Overview

The conventional datacenter construction playbook — custom-designed facility, long procurement cycles, sequential trades — is being disrupted by the pace of AI infrastructure demand. McKinsey projects $6.7 trillion in cumulative global datacenter capex through 2030, with demand growing at 22% CAGR. Operators cannot build fast enough using traditional methods. The response is a shift toward modular/prefabricated construction, standardized designs that can be replicated across sites, and early contractor involvement to compress timelines. The intersection of high-density AI workloads and construction timelines is creating demand for purpose-designed AI datacenter facilities that differ significantly from legacy enterprise colocation buildings.

Key Themes

  • Traditional datacenter construction: 24–36 months from greenfield to live; AI urgency pushing toward 12–18 month targets
  • Prefabricated and modular datacenter (PFMD) construction: factory-built power and cooling modules reduce on-site labor and compress schedule
  • Zoned hybrid design: air-cooled zones for legacy/moderate-density workloads alongside liquid-cooled zones for AI/HPC — phased transition rather than all-or-nothing
  • AI datacenter power density planning: designing for 30–100+ kW/rack from the start, not retrofitting
  • Site selection criteria shifting: power availability > fiber availability; proximity to generation, substation capacity, and water supply now primary constraints
  • Design standardization and replication: hyperscalers using “copy-exact” facility designs across global sites to compress schedule and reduce engineering cost
  • Time-to-power as a competitive differentiator: colocation operators who can deliver power faster win hyperscaler commitments

Companies

Startups & Development Partners

Company HQ Stage Mission
MODLOGIQ Bend, OR, USA Growth Modular datacenter construction; prefabricated mechanical and electrical (M&E) modules for accelerated deployment.
Compass Datacenters Dallas, TX, USA Private (OMERS Infrastructure) Purpose-built wholesale colocation; single-tenant design optimized for hyperscaler and large enterprise tenants; known for fast delivery cycles.
Stack Infrastructure Chicago, IL, USA Private (IFM Investors) Wholesale colocation with AI-optimized facilities; partnership with hyperscalers for dedicated campuses.

Public Companies

Ticker Company Mission
DLR Digital Realty Largest global colocation/interconnection REIT; PlatformDIGITAL modular design approach; relevant for construction/design trends at scale.
EQIX Equinix Global colocation and interconnection; xScale hyperscaler joint ventures; International Business Exchange (IBX) standardized design.
IRON Ironmountain Data management and datacenter; Project Raven AI-optimized datacenter development.

Incumbents

Ticker Company Relevance
CBRE CBRE Group Largest commercial real estate firm; datacenter advisory, development management, and construction management services; Data Center Solutions practice.
Mortenson Mortenson (private) One of the largest US datacenter construction contractors; hyperscaler campus builder; modular construction expertise.
Turner Construction Turner (HOCHTIEF subsidiary) Major datacenter construction contractor; AI campus construction experience.

Key Design Approaches Being Tracked

Zoned Hybrid Cooling Design

The emerging standard for new AI datacenters mixes cooling technologies by zone rather than committing to a single approach. Air-contained zones handle legacy and moderate-density workloads; rear-door heat exchanger (RDHx) zones handle 20–40 kW/rack AI inference; direct-to-chip liquid zones handle 40–120 kW/rack GPU training clusters. A shared warm-water secondary loop connects the liquid zones to a central cooling plant (dry coolers preferred for water efficiency, chillers for hot climates). This approach allows phased densification as workloads evolve without stranded cooling infrastructure.

Prefabricated Mechanical and Electrical (M&E) Modules

Factory-built modules containing complete power (switchgear, UPS, PDU) or cooling (CDU, dry coolers, piping) subsystems are delivered to site pre-tested and pre-commissioned, then craned into position and connected. This approach reduces on-site labor (skilled labor is the construction timeline bottleneck), improves quality consistency, and compresses schedule. Schneider Electric, Vertiv, and specialist PFMD contractors (MODLOGIQ, Corscale) offer modular approaches. Hyperscalers have largely developed proprietary modular designs.

Copy-Exact Facility Design

Hyperscalers (Google, Microsoft, Amazon) increasingly deploy standardized facility designs across global sites rather than custom-engineered buildings. A proven design that meets power, cooling, and construction requirements is replicated with minimal modification at each new site. This compresses design timeline, enables contractor familiarity, and reduces commissioning risk. The design-standardization trend is relevant to construction contractors and M&E suppliers who gain volume through preferred vendor relationships.

Supply Chain — Last Reviewed: 2026-03-24

Entries

  • Applied Digital — Dallas-based AI datacenter developer and operator (NASDAQ: APLD); founded 2021 by Wes Cummins and Jason Zhang; 100 MW live, 900 MW under construction, ~3.5 GW pipeline; Polaris Forge 1 (Ellendale, ND) leased to CoreWeave for $7B over 15 years; Polaris Forge 2 (Harwood, ND) $5B lease with unnamed investment-grade hyperscaler; $5B perpetual preferred equity facility from Macquarie Asset Management; proprietary waterless direct-to-chip cooling.
  • CoreWeave — Livingston, NJ GPU cloud company; co-founded 2017 (as Atlantic Crypto) by Michael Intrator, Brian Venturo, Brannin McBee, and Peter Salanki; pivoted from Ethereum mining to AI cloud 2019; IPO March 2025 (CRWV, ~$35B market cap); $5.1B revenue FY2025; $66.8B contracted backlog; designs entire data centers for 130kW+ liquid-cooled racks; NVIDIA strategic partner and preferred GPU allocatee; 5 GW capacity expansion target by 2030; largest 'neocloud' in the AI infrastructure market.
  • Corgan — Dallas-based architecture and engineering firm; #1 ranked data center architecture firm globally (BD+C, 2024–2025); founded 1938; data center practice since 1979; 300 of 1,200 employees on data center projects; ~$135M data center revenue in 2024; 15M+ sq ft designed; portfolio spans QTS, CoreSite, CyrusOne, Vantage, Cologix, and confidential hyperscalers; expertise in modular power rooms, adiabatic dry cooling, multistory high-density design, and AI infrastructure architecture.
  • Crusoe — Denver-based AI Factory developer and cloud provider; co-founded 2018 by Chase Lochmiller and Cully Cavness; $2.77B raised; primary developer for OpenAI's Stargate campus in Abilene, TX ($11.6B financing, ~$15B total committed); zero-water-evaporation direct-to-chip cooling; vertically integrated from power sourcing through managed AI inference. Divested bitcoin mining to NYDIG in 2024 to focus entirely on AI infrastructure.
  • CyrusOne — Dallas-based colocation datacenter operator; former REIT (CONE), taken private by KKR and GIP (now BlackRock) in March 2022 for ~$15B; 55+ data centers across North America and Europe; Intelliscale AI-native design platform supporting 12–300 kW/rack via hybrid liquid cooling; ~$12B in debt financing raised in 2025; major Texas expansion underway (DFW7, DFW10, DFW17); E.ON power partnership for European grid-constrained markets; experienced CME cooling outage November 2025.
  • Digital Realty — Largest global colocation/interconnection REIT (NYSE: DLR); founded 2004; HQ Austin TX; 310+ data centers across 50+ metros in 25+ countries; ~43M sq ft operating; $6.1B FY2025 revenue; PlatformDIGITAL modular design framework; direct liquid cooling available at 170+ facilities supporting 30–150+ kW/rack; NVIDIA AI Factory Research Center partner; $7B Blackstone JV and $3.25B inaugural hyperscale fund; ~$62B market cap.
  • HostDime — Orlando-based global edge colocation operator; founded 2003 by Manny Vivar; self-funded with no outside investors; purpose-built Tier IV datacenter facilities in 8+ countries across Latin America, US, Europe, and Asia; new-generation facilities designed for 30–100 kW/rack AI workloads with direct-to-chip and immersive liquid cooling; 1.3 PUE target; carrier-neutral with local internet exchange points; flagship new SuperNova Orlando facility (100,000 sq ft, 5MW IT load, ~1,000 racks) completing Q1 2026; pursuing potential $100M+ capital raise for expansion.
  • Novva Data Centers — West Jordan, UT wholesale colocation operator; founded 2019 by Wes Swenson and CIM Group; $2B financing (J.P. Morgan + Starwood, March 2025) to complete 175 MW Salt Lake City campus — one of the largest direct-to-chip cooled AI datacenters in the world; waterless cooling; autonomous robot dog and drone security monitoring; Western US footprint across 6 markets.
  • Ramboll — Danish global engineering and architecture firm (founded 1945, 18,000+ employees, Foundation-owned); one of the world's largest datacenter consulting practices with 300+ dedicated experts, 15+ GW designed, $60B+ in due diligence support; built via acquisitions of EYP Mission Critical Facilities (2022) and i3 Solutions Group (2024); flagship work includes Meta Odense heat recovery (45 MW, 12,000 homes), TeraWulf Lake Mariner brownfield-to-compute ($65M, 360+ MW), and Vantage Data Centers Lighthouse Campus (902 MW, 40% biodiversity net gain); known for CFD Digital Wind Tunnel cooling optimization, high-density design methodology (30–100+ kW/rack), and first-of-its-kind 2025 net zero datacenter roadmap.