Computing Infrastructure, Data Processing, and Web Hosting (NAICS 518210): An Investor's Primer

1. Overview

NAICS 518210 — its full 2022 title is Computing Infrastructure Providers, Data Processing, Web Hosting, and Related Services — is the federal statistical home of the physical and digital plumbing of the internet: the companies that own the computers, run the data centers, and rent out storage, processing power, and hosting to everyone else. (NAICS is the North American Industry Classification System, the standard the U.S. government uses to sort businesses by activity.) In plain terms, this is the cloud and data-center industry — the businesses that let a startup, a bank, a hospital, or an artificial-intelligence (AI) lab use large-scale computing without owning the machines. The Census Bureau defines the industry as establishments primarily providing computing infrastructure, data processing, web hosting, and related services — including Infrastructure as a Service (IaaS, renting raw servers and storage), Platform as a Service (PaaS, renting a ready-made software environment), cloud storage, application hosting, colocation, and technical support for streaming [1].

This has become one of the fastest-growing and most capital-hungry parts of the economy, because AI demand has turned computing capacity into a scarce, buildable asset. Combined 2026 capital spending (capex) by the largest cloud operators is guided toward roughly $700 billion, up from about $410 billion in 2025 [9]. That money flows into land, buildings, power, chips, and cooling — and the owners of that infrastructure collect the rent.

The central question for anyone putting money in is not simply revenue growth. It is whether an operator can secure power, build capacity on budget, fill it with creditworthy customers, keep utilization high, and earn an acceptable return on very heavy spending. There are two doors in:

• Public markets — the diversified technology giants whose cloud arms dominate the industry, the specialist data-center landlords (structured as real estate investment trusts, or REITs — companies that own income-producing property and pay out most profit as dividends), a growing set of AI-cloud and hosting specialists, and the listed asset managers that own private facilities.
• Private markets — infrastructure and private-equity funds, pension and sovereign investors, joint ventures, project finance, and private credit that funds data-center construction, plus direct stakes in privately held operators.

2. What it is and how it is structured

The industry sells computing capacity as a service. The National Institute of Standards and Technology (NIST) defines cloud computing as on-demand network access to a shared pool of configurable computing resources that can be rapidly provisioned and released [6]. In practice the code spans several overlapping layers:

• Hyperscale cloud — very large platforms (Amazon Web Services, Microsoft Azure, Google Cloud, Oracle Cloud) that sell computing, storage, databases, networking, and AI capacity on pay-as-you-go or contract terms.
• Colocation — renting customers space, power, cooling, security, and network connections inside a data center ("colocation" means you put your servers in their building). Equinix and Digital Realty are the public leaders.
• Wholesale data centers — large blocks of power and space leased to cloud companies and other big customers.
• "Neoclouds" — newer operators (CoreWeave, Nebius, Lambda, Crusoe) that specialize in renting graphics-processing-unit (GPU) capacity for AI training and inference.
• Web hosting, managed hosting, and data processing — the long tail: hosting websites and applications, operating infrastructure, and running customer data through batch processing, optical scanning, and related back-office services (GoDaddy, Newfold Digital, DigitalOcean, plus edge/delivery names like Cloudflare and Akamai).
• Edge and specialized services — smaller facilities near users to cut latency, plus streaming support, computer time rental, and cryptocurrency mining.

Ownership is mixed and skews toward those who can fund multi-billion-dollar builds: publicly traded tech giants, data-center REITs, and — increasingly — private-equity and infrastructure funds, pension and sovereign investors. Hyperscalers own substantial infrastructure but also lease from data-center operators; smaller hosts and resellers add fragmentation.

What the code excludes (this matters for reading the statistics). The boundary is drawn around hosting the infrastructure, not making the software [1]. Adjacent activities are classified elsewhere:

• 513210 Software Publishers — writing and licensing software (where most of Microsoft's and Oracle's software businesses sit).
• 541511 Custom Computer Programming and 541512 Systems Design / Integration — writing custom code and integrating systems without hosting them; 541513 covers on-site facility management.
• 517111 Wired Telecom Carriers and 517810 Other Telecom — the network carriers that move the bits between data centers.
• 516210 Streaming Distribution and content platforms — streaming distribution is carved out even though streaming support is included.
• 519290 Web Search Portals, 522320 Financial Transaction Processing, and 541214 Payroll Processing — search engines and specialized transaction/payroll processors.

3. How big it is

From our federal statistics:

Average pay near $147,700 confirms a high-wage, high-skill workforce [2]. Note the modest headcount (about 630,000) against the enormous capital deployed: the output is produced by machines and power, not people, so employment badly understates the industry's economic weight. A concentration ratio measures the share of revenue held by the largest firms; the Herfindahl-Hirschman Index (HHI), a finer concentration gauge, is suppressed in the federal data, so no HHI-based conclusion is appropriate.

The undercount caveat — read the concentration numbers carefully. A four-firm concentration ratio of only 37.2% looks surprisingly un-concentrated for an industry most people picture as a three-company cloud oligopoly [4]. Two things are going on:

1. The giants are largely classified elsewhere. Federal firm-based tabulations file each company under its primary activity, so most of AWS's revenue rolls up under Amazon (retail), Azure under Microsoft (software publishing), and Google Cloud under Alphabet (search/internet publishing). AWS alone generated roughly $129 billion in 2025 [13] — by itself nearly 40% of the entire measured industry's $329.5 billion of 2022 receipts, yet almost none of it is counted inside 518210. Add Azure and Google Cloud and the three hyperscalers' cloud revenue comfortably exceeds the whole measured industry. The federal figures therefore best describe the independent hosting, colocation, and data-processing firms — the long tail — while the giants sit in adjacent codes.
2. Coverage gaps. County Business Patterns covers only employer establishments with paid staff, excluding self-employed operators and most government facilities [3]; the Economic Census likewise omits nonemployer businesses. The supplied data does not quantify the missing share, so no adjustment is made here — but small operators and government-run infrastructure are understated.

For scale outside the code, private-market trackers put U.S. data-center revenue around $135 billion in 2025 [11] (U.S. colocation alone about $38.8 billion [12]), and global cloud-infrastructure spending in the region of $400 billion annualized, with the three largest providers holding roughly 60–65% of it [10] — a very different concentration picture from the federal 37.2%, precisely because these tabulations do count the hyperscalers.

4. The investable universe

NAICS is an activity classification, not a list of pure-play companies. Because the hyperscalers are embedded in diversified parents, few public stocks are "pure" plays on this industry; the closest are data-center REITs and AI-cloud specialists. Figures are approximate and change with the market — this is the section where tickers and scale belong.

(Nebius (NBIS) and CoreSite/American Tower are further examples with cloud/data-center exposure.)

Major private and other owners. A large share of U.S. data-center capacity is privately held, much of it via private equity and infrastructure funds:

• QTS — owned by Blackstone funds [39].
• CyrusOne — owned by KKR and Global Infrastructure Partners (GIP), a ~$15 billion take-private [40].
• Vantage Data Centers — backed by a consortium including DigitalBridge, Silver Lake, and the Public Sector Pension Investment Board (PSP) [41].
• DataBank, Switch, and Yondr — associated with DigitalBridge's digital-infrastructure portfolio [42].
• Flexential — backed by GI Partners and Morgan Stanley Infrastructure Partners [43].
• Aligned Data Centers — acquired by a consortium including the AI Infrastructure Partnership, MGX, and BlackRock's GIP [44].
• Among neoclouds, Lambda and Crusoe remain private.

Public investors can also reach private facilities indirectly through the listed managers that own them — BlackRock (BLK), Blackstone (BX), KKR (KKR), and DigitalBridge (DBRG). Private ownership is typically structured through funds and joint ventures, so leverage, customer contracts, and project economics are far less transparent than for public companies.

5. How the money works

Owners make money by converting capital and power into rentable capacity, then keeping it full at a healthy spread over its cost. REIT economics genuinely apply here (Equinix and Digital Realty are REITs); utility rate-base or mining-cost frameworks do not. The models differ by layer:

• Colocation / data-center REITs behave like specialized landlords: long-term leases, recurring monthly revenue, occupancy, tenant churn, and re-leasing spreads. Because they are REITs, the headline profitability metric is funds from operations (FFO) — roughly, cash earnings after adding back property depreciation — most of which is distributed as dividends [18][19]. The moat is interconnection density: the more networks and customers meet inside a facility, the more valuable a spot there becomes.
• Hyperscale and neocloud sell capacity on consumption (pay-as-you-go) or committed terms, so the metrics are gross margin, operating leverage as fixed data centers fill up, net revenue retention (do existing customers spend more?), and backlog / remaining performance obligations (RPO) — contracted future revenue. Neoclouds like CoreWeave run on large take-or-pay contracts (the customer pays whether or not it uses the capacity) with a few big AI buyers, turning backlog into near-bond-like cash flow — but concentrating risk. Note that RPO is contracted, not realized: it can be delayed, renegotiated, or canceled [20].
• Web hosting and managed services are subscription businesses: recurring revenue, average revenue per user (ARPU), churn, and gross margin, with growth from cross-selling domains, security, and tools; less real-estate-heavy, more service-quality-dependent.

A few unifying measures cut across layers:

• Capacity is priced in power, not floor space. Electricity — to run and cool the chips — is the binding constraint, so contracted and energized megawatts (MW) and gigawatts (GW), revenue per megawatt, and how much capacity is actually leased and occupied are the core levers.
• Power Usage Effectiveness (PUE) — total facility power divided by the power reaching the computing hardware — measures efficiency; a lower PUE means less electricity wasted on cooling, directly improving margins.
• Uptime, outage history, and service-level-agreement (SLA) performance, plus construction cost and time-to-energize, determine whether a well-located site actually earns.

The whole model is capital-intensive and front-loaded: operators spend billions before revenue arrives, then depreciate the assets. When utilization is high and financing is cheap, returns are attractive; when capacity is overbuilt, rates rise, or GPUs obsolesce, the same leverage cuts the other way. A facility can have strong demand yet weak returns if construction, interest, power, or depreciation outrun revenue. The single most important judgment is whether contracted demand will fill the capacity being built.

6. What drives demand

• Artificial intelligence. Training and running large AI models is now the dominant driver — the reason hyperscaler capital budgets are heading toward ~$700 billion in 2026, why neoclouds exist, and why power has become the scarce input. Roughly three-quarters of hyperscaler capex is tied directly to AI infrastructure [9].
• Cloud migration. Enterprises keep shifting workloads from their own server rooms to rented cloud — a multi-year "digital transformation" with room to run, especially in regulated finance, healthcare, and government.
• Data growth, analytics, streaming, and cybersecurity. More devices, video, software-as-a-service, and connected sensors generate data that must be stored, processed, and protected.
• Power availability — increasingly the ceiling. The Department of Energy's Lawrence Berkeley National Laboratory (LBNL) estimated U.S. data centers used 4.4% of national electricity in 2023, and projects a central estimate of 11.8% by 2030 (a range of 9.5%–15.3%) [7]; the International Energy Agency corroborates a steep 2025 surge in demand [8]. Many announced AI projects are now gated by grid connections, transformers, and cooling rather than by chips — Futurum estimates about 40% face power-related delays [9]. Where you can get electricity now shapes where the industry can grow.

Judgment: AI should support strong long-term demand, but demand will not translate evenly into returns. Operators with secured, energized power and repeatable construction should outperform projects that hold only land or a speculative customer pipeline. Demand is also cyclical and increasingly tied to the AI capital cycle — a strength while it expands, a concentrated risk if AI spending pauses.

7. Regulation

There is no single regulator; the industry sits at the intersection of several regimes, mostly a patchwork of federal, state, and local rules.

• Land, power, and environmental approvals. Data centers face zoning, building, fire, water, utility, and environmental requirements. Backup generators and other stationary equipment can trigger Environmental Protection Agency (EPA) emissions standards and Clean Air Act permitting [32]. This is now the most active front: between May 2024 and June 2025 at least 36 U.S. projects were delayed or blocked, disrupting an estimated $162 billion of investment [35]; several states are weighing moratoriums (Maine's governor vetoed a statewide bill in April 2026) [35]. In July 2025 a federal executive order sought to streamline permitting for facilities drawing more than 100 MW, creating tension with stricter state and local rules [36].
• Health data. Under the Health Insurance Portability and Accountability Act (HIPAA), a cloud provider that stores or processes electronic protected health information becomes a "business associate" — even when the data are encrypted and it holds no decryption key — and must sign a Business Associate Agreement (BAA) and meet security safeguards [30].
• Government cloud. Selling covered services to federal agencies generally requires FedRAMP (Federal Risk and Authorization Management Program) authorization, a standardized cloud-security assessment [31].
• Competition. The Federal Trade Commission (FTC) has examined switching costs, software licensing, data-transfer ("egress") fees, security, and single points of failure in cloud [33], and has flagged possible competition concerns in partnerships between large cloud providers and AI developers, including access to scarce computing and higher switching costs [34].
• Privacy, cybersecurity, and export controls. The U.S. has no comprehensive federal privacy law, so operators navigate a growing set of state privacy statutes and sectoral rules that can dictate where data may be stored. Separately, U.S. export controls on advanced AI chips indirectly shape which operators can obtain capacity and where. For private investors, permitting and utility contracts can matter as much as the technology.

8. Competitive dynamics and consolidation

The industry is tiered and consolidating. At the top, hyperscale cloud is a tight oligopoly: the three largest providers together hold roughly 60–65% of global cloud-infrastructure spending [10], protected by scale, switching costs, "data gravity" (data is expensive to move once it accumulates), and ecosystem lock-in. Scale matters but is not sufficient — a large operator without power, usable capacity, or competitive pricing can underperform a smaller one with better sites and contracts.

Below them, the data-center layer is fragmenting between two well-capitalized public landlords (Equinix, Digital Realty) and a wave of private-equity roll-ups. M&A hit an all-time record in 2024, with disclosed private-equity data-center spending of at least $115 billion across ~95 deals — nearly double 2022–2023 combined [37]. Landmark transactions include Blackstone's ~$10 billion purchase of QTS, KKR/GIP's ~$15 billion take-private of CyrusOne, and large investments into Vantage and Aligned [37][38]. The logic: data centers throw off long-term, contracted, often inflation-linked cash flows that suit infrastructure funds and their low-cost capital. Nearly half of the top 25 U.S. data-center owners are now owned by, or in joint ventures with, private-equity firms [38].

The newest entrants, the neoclouds, compete on GPU availability and speed to deploy, and have introduced a much-watched circular-financing pattern — chipmakers, AI labs, and neoclouds investing in and contracting with one another — that amplifies both growth and fragility [20]. At the bottom, web hosting is fragmented and slowly consolidating; even the largest players hold only low-single-digit shares of a global market worth well over $100 billion.

Judgment: consolidation should continue, because power, capital, compliance, and interconnection favor scaled platforms. But higher leverage and customer concentration can make an apparently stable infrastructure asset fragile.

9. Principal risks

• AI overbuild / demand air-pocket. Capacity is being built against forecasts and funded heavily with debt. If AI monetization disappoints or demand pauses, expensive, half-full data centers become a drag — the classic capital-cycle risk, magnified by leverage.
• Power cost and availability. Electricity is the binding constraint; interconnection queues can delay revenue for years, and rising grid demand can raise prices for local residents — a growing political flashpoint.
• Community and regulatory backlash. Moratoriums, zoning fights, air/water restrictions, and noise complaints can strand projects [35].
• Customer concentration. Neoclouds especially lean on a handful of large AI buyers; loss or renegotiation of one contract can be existential. RPO and backlog can be canceled or delayed [20].
• Hardware obsolescence. GPUs and networking gear can lose economic value before facilities are fully depreciated; a more efficient chip generation can undercut capacity bought at today's prices.
• Financing and rate risk. The debt-heavy buildout is sensitive to interest rates and capital-market conditions.
• Cybersecurity and outages. Breaches, ransomware, and physical failures can trigger SLA penalties, lost customers, and reputational damage.
• Disclosure and structure risk. Public conglomerates may not break out cloud economics; private owners disclose even less, and exposure there is illiquid and often highly levered.

10. How to invest and the outlook

Public-market routes. First choose the exposure you want:

• Cloud-platform parents (AMZN, MSFT, GOOGL, ORCL, IBM) — diversified growth with strategic AI exposure, but you buy the whole company, and segment transparency is limited [13][14][15][16][17].
• Data-center REITs (EQIX, DLR; IRM adjacent) — landlord-style, dividend-paying exposure to the physical buildout, sensitive to interest rates and capital costs [18][19][22].
• AI-cloud / high-density developers (CRWV, APLD; peers such as NBIS) — higher growth, higher leverage, customer-concentration and technology risk [20][23].
• Managed cloud, hosting, and edge (RXT, GDDY, DOCN, AKAM, NET, FSLY, SNOW) — more service/software exposure with less direct ownership of large facilities [21][24][25][26][27][28][29].
• Listed alternative managers (BLK, BX, KKR, DBRG) — indirect exposure to privately owned platforms and infrastructure funds. Thematic exchange-traded funds bundle several of these "digital infrastructure" names together.

Private-market routes. Because much of the best capacity is privately held, allocators reach it through infrastructure and private-equity funds, direct or co-investment stakes in operators (Vantage, Aligned, Switch, QTS, DataBank), private credit that finances construction, equipment financing, and data-center real estate. Venture capital is the route into earlier-stage neoclouds. These offer concentrated, income-generating exposure but are illiquid, disclosure-light, and leverage-dependent — and have recently drawn Congressional scrutiny over their effect on local utility bills [46].

A practical diligence checklist (public or private): Is power merely planned, or contracted and energized? How much capacity is leased, occupied, and generating revenue? Who are the customers, what are their credit ratings and termination rights, and are contracts take-or-pay, usage-based, or cancellable? How much revenue rests on one customer? What are PUE, cooling, water, and backup-power needs, and the related environmental exposure [45]? What is the cost and timeline to finish each project? How old is the GPU fleet? Can debt service be covered if utilization or pricing falls?

Outlook (forward-looking judgment, not fact). The near-term story is dominated by the AI capital cycle: capacity, power, and financing are being added at an unprecedented pace, and demand — for now — is racing to keep up. The questions to track are (1) whether AI revenue grows fast enough to justify ~$700-billion-a-year capital budgets, (2) whether power and permitting keep pace or throttle growth, and (3) whether the debt funding the buildout stays cheap. Mature colocation may offer steadier cash flows; AI cloud offers greater upside but greater execution and financing risk. The likely winners hold secured power, dense connectivity, strong counterparties, disciplined capital spending, and manageable leverage — not just the AI label. Either way, computing infrastructure has moved from a back-office cost line to one of the defining capital-formation stories of the decade, and the federal statistics — anchored on independent firms and a 2022 revenue snapshot — will keep understating just how large it has become.

---

Sources

1. U.S. Census Bureau, "2022 NAICS: 518210 Computing Infrastructure Providers, Data Processing, Web Hosting, and Related Services." https://www.census.gov/naics/?details=518210&input=518210&year=2022
2. U.S. Census Bureau, County Business Patterns 2023 (establishments 18,544; employment 629,527; annual payroll $92.96B; Q1 payroll $24.05B). https://www.census.gov/data/datasets/2023/econ/cbp/2023-cbp.html
3. U.S. Census Bureau, County Business Patterns Methodology (employer-only coverage; excludes nonemployers, self-employed, most government). https://www.census.gov/programs-surveys/cbp/technical-documentation/methodology.html
4. U.S. Census Bureau, 2022 Economic Census — Concentration of Largest Firms, NAICS 518210 (receipts $329.47B; 12,054 firms; CR4 37.2%, CR8 41.9%, CR20 50.3%, CR50 61.2%; HHI suppressed). https://data.census.gov/table/ECNSIZE2022.EC2200SIZECONCEN?q=EC2200SIZECONCEN
5. U.S. Small Business Administration, "Table of Size Standards" (NAICS 518210 = $40 million average annual receipts), 2023. https://www.sba.gov/document/support-table-size-standards
6. National Institute of Standards and Technology, "The NIST Definition of Cloud Computing" (SP 800-145). https://csrc.nist.gov/pubs/sp/800/145/final
7. Lawrence Berkeley National Laboratory (DOE), "United States Data Center Energy Usage Report: 2025 Update" (4.4% of U.S. electricity in 2023; central 11.8% by 2030; range 9.5%–15.3%). https://eta-publications.lbl.gov/publications/united-states-data-center-energy-2025
8. International Energy Agency, "Data centre electricity use surged in 2025." https://www.iea.org/news/data-centre-electricity-use-surged-in-2025-even-with-tightening-bottlenecks-driving-a-scramble-for-solutions
9. Futurum Group, "AI Capex 2026: The Infrastructure Sprint" (2026 capex ~$700B; ~$410B in 2025; ~75% AI-tied; ~40% of projects power-delayed). https://futurumgroup.com/insights/ai-capex-2026-the-690b-infrastructure-sprint/
10. Synergy Research Group / Canalys, cloud market-share tracking, 2025 (global cloud-infrastructure spending ~$400B annualized; Big Three ~60–65%). https://www.srgresearch.com/
11. Precedence Research, "Data Center Market Size" (U.S. ~$134.77B in 2025). https://www.precedenceresearch.com/data-center-market
12. MarketsandMarkets, "US Data Center Colocation Market 2025–2030" (U.S. colocation ~$38.8B in 2025). https://www.marketsandmarkets.com/Market-Reports/us-data-center-colocation-market-251053035.html
13. Amazon.com, Form 10-K for FY ended December 31, 2025 (AWS segment ~$129B). https://www.sec.gov/Archives/edgar/data/1018724/000101872426000004/amzn-20251231.htm
14. Microsoft, 2025 Annual Report (Azure / Microsoft Cloud). https://www.microsoft.com/investor/reports/ar25/
15. Alphabet, Form 10-K for FY ended December 31, 2025 (Google Cloud). https://www.sec.gov/Archives/edgar/data/1652044/000165204426000018/goog-20251231.htm
16. Oracle, "Record Q4 and FY 2026 Results" (Oracle Cloud Infrastructure growth; large RPO). https://www.oracle.com/news/announcement/q4fy26-earnings-release-2026-06-10/
17. IBM, 2025 Form 10-K. https://www.sec.gov/Archives/edgar/data/51143/000005114326000010/ibm-20251231_d2.htm
18. Equinix, Form 10-K FY2025 (~$9.2B revenue; ~77% cabinet utilization; ~$1.6B annualized gross bookings). https://www.sec.gov/Archives/edgar/data/1101239/000110123926000032/eqix-20251231.htm
19. Digital Realty Trust, Form 10-K FY2025 (~$6.1B revenue; 310 data centers, 118 in the U.S.). https://www.sec.gov/Archives/edgar/data/1297996/000110465926015365/dlr-20251231x10k.htm
20. CoreWeave, Form 10-K FY2025 (~$5.1B revenue; ~$60.7B RPO; IPO March 2025). https://www.sec.gov/Archives/edgar/data/1769628/000176962826000104/crwv-20251231.htm
21. Rackspace Technology, Q4 and Full-Year 2025 Results. https://www.sec.gov/Archives/edgar/data/1810019/000181001926000005/rxt991_q42025.htm
22. Iron Mountain, 2025 Form 10-K (data-center segment). https://www.sec.gov/Archives/edgar/data/1020569/000102056926000013/irm-20251231.htm
23. Applied Digital, Annual Reports. https://ir.applieddigital.com/sec-filings/annual-reports
24. Akamai Technologies, FY2025 Financial Results (~$4.2B revenue). https://www.ir.akamai.com/financial-information/annual-reports
25. Fastly, Q4 and Full-Year 2025 Financial Results. https://investors.fastly.com/news-releases/news-release-details/fastly-announces-both-record-fourth-quarter-and-full-year-2025
26. Snowflake, FY2025 Financial Results (~$3.63B revenue). https://www.snowflake.com/en/news/press-releases/snowflake-reports-financial-results-for-the-fourth-quarter-and-full-year-of-fiscal-2025/
27. Cloudflare, Form 8-K Q2 2025 (revenue $512.3M; >$2B annualized). https://www.sec.gov/Archives/edgar/data/1477333/000147733325000136/q225exhibit991.htm
28. GoDaddy revenue (~$5.0B 2025), Macrotrends compilation of filings. https://www.macrotrends.net/stocks/charts/GDDY/godaddy/revenue
29. DigitalOcean Holdings, SEC filing FY2025 (~$0.9B revenue). https://www.sec.gov/Archives/edgar/data/1582961/000158296125000084/a2025-q1dopressrelease.htm
30. U.S. Department of Health and Human Services, "Guidance on HIPAA and Cloud Computing" (business-associate status; encrypted data). https://www.hhs.gov/hipaa/for-professionals/special-topics/health-information-technology/cloud-computing/index.html
31. FedRAMP, "Is FedRAMP Right for You?" https://www.fedramp.gov/docs/rev5/playbook/csp/authorization/getting-started/
32. Environmental Protection Agency, "Clean Air Act Resources for Data Centers." https://www.epa.gov/stationary-sources-air-pollution/clean-air-act-resources-data-centers
33. Federal Trade Commission, "Cloud Computing RFI: What We Heard and Learned," 2023. https://www.ftc.gov/policy/advocacy-research/tech-at-ftc/2023/11/cloud-computing-rfi-what-we-heard-learned
34. Federal Trade Commission, "Staff Report on AI Partnerships and Investments Study," 2025. https://search.ftc.gov/news-events/news/press-releases/2025/01/ftc-issues-staff-report-ai-partnerships-investments-study
35. Built In / MultiState, "States Push Data Center Moratoriums" (36 projects delayed May 2024–June 2025, ~$162B; Maine veto April 2026). https://builtin.com/articles/state-data-center-moratoriums
36. MultiState, "Federal AI Data Center Policy Meets Resistance" (July 2025 executive order for facilities >100 MW). https://www.multistate.us/insider/2026/4/14/federal-ai-data-center-policy-meets-resistance-from-state-lawmakers
37. Synergy Research Group, "Data Center M&A Deals Broke All Records in 2024" (~$115B / ~95 deals). https://www.srgresearch.com/articles/its-official-data-center-ma-deals-broke-all-records-in-2024
38. STL Partners, "Top Data Centre Investors 2025" (private-equity ownership of the top-25 U.S. data centers). https://stlpartners.com/articles/data-centres/top-data-centre-investors-2025/
39. Blackstone, "Blackstone Funds Complete Acquisition of QTS Realty Trust," 2021. https://www.blackstone.com/news/press/blackstone-funds-complete-acquisition-of-qts-realty-trust/
40. Global Infrastructure Partners, "KKR and GIP Complete Acquisition of CyrusOne," 2022. https://www.global-infra.com/news/kkr-and-gip-complete-acquisition-of-cyrusone/
41. Public Sector Pension Investment Board, "PSP Investments Completes Sale of Majority of Its Investment in Vantage Data Centers," 2024. https://www.investpsp.com/en/news/psp-investments-completes-sale-of-majority-of-its-investment-in-vantage-data-centers/
42. DigitalBridge, "Portfolio" (DataBank, Switch, Yondr). https://www.digitalbridge.com/portfolio
43. GI Partners, "Flexential." https://www.gipartners.com/private-equity/portfolio/peak-10
44. Aligned Data Centers, "AIP, MGX and BlackRock's GIP Complete Acquisition of Aligned Data Centers," 2026. https://aligneddc.com/
45. Li et al., "Making AI Less 'Thirsty': Uncovering and Addressing the Water Footprint of AI Models," arXiv, 2023. https://arxiv.org/pdf/2304.03271
46. U.S. Senate Committee on Banking, Housing, and Urban Affairs, "Warren Probes Major Private Equity Firms on Investments in Data Centers as Utility Costs Rise," 2025. https://www.banking.senate.gov/newsroom/minority/warren-probes-major-private-equity-firms-on-investments-in-data-centers-as-utility-costs-rise