Data centres

The data centres market covers the physical real estate and infrastructure assets that house servers, storage and networking equipment: colocation facilities (retail and wholesale), hyperscale self-build campuses, powered-shell developments, and the land, power-interconnection rights and cooling infrastructure that make them operable. Viewed through this real estate lens, the market spans site acquisition, entitlement and construction; wholesale and retail leasing to tenants (hyperscalers, enterprises, AI/GPU-cluster operators); and the ownership vehicles -- publicly listed data centre REITs (e.g. Equinix, Digital Realty) and private infrastructure funds -- that hold these assets long-term.

Included: colocation (retail cabinet/cage and wholesale suite leasing), hyperscale and enterprise self-build data centre campuses, powered-shell development, land and grid-interconnection arrangements, and data centre REIT/infrastructure-fund ownership structures.

Excluded: the cloud computing services layered on top of this physical infrastructure (tracked separately under this site's Cloud infrastructure subsector); servers, chips and networking-equipment manufacturing (a distinct hardware market); submarine cables and long-haul fibre networks (tracked under their own Telecom and Digital Infrastructure subsectors); and small enterprise server closets or single-rack deployments below commercially meaningful colocation/wholesale scale.

• AI training and inference workloads: the IEA reports AI-focused data centre electricity consumption rose 50% in 2025 alone (Electricity 2026).
• Hyperscaler capex supercycle: quarterly hyperscale operator capex reached US$142 billion in Q3 2025, up roughly 180% versus three years earlier (Synergy Research Group).
• Continued enterprise cloud migration and SaaS growth, sustaining baseline colocation and wholesale demand independent of the AI cycle.
• Data sovereignty and localisation rules pushing governments and regulated enterprises toward in-country capacity rather than cross-border cloud regions.
• 5G and edge computing pushing a smaller but growing share of capacity toward distributed, latency-sensitive edge facilities.

• Hyperscaler self-build campuses (Amazon, Microsoft, Google, Meta and increasingly AI-native 'neocloud' operators) now account for the largest share of new capacity additions, often bypassing third-party colocation entirely.
• Third-party wholesale and retail colocation operators (Equinix, Digital Realty, NTT, Vantage, QTS and others) supply the remaining capacity, increasingly specialised toward AI/GPU-dense, liquid-cooled suites.
• A smaller 'powered shell' developer segment builds and leases core-and-shell facilities to both hyperscale and colocation tenants, monetising land and power-interconnection positions without operating the completed facility.
• Land and grid-interconnection queue position, not capital, is now the binding constraint on new supply in most primary markets (Cushman & Wakefield, 2026; JLL, 2026).

• Hyperscale cloud providers (AWS, Microsoft Azure, Google Cloud) leasing wholesale or build-to-suit capacity for public cloud regions.
• AI-native compute providers ('neoclouds') leasing large contiguous blocks for GPU training and inference clusters.
• Regulated enterprises (financial services, healthcare, government) requiring colocation for compliance, disaster recovery and latency-sensitive workloads.
• Content, media and telecom carriers requiring interconnection-dense retail colocation for peering and content delivery.
• Sovereign and government cloud programmes requiring in-country, sometimes air-gapped, capacity.

• Power availability and time-to-energisation, now the dominant site-selection variable ahead of latency (Cushman & Wakefield, May 2026).
• Land cost, availability and entitlement speed.
• Fibre and interconnection density; carrier-neutral facilities command a premium for latency-sensitive tenants.
• Sustainability commitments: renewable-energy access and PUE/WUE performance, increasingly a contractual requirement for hyperscale tenants.
• Regulatory stability: exposure to moratorium or permitting risk is now an explicit site-selection filter.

The colocation and wholesale segment is moderately concentrated at the top and highly fragmented below it: Equinix, Digital Realty, NTT Global Data Centers, KDDI and QTS together hold roughly 20-25% of global colocation revenue (MarketsandMarkets, 2025), while in the US specifically, Equinix, Digital Realty, QTS, CoreSite, Cyxtera and Iron Mountain account for an estimated 45-50% of installed colocation capacity. Institutional capital has funded a wave of AI-focused challengers, including Vantage Data Centers and Applied Digital, built specifically around large contiguous GPU-cluster leases rather than traditional retail cabinet colocation. The larger competitive dynamic is hyperscalers increasingly self-developing capacity rather than leasing it, which shifts share away from third-party operators even as total market size grows.

• Land acquisition and entitlement in markets with available grid capacity.
• Securing grid interconnection and power-purchase agreements, increasingly including behind-the-meter generation.
• Powered-shell construction (building envelope, electrical and mechanical backbone).
• IT fit-out: cooling (increasingly liquid cooling for AI racks), power distribution, fire suppression, security.
• Operation and leasing, either self-operated (hyperscale) or leased to tenants (colocation/wholesale).
• Interconnection and connectivity layer linking the facility to carriers, internet exchanges and cloud on-ramps.
• End customer: cloud tenant, enterprise IT department or AI compute buyer.

• Direct wholesale leases and build-to-suit agreements negotiated directly with hyperscale/AI tenants.
• Retail colocation sold via operator sales teams and channel/reseller partners for cabinet- and cage-scale deployments.
• REIT-owned portfolios leased through in-house leasing and asset-management teams.
• Brokered transactions arranged by commercial real estate advisory practices specialising in data centres (JLL, CBRE, Cushman & Wakefield).

• Liquid cooling adoption for AI/GPU racks, as rack power density rises from a legacy ~5-10kW to 100kW+ for dense AI clusters.
• On-site and behind-the-meter power generation (gas turbines, small modular reactor proposals, co-located solar-plus-storage) used to bypass multi-year grid interconnection queues.
• Modular and prefabricated construction methods to compress build timelines in a power- and schedule-constrained market.
• Growth of edge/micro data centres for latency-sensitive AI inference, distinct from the centralised hyperscale training buildout.
• Rising scrutiny of water usage effectiveness (WUE) alongside PUE, as cooling-related water consumption draws local political attention in water-stressed markets.

The EU's Energy Efficiency Directive (Article 12) requires data centres in the EU with installed IT power demand of 500kW or more to report annual energy-performance and sustainability KPIs, including PUE, WUE, Energy Reuse Factor and Renewable Energy Factor, to a central European database by 15 May each year; individual member states can set a lower threshold (Germany: 300kW; France: 100kW) (Concerted Action EED, DG ENER).

In the United States, moratorium and restriction activity has accelerated sharply: Good Jobs First counted at least 12 in-session state moratorium bills and more than 100 local moratorium actions in the 2025-2026 cycle, with 54 of at least 63 tracked local actions passing. Virginia's HB 1515 would block final rezoning/permit approvals for new data centres until specified grid-interconnection conditions are met.

Ireland's Commission for Regulation of Utilities ended a four-year de facto moratorium on new data-centre grid connections in December 2025, replacing it with strict conditions: large (>10MVA) schemes must now provide behind-the-meter generation sized to 100% of their grid connection, site in 'unconstrained' parts of the grid, and match 80% of annual demand with new Irish renewable investment (KPMG Ireland).

Singapore is emerging from its own three-year moratorium via a phased pilot, having awarded 80MW of new data-centre capacity to four operators in mid-2025.

• Powered-shell development in secondary/tertiary markets with available grid capacity (e.g. Columbus, Ohio; Johor, Malaysia), where land and interconnection queues are less congested than saturated primary markets.
• Behind-the-meter power development and on-site generation partnerships, positioned as a way for tenants and developers to bypass grid interconnection bottlenecks.
• Edge/micro data centre deployment for latency-sensitive AI inference, a smaller but structurally distinct niche from the centralised hyperscale training buildout.
• Liquid-cooling retrofit and specialist mechanical/electrical services for existing air-cooled facilities being converted to support AI/GPU racks.

Granular multi-year historical back-series (e.g. year-by-year global market size for 2019-2024) are typically sold as proprietary datasets by firms such as Synergy Research Group, Cushman & Wakefield and JLL; only their most recently published headline figures are cited here, so the Historical market size module intentionally shows only the most recent 1-2 years rather than a fabricated back-series. Cap rates, transaction yields and detailed unit-economics for data centre real estate are similarly proprietary and are not included here; occupancy and PUE are used as public proxies instead. Figures blend multiple third-party research methodologies (real-estate brokerage market comparisons, technology-market sizing firms, and an intergovernmental energy agency) that do not all define 'data centre market' identically; see Scope and exclusions above for this page's own boundary.

This page draws on: (1) commercial real estate brokerage data centre market reports (Cushman & Wakefield, JLL) for pricing, vacancy and market rankings; (2) technology-market sizing firms (Grand View Research, Precedence Research, MarketsandMarkets) for market-size and forecast figures; (3) the International Energy Agency for electricity-demand figures; (4) Synergy Research Group for hyperscale capacity and capex figures; (5) the Uptime Institute's annual global survey for PUE/efficiency benchmarks; and (6) policy-tracking organisations (Good Jobs First) and professional-services firms (KPMG) for regulatory and moratorium developments. Every figure above is individually dated and sourced (see Sources and review below); no figure on this page is estimated or interpolated by this site itself.