Cost to build a data centre: capital benchmarks and cost drivers

Institutional investors and developers evaluating project feasibility must verify the capacity denominator and technical scope before applying capital expenditure per megawatt (MW) rates. In a benchmark published in September 2026, Cushman & Wakefield reports that all-in greenfield development costs in the United States and Canada average US$17.6 million per MW for the most modern facilities, excluding chips and GPUs. This figure serves as a published benchmark rather than a project quotation. Published metrics vary globally because reporting methodologies measure different scopes of work, structural specifications and regional cost bases.

Global comparisons demonstrate divergence across international metro areas. The Turner & Townsend Data Centre Construction Cost Index 2025–2026, drawn from a survey of 250 experts across 52 markets, provides an index ranking and indicative US$/W benchmarks to illustrate regional variations.

When interpreting published metrics, capital allocators must verify scope boundaries and underlying exclusions. A CapEx and OpEx benchmark published by KPMG is based on cloud-only projects, greenfield developments, 2025 pricing, and explicitly excludes abnormal items: major grid reinforcements, abnormal ground conditions, exceptional planning obligations, and bespoke customer requirements, so figures reflect steady-state builds, not hyperscale retrofits or customised AI/HPC deployments.

Evaluating development proposals requires verifying the exact scope of works included in quoted unit rates. Financial presentations can compress complex capital expenditure into single per-megawatt figures, masking critical boundary differences:

• Basis of capacity measurement: Unit rates can be stated against incoming utility substation capacity, gross facility power, or critical IT load. Calculating capital against incoming substation capacity deflates apparent unit rates because gross power supports cooling systems and transformation losses. Feasibility models must evaluate capital against usable critical IT load.
• Active IT hardware scope: Does delivery terminate at the power distribution unit or rack tap-off, or extend to turnkey white space? As boundaries depend on contracted powered-shell versus turnkey scope, buyers should establish the electrical and mechanical handover boundary, verify active IT equipment inclusion or exclusion, and confirm any separate equipment budget.
• Powered shell versus turnkey fit-out: A powered shell provides the bare building envelope, foundation and primary high-voltage utility connection, leaving data halls unimproved. Fit-out capital for UPS modules, switchgear, generators and cooling plant is deferred until tenant commitments occur, requiring separate secondary fit-out budgeting.
• Redundancy topologies: Which systems are included, and what topology is specified for each critical path? Diligence should review buyer-specified topologies such as N+1 or 2N, establish the equipment duplication required for each system, and verify what capital allowance is quoted for switchgear, UPS modules and generators.

A data centre development budget divides into five primary stages: site acquisition and civil works, electrical infrastructure, structural shell construction, mechanical cooling systems, and professional fees alongside commissioning.

Land acquisition and site preparation

Site preparation and civil enablement costs can escalate rapidly. Enablement works require geotechnical stabilisation, cut-and-fill grading, acoustic berms, security perimeters and remediation. Poor subsurface bearing capacity requires piling, increasing direct costs and extending schedules before structural foundations commence. Diligence must establish whether the vendor has completed geotechnical boreholes and environmental soil assessments before site pricing is agreed.

Electrical infrastructure and utility connection

Securing high-voltage electrical capacity represents a critical cost driver in modern data centre delivery. In established metro markets, direct substation connections require substantial advance capital and extensive coordination with transmission system operators. As documented by KPMG, utility connection works and external electrical infrastructure are becoming a larger component of total project capital in a number of Tier-1 markets, particularly where grid scarcity persists.

Developers frequently fund contestable works, including high-voltage switchyards, step-down substations, underground transmission cable routes and regional network reinforcements beyond the site boundary. On-site distribution requires UPS modules, medium-voltage switchgear, busways and standby generators with bulk fuel storage.

Structural shell and civil engineering

The physical envelope comprises concrete foundations, structural steel framing, precast cladding, roofing systems and loading bays. Structural design must support heavy equipment loading: enterprise facilities accommodate standard cabinets, whereas artificial intelligence halls require reinforced slabs for dense racks, coolant distribution units (CDUs) and fluid piping manifolds.

Mechanical fit-out and cooling architectures

Mechanical and electrical plant specifications depend on rack density and cooling design. What proportion of the mechanical, electrical and plumbing budget is allocated to cooling plant versus electrical distribution? Conventional facilities rely on air cooling architectures, such as direct evaporative cooling, computer room air handlers (CRAHs) and chilled-water loops paired with dry coolers, handling standard enterprise power densities.

High-density artificial intelligence clusters generate substantial heat per cabinet, which can make air cooling insufficient on its own for certain dense configurations. Where liquid cooling is deployed, designs incorporate direct-to-chip liquid loops or rear-door heat exchangers. What heat rejection method, fluid loop design and tenant heat-rejection interface does the intended compute workload require? Does quoted cooling capital cover facility mechanical plant alone, or does it incorporate active IT hardware across planned rack densities?

Before allocating development equity or committing debt financing, institutional investors and developers should examine core technical, utility and planning assumptions against projected financial models:

• Grid connection status and contestable works: Has the utility issued a binding connection offer with fixed capacity and energisation dates? What portion of utility infrastructure is contestable, and who assumes cost-overrun exposure for off-site reinforcement?
• Geotechnical and environmental site conditions: Have geotechnical boreholes, contamination surveys and flood models been completed? Do foundation specifications account for seismic requirements, ground settlement risks or high groundwater levels?
• Cooling design and municipal resource limits: Does mechanical cooling comply with local water consumption limits, acoustic boundaries and ambient temperature extremes? What is the modelled power usage effectiveness (PUE) across seasonal peaks?
• Supply chain commitments and equipment lead times: Have critical electrical and mechanical assets (transformers, switchgear, UPS batteries, generators, CDUs) been secured via reservation agreements or production slot deposits?
• Procurement structure and risk allocation: Is the project procured under a turnkey EPC contract, a fixed maximum price contract, or a design-bid-build structure? Where are commodity inflation, currency risks and contractor delay liabilities allocated?

Procurement structure directly affects total capital expenditure and delivery schedules. Turnkey Engineering, Procurement and Construction (EPC) contracts consolidate design and construction responsibility under a single contractor. This approach provides cost certainty through fixed pricing models, though contractors incorporate risk premiums for utility interconnection and supply chain lead times. Conversely, construction management or design-bid-build models enable developers to procure long-lead electrical equipment directly from equipment manufacturers. This reduces contractor mark-ups but transfers coordination and schedule risks to the project sponsor.

Controlling build cost per MW starts with selecting a site that already sits near firm grid capacity and matching the build programme to committed customer demand. Our site-finding and signals engine tracks 2,004 project signals across Australia (215) and the UK (1,789), including 1,340 planning records and 663 operator announcements, and screens 683,150 Australian land parcels against published network capacity. UK candidate sites are scored on firm grid headroom, water, fibre and zoning, with flood and protected areas excluded.

What we deliver for developers, buyers and site owners:

• Find a site: Off-market and pre-market powered land and data centre sites, sourced and screened against your MW, land, grid and timing requirements, with a shortlist you can act on.
• Find tenants and offtakers: Anchor tenants and pre-lets for greenfield builds and powered shells so fit-out capital is backed by contracted revenue.
• Sell or lease a site: Land, powered sites and consented projects put in front of the right developers, operators and buyers discreetly. We partner with developers and pair you with the right one for your project.
• Find capacity: Colocation, wholesale and AI-ready capacity sourced against power, density, location and delivery date, with operator proposals compared side by side.

Projects that need funding: we introduce investors and JV partners. Our market directories also cover 115 facility profiles and 44 regional operator profiles across Australian data centres and UK data centres, with facility and colocation coverage across the United States and Canada.