquid cooling, floor loading, network, and accelerator requirements for AI or HPC.
- Leaving out power-interface work: confirm substations, utility upgrades, fibre, easements, studies, and connection fees.
- Counting only construction: add energy, maintenance, people, taxes, insurance, financing, and refresh cycles.
- Assuming a published market figure is a quote: use benchmarks for screening, then obtain local design and vendor pricing.
- Assuming power is available when the building is complete: put the utility milestone on the critical path and price interim options.
- Comparing unlike alternatives: normalize service level, geography, growth, migration, network, contract term, and refresh responsibility.
A practical approval checklist
- Approve the estimate basis: location, currency, date, IT MW, workload, density, redundancy, cooling, schedule, ownership, and useful lives.
- Select a benchmark whose scope matches the current decision and record all exclusions.
- Build capital lines for land, civil works, power, cooling, utility interface, security, fit-out, IT, professional fees, financing, and contingency.
- Obtain a utility feasibility statement and a dated connection path before treating capacity as committed.
- Model annual energy from the local tariff and load profile, with an explicit PUE assumption.
- Set facility and IT replacement cycles, then calculate annual cash expense and lifecycle cost.
- Normalize build, colocation, and cloud alternatives against the same workload and service level.
- Run schedule, cost, tariff, utilization, cooling, and refresh sensitivities.
- Replace screening assumptions with design quantities, supplier bids, utility terms, and contract prices as the project advances.
A defensible business case is therefore a dated, scoped model—not a headline cost-per-megawatt figure. The model should make exclusions visible, show who bears each risk, and identify the date and conditions under which usable computing capacity actually arrives.
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.
