Chapter 18
Data Center Operators
Chapter 18: Data Center Operators
18.1 Overview
Data centers are where every other layer in this report converges. The chips (Chapter 6), memory (Chapter 8), networking (Chapter 10), optics (Chapter 11), power (Chapters 13-14), cooling (Chapter 15), and servers (Chapter 18) all live inside facilities owned or leased by the operators covered in this chapter. The data center operator layer determines where AI infrastructure gets deployed, how quickly capacity comes online, and who bears the capital cost.
The market splits into four categories. First, hyperscaler self-build: Amazon (AWS), Microsoft (Azure), Google (Google Cloud), and Meta operate the largest data center fleets globally, building massive campuses designed specifically for their own AI and cloud workloads. In 2024, these four spent over $228 billion in combined capex, and 2025 plans exceeded $320 billion. Each builds data centers with 100 MW to multi-GW power capacity, custom-designed for their specific chip architectures (Google TPU, Amazon Trainium, Meta MTIA, NVIDIA GPUs) 1.
Second, colocation REITs: Equinix and Digital Realty are the two dominant public data center REITs. Equinix operates 260+ facilities globally with 2025 revenue guidance of $9.175-9.275 billion and Q3 2025 revenue of $2.316 billion (+5% YoY as reported, +8% normalized constant currency). Its xScale hyperscale brand has approximately 415 MW leased across 21 facilities, with 730 MW in development. Digital Realty controls 290+ facilities with 45 million square feet. Q3 2025 revenue was $1.6 billion (+10% YoY), with adjusted EBITDA of $868 million (+14% YoY). The US colocation market grew from $40.21 billion in 2025 to a projected $72.37 billion by 2030 at 11.5% CAGR 234.
Third, neoclouds: CoreWeave, Lambda Labs, and Crusoe Energy represent a new category of GPU-as-a-service providers. CoreWeave went public in 2025 and reported Q2 2025 revenue of $1.213 billion (up from $395 million a year earlier), with a $30.1 billion revenue backlog. The company ended Q2 with approximately 470 MW of active power and 2.2 GW of total contracted power. CoreWeave builds and operates its own data centers specifically for AI workloads, owning the GPU hardware and selling compute time rather than rack space 5.
Fourth, sovereign AI initiatives: Governments in the Middle East (Saudi Arabia’s Humain, UAE), Southeast Asia, and Europe are building national AI data center capacity to ensure domestic AI capabilities. Cisco’s partnership with Humain to build sovereign AI infrastructure in Saudi Arabia, and various European data sovereignty initiatives, represent a growing segment of the operator market.
The competitive dynamic between these categories is shifting. Hyperscalers are increasingly building their own facilities rather than leasing from colocation providers for their largest AI clusters, because custom-designed campuses can incorporate purpose-built power, cooling, and networking. Colocation providers are responding by offering high-density, AI-ready capacity with liquid cooling and high-power density. Equinix’s xScale program and Digital Realty’s high-density expansions specifically target GPU-intensive workloads from AI customers.
18.2 Market Sizing & Growth
US colocation market: $40.21 billion (2025), projected $72.37 billion by 2030 at 11.5% CAGR 4.
Hyperscaler capex: Amazon, Microsoft, Google, and Meta combined capex exceeded $228 billion in 2024 (+55% YoY) and planned $320 billion+ in 2025. Projected to exceed $600 billion in 2026 16.
Equinix: FY2025 revenue guidance $9.175-9.275B. Q3 2025 revenue $2.316B (+5% as reported). Adjusted EBITDA margin ~49%. AFFO guidance $3.675-3.755B. 499,000+ interconnections. xScale: 415 MW leased, 730 MW in development. $15B plan announced for 25 new IBX facilities in Europe and APAC through 2028 23.
Digital Realty: Q3 2025 revenue $1.6B (+10% YoY). Adjusted EBITDA $868M (+14% YoY). 290+ facilities, 45M sq ft. Broke ground on 150 MW Northern Virginia campus (Dec 2025) 2.
CoreWeave: Q2 2025 revenue $1.213B (vs $395M YoY). Revenue backlog $30.1B. 470 MW active power. 2.2 GW contracted power. TIME100 most influential companies (2024). IPO in 2025 5.
Lambda Labs: Private. GPU cloud for AI researchers. Growing from enterprise AI demand.
Crusoe Energy: Private. Operates data centers powered by stranded natural gas (flare gas). Targeting AI inference workloads.
18.3 Supply Chain Flowchart
DATA CENTER OPERATORS
|
|---> HYPERSCALER SELF-BUILD (own design, own operation)
| Amazon/AWS: largest DC fleet globally; custom Graviton/Trainium
| Microsoft/Azure: Stargate ($100B+), TMI nuclear PPA
| Google/GCP: TPU-optimized DCs; Kairos nuclear PPA (500 MW)
| Meta: $600B US infrastructure plan; 1GW Prometheus, 5GW Hyperion
| Oracle: OCI expansion; SMR-powered DC design announced
| Trend: 100 MW → multi-GW campus scale; custom power + cooling
|
|---> COLOCATION REITs (lease space to tenants)
| Equinix (EQIX): 260+ DCs; $9.2B revenue; xScale hyperscale brand
| Digital Realty (DLR): 290+ facilities; 45M sq ft; wholesale + retail
| QTS (Blackstone): major US colo operator; acquired by Blackstone 2021
| CyrusOne (KKR): hyperscale colo; acquired by KKR 2022
| Iron Mountain (IRM): expanding DC footprint (Phoenix, Denver)
| NTT Global Data Centers: $1.8B revenue; strong in APAC
| Vantage Data Centers: hyperscale colo in NA and EMEA
| Switch: hyperscale specialist; Oklo nuclear partnership (12 GW)
|
|---> NEOCLOUDS / GPU-AS-A-SERVICE
| CoreWeave (CRWV): $1.2B Q2 revenue; $30.1B backlog; 2.2 GW contracted
| Lambda Labs: GPU cloud for AI researchers
| Crusoe Energy: flare gas-powered AI data centers ($10B+, 1.2 GW)
| Nebius (ex-Yandex): AI infrastructure ($46B backlog, $3-3.4B 2026 guide)
| Applied Digital (APLD): $23B+ contracted; developer model
|
|---> MINING-TO-AI OPERATORS (~$59B combined AI contract backlog)
| Iris Energy (IREN): $9.7B Microsoft; 4.5+ GW power; Sweetwater 1.4 GW
| TeraWulf (WULF): $12.8B contracts; HPC now majority of revenue
| Hut 8 (HUT): $16.8B leased; $9.8B Beacon Point; BBB- rated
| Core Scientific (CORZ): $10.2B CoreWeave; 590 MW expansion
| Cipher Digital (CIFR): $9.3B (AWS $5.5B); stranded gas model
| Keel Infrastructure (KEEL, fmr. Bitfarms): 2.2 GW; seeking anchor
| Marathon Digital (MARA): Starwood JV; cautious pivot
| Riot Platforms (RIOT): AMD 50 MW; nuclear SMR exploration
|
+---> SOVEREIGN AI DATA CENTERS
Humain (Saudi Arabia): Cisco partnership for sovereign AI infra
G42 (UAE): AI infrastructure provider
European sovereign cloud initiatives
India: rapid DC expansion (NTT: 200 MW; multiple campuses)
Japan/Korea: national AI compute initiatives18.4 Key Companies
| Company | Ticker | Exchange | Approx. Mkt Cap | Role in Buildout | Key Metric |
|---|---|---|---|---|---|
| Equinix | EQIX | NASDAQ | ~$106B | Largest colocation REIT; 260+ DCs; interconnection leader | FY2025 revenue ~$9.2B; xScale 415 MW leased; 730 MW in dev |
| Digital Realty | DLR | NYSE | ~$68.0B | #2 colo REIT; 290+ facilities; wholesale hyperscale focus | Q3 2025 revenue $1.6B (+10%); 150 MW NoVA campus |
| CoreWeave | CRWV | NASDAQ | ~$60.0B | Neocloud GPU-as-a-service; IPO 2025 | Q2 2025 revenue $1.213B; backlog $30.1B; 2.2 GW contracted |
| Iron Mountain | IRM | NYSE | ~$30.0B | Expanding DC REIT; cross-selling from records management | Growing DC footprint in Phoenix, Denver, secondary markets |
| QTS (Blackstone) | Private | Private (Blackstone) | Private | Major US hyperscale colo; taken private 2021 | Significant Virginia, Texas, Georgia campuses |
| Switch | SWCH | Private | Private | Hyperscale DC specialist; Oklo nuclear partnership (12 GW) | Las Vegas, Atlanta, Michigan campuses |
| NTT Global Data Centers | Private | Subsidiary (NTT) | Private | Major global DC operator; strong APAC/Europe | Revenue $1.8B (2024); acquired Mumbai/Bangalore sites ($800M) |
| Vantage Data Centers | Private | Private | Private | Hyperscale colo in NA and EMEA | Growing rapidly with hyperscaler leasing |
| Lambda Labs | Private | Private | ~$2.5B (est.) | GPU cloud for AI researchers and enterprises | Growing from AI training/inference demand |
| Crusoe Energy | Private | Private | ~$10.0B (Series E, Oct 2025) | Flare gas-powered AI data centers; Digital Flare Mitigation | $1.375B Series E (Valor/Mubadala); 17x YoY contract value growth; Abilene expanding to 1.2 GW |
| Applied Digital | APLD | NASDAQ | ~$11.8B | AI data center developer; GPU cloud operator | $23B+ total contracted value; Polaris Forge 1 $11B/15yr CoreWeave lease; Delta Forge 1 $7.5B lease; 38% revenue growth guidance 2026 |
| Nebius Group | NBIS | NASDAQ | ~$45.0B | AI infrastructure company (spun from Yandex); GPU cloud | $46B+ contracted backlog (Meta $27B, Microsoft $19.4B); 2026 revenue guidance $3.0-3.4B (vs $530M in 2025, +479% YoY) |
| Cipher Digital | CIFR | NASDAQ | ~$9.0B | Bitcoin mining pivot to AI infrastructure; stranded gas model | $9.3B in long-term AI contracts (AWS $5.5B/15yr, 300 MW Barber Lake); dual-use crypto/HPC |
| Iris Energy | IREN | NASDAQ | ~$22.0B | Mining-to-AI pivot; hyperscale AI cloud infrastructure | $9.7B Microsoft 5-yr deal; 4.5+ GW secured power; Sweetwater 1 (1.4 GW) operational; acquired Mirantis ($625M) for cloud delivery |
| TeraWulf | WULF | NASDAQ | ~$11.5B | Mining-to-AI pivot; low-cost nuclear/hydro-powered AI hosting | $12.8B contracted HPC revenue; Q1 2026 HPC revenue exceeded mining for first time; Fluidstack 200 MW/10-yr ($9.5B) deal |
| Hut 8 | HUT | NASDAQ | ~$7.3B | Mining-to-AI pivot; investment-grade AI data center leasing | $16.8B leased AI capacity; $9.8B Beacon Point 15-yr deal (352 MW); BBB- rated $3.25B senior notes at 6% |
| Core Scientific | CORZ | NASDAQ | ~$11.0B | Mining-to-AI pivot; CoreWeave anchor tenant model | $10.2B CoreWeave 12-yr deal; 590 MW expansion; 350 MW energized Q1 2026; selling BTC to fund AI buildout |
| Keel Infrastructure (fmr. Bitfarms) | KEEL | NASDAQ/TSX | ~$2.2B | Mining-to-AI pivot (rebranded April 2026); 2.2 GW power secured | Early-stage: seeking first major AI anchor tenant; $520M liquidity; PA, WA, Quebec sites |
| Marathon Digital | MARA | NASDAQ | ~$4.8B | Bitcoin miner with AI/HPC ambitions; Starwood Capital JV | $910M TTM revenue; Starwood JV for 1+ GW AI capacity (Feb 2026); still primarily mining-focused |
| Riot Platforms | RIOT | NASDAQ | ~$9.0B | Bitcoin miner with emerging AI hosting; nuclear SMR exploration | 1.7 GW Texas power; AMD 50 MW partnership; Terrestrial Energy nuclear MoU (4 GW potential) |
| CleanSpark | CLSK | NASDAQ | ~$3.7B | Mining-to-AI pivot; building multi-GW AI data center platform | 1.8 GW under contract; AI-ready facilities in Texas and Georgia; still early execution |
| GDS Holdings | GDS | NASDAQ / HKEX | ~$8.9B | Largest independent DC operator in China | 130+ DCs across China and SE Asia. Dual-listed. |
| Chindata Group | Private | Private (taken private by Bain Capital, $3.16B, Q4 2023) | N/A | China/SE Asia hyperscale DC operator | Delisted from NASDAQ. Focuses on hyperscale campuses; Hebei, Malaysia operations. |
18.4.2 Middle East & Sovereign AI Operators
| Company | Ticker | Exchange | Approx. Mkt Cap | Role | Key Metric |
|---|---|---|---|---|---|
| Humain (Saudi Arabia) | Private | Sovereign | Sovereign-backed | Saudi government AI infrastructure company | Building 5 GW+ of AI data center capacity. Estimated $100B+ investment program. Part of Saudi Vision 2030 diversification. |
| G42 (UAE) | Private | Private | ~$10.0B (est.) | Abu Dhabi AI infrastructure and AI applications | Microsoft strategic investment ($1.5B, 2024). MGX sovereign fund affiliate. AI compute buildout in UAE. |
| Gulf Data Hub | Private | Private | Private | UAE data center operator | Dubai and Abu Dhabi campuses serving regional hyperscale demand. |
| Khazna Data Centers | Private | Private | Private | UAE sovereign DC operator | Managed by Mubadala Investment Company. Major Abu Dhabi campus. |
18.4.3 International Operators (Expanded)
| Company | Ticker | Exchange | Approx. Mkt Cap | Role | Key Metric |
|---|---|---|---|---|---|
| atNorth | Private | Private (40% Equinix) | Private | Nordic DC operator (Iceland, Sweden, Finland) | Equinix acquired 40% stake ($4B deal, Feb 2026). Renewable-powered AI compute. |
| Green Mountain | Private | Private | Private | Norwegian DC operator; hydropower-cooled | Rjukan facility in former industrial complex. 100% renewable. |
| ST Telemedia GDC | Private | Private (Temasek) | Private | Singapore-based DC operator; APAC + India | Temasek-backed. Major APAC presence. |
| AirTrunk | Private | Private (acquired by Blackstone/CPP Investments from Macquarie, Sep 2024) | AU$24B (acquisition value) | Australia/Asia-Pacific hyperscale DC | Blackstone and CPP Investments acquired AirTrunk from Macquarie for AU$24B. Largest DC deal globally in 2024. Australia, Singapore, Japan, Malaysia. |
| NEXTDC | NXT | ASX | Australian/APAC pure-play DC operator; AI-ready liquid cooling | 17 facilities operational + 15 in development; 28% mkt cap growth YoY; revenue growing 12.8% | |
| Keppel DC REIT | AJBU | SGX | First pure-play Asia DC REIT; 25 properties across 10 countries | 20.7% distributable income growth early 2026; Singapore, Japan, Europe exposure | |
| AdaniConneX | Private | Private (JV) | Private | Indian DC operator (Adani + EdgeConneX JV) | Building 1 GW+ across India. Chennai, Mumbai, Hyderabad campuses. |
| STACK Infrastructure | Private | Private (IPI Partners) | ~$30.0B (Asia valuation) | Global hyperscale DC developer | US, EMEA, APAC operations. Major powered land portfolio. |
18.5 Bottleneck Analysis
Available powered capacity (SEVERE): This is the central bottleneck. Hyperscalers want to deploy GPUs immediately, but facilities with available power and cooling take 2-5 years to build. CoreWeave’s $30.1 billion backlog 5 exists because customers want GPU compute but cannot get it fast enough. The gap between demand for AI compute capacity and physical data center capacity is the primary constraint on AI deployment speed.
Land with grid access (SEVERE): Finding sites with available power, fiber connectivity, water access, and permitting feasibility is increasingly difficult. The best sites (Northern Virginia, central Texas, central Ohio) are saturated. Secondary markets (Idaho, Louisiana, Oklahoma, West Texas) are emerging, but they often lack fiber infrastructure and skilled labor.
Capital intensity (MODERATE-HIGH): Building a 100 MW data center campus costs $1-2 billion 7. A 1 GW campus costs $10-20 billion. Only hyperscalers and well-capitalized REITs can fund projects at this scale. CoreWeave raised significant debt and equity (IPO plus debt facilities) to fund its expansion. The capital barrier limits competition and favors incumbents.
18.6 Risks
Hyperscaler vertical integration reduces colocation demand: If hyperscalers build all their own facilities, the addressable market for colocation REITs shrinks. Equinix and Digital Realty mitigate this by serving enterprise customers, offering interconnection services, and providing high-density AI-ready capacity that smaller customers cannot build themselves. But the largest AI clusters will increasingly be hyperscaler-owned.
Neocloud sustainability risk: CoreWeave’s $30.1 billion backlog is impressive, but the company is loss-making (Q2 2025 operating income was only $19 million on $1.2 billion revenue) 5 and heavily leveraged. If AI demand slows or customer concentration risk materializes (Microsoft is a major customer), CoreWeave could face financial stress. The neocloud model depends on sustained GPU demand at premium pricing.
Mining-to-AI execution and leverage risk: Bitcoin miners pivoting to AI hosting (IREN, WULF, HUT, CORZ, CIFR, KEEL, MARA, RIOT) collectively hold over $70B in announced AI/HPC contracts and control 15+ GW of secured or contracted power capacity. Their competitive advantage is real: existing power infrastructure, cooling systems, and permitting in place. However, the economics differ sharply from mining. AI hosting requires $8-15M per MW in buildout capital (vs. $0.7-1M for mining), meaning the capex intensity is 10-15x higher. Several companies are taking on unprecedented debt to fund the transition (IREN: $3.7B convertible 8; WULF: $5.7B total 9). If AI demand disappoints or hyperscaler contracts are delayed, these companies face debt service on facilities that cannot easily revert to mining at current BTC economics. The winners will be determined by speed to hyperscaler contract execution, capital efficiency, and power delivery timelines. By Q1 2026, TeraWulf’s HPC revenue exceeded mining for the first time, marking the inflection point for the sector.
Sovereign AI fragmenting the market: If every country insists on domestic AI compute, the market fragments into dozens of smaller national deployments rather than concentrating in optimal locations. This raises costs (suboptimal sites, duplicated infrastructure) and reduces economies of scale. For global operators like Equinix, fragmentation could be an opportunity (operating in 70+ markets) or a burden (customizing for each jurisdiction).
First principles check: Why can’t hyperscalers just build faster? Because data center construction is constrained by physical limits: concrete cures at the same rate regardless of who pours it, electrical permits take the same time regardless of who files them, and grid interconnection queues do not recognize corporate urgency. Money can buy more construction crews, but it cannot compress the critical path of a 24-month build.