Data centres have become one of the most structurally driven asset classes in commercial real estate, with demand underpinned by AI and machine learning compute requirements, enterprise cloud migration, and the ongoing buildout of global digital infrastructure. Hyperscale cloud providers (Amazon Web Services, Microsoft Azure, Google Cloud) are committing to multi-gigawatt capacity expansions through the late 2020s, while enterprise co-location demand from financial services, healthcare, and media sectors continues to compound.
Unlike most CRE demand drivers, data centre absorption is not cyclically correlated with office employment or retail spending; it tracks compute load, which is structurally growing. Markets with abundant low-cost power, access to fibre routes, and favourable permitting environments (Northern Virginia, Phoenix, Dallas, Toronto, and Amsterdam) have seen land and power easement values escalate significantly as a consequence.
Power is the primary value driver in data centre underwriting, and two metrics define the conversation: critical IT load measured in kilowatts per rack (kW/rack), and Power Usage Effectiveness (PUE). Critical IT load refers to the power consumed by computing equipment itself, exclusive of cooling, lighting, and facility systems.
PUE is the ratio of total facility power consumption to critical IT load; a PUE of 1.0 would mean all power goes directly to computing, while a PUE of 2.0 means the facility consumes twice as much power as the IT load it supports. Per the Uptime Institute's annual Global Data Centre Survey, the industry average PUE has declined from approximately 2.0 in 2007 to 1.58 in 2022, with hyperscale facilities achieving 1.1 to 1.3 through advanced cooling architectures.
Low PUE commands meaningful rent premiums in powered-shell and build-to-suit deals because tenants' total power cost is the product of IT load multiplied by PUE multiplied by the utility rate; a 0.3 reduction in PUE at 10 MW of critical load is worth millions of dollars per year in operating savings.
Lease structure is the most important determinant of cap rate in data centre investing, because it governs who bears operational risk. Hyperscale NNN deals (a single investment-grade cloud provider on a 10-to-15-year absolute NNN lease covering an entire campus) are priced as credit instruments rather than traditional real estate: cap rates for newly executed hyperscale leases from Amazon, Microsoft, or Google have compressed to the 5-6% range, per CBRE's North America Data Centre Trends report, with investors prioritising counterparty credit quality and lease duration over underlying replacement cost.
Multi-tenant colocation leases, by contrast, are structured on a per-cabinet or per-kilowatt basis with terms of 1-5 years, monthly recurring revenue from dozens of tenants, and meaningful rollover risk when a single large tenant represents 20-30% of revenue. Colocation assets trade at wider cap rates (6.5-8%) reflecting operational complexity, leasing velocity assumptions, and management intensity.
Valuation of data centres presents challenges that standard income approach models do not fully capture. The income approach dominates; replacement cost alone cannot anchor value when a 100 MW hyperscale campus requires both a purpose-built power substation and a utility interconnection agreement that may take 18-36 months to replicate.
Single-tenant hyperscale assets are frequently valued by reference to the tenant's investment-grade credit spread and the lease's remaining term, treating the asset more like a corporate bond with a real estate residual than a property with a going-in cap rate. The terminal value assumption is particularly sensitive to power infrastructure: if the utility interconnection and cooling infrastructure cannot be re-leased to a successor tenant at similar power density, the residual real estate value without the operating tenancy is substantially lower.
Appraisers underwriting data centres for institutional lenders apply USPAP Standards Rule 1-4 and must explicitly address the going-concern component versus the underlying real property value.