Global Data Center Market Analysis: The AI-Driven Infrastructure Boom

Executive Summary

The global data center market has entered a phase of investment unlike anything in the history of technology infrastructure. In 2025, the five largest hyperscalers – Amazon, Alphabet, Microsoft, Meta, and Oracle – spent approximately $443 billion on capital expenditure, a 73% increase from $256 billion in 2024. For 2026, that figure is projected to reach $602-690 billion, with roughly 75% directed toward AI compute infrastructure. Amazon alone plans to spend $200 billion in 2026, while Alphabet has guided $175-185 billion, Meta $115-135 billion, and Microsoft is tracking toward $120 billion or more.

This spending surge is remaking the physical infrastructure of the internet. North American data center vacancy rates have held at or below 1.6% for two consecutive years, while colocation pricing has doubled from $70-80 per kilowatt to $150-160 per kilowatt. Global data center electricity consumption reached approximately 415 terawatt-hours (TWh) in 2024, according to the International Energy Agency, and is projected to double to 945 TWh by 2030. The United States consumed 183 TWh of data center electricity in 2024, representing more than 4% of total national electricity consumption.

Colocation generated $104.2 billion in revenue in 2025 and is projected to reach $204.4 billion by 2030. Equinix, the largest operator, posted $9.2 billion in 2025 revenue. Digital Realty reported $6.1 billion. Iron Mountain’s data center revenue surged 39% in Q4 2025 alone.

Power procurement has become the binding constraint on growth. Big tech companies signed more than 10 gigawatts of new U.S. nuclear capacity commitments in 2025, while data center operators accounted for 43% of all clean energy power purchase agreements signed in 2024. Transformer lead times have stretched to 128-144 weeks, and prices have risen 45-95% since 2019. At least 16 GW of new data center capacity is slated to come online globally in 2026, but analysts estimate that 30-50% of that pipeline will not materialize on schedule due to power constraints and construction delays.

Introduction

$690 billion. That is the upper-end estimate for what the world’s largest technology companies plan to spend on data center infrastructure in 2026, according to Futurum Group. To put that figure in perspective, it exceeds the 2025 GDP of Switzerland and approaches the entire U.S. defense budget. The scale of this investment wave reflects a single, dominant conviction: that AI compute capacity is the most important strategic asset a technology company can own, and that the companies that secure the most capacity today will hold the strongest competitive positions for the next decade.

The data center industry was already large before generative AI. Cloud computing, streaming video, e-commerce, and SaaS applications had driven steady growth for years. But the emergence of large language models, multimodal AI systems, and AI agents has fundamentally changed the demand curve. Training a single frontier AI model now requires tens of thousands of GPUs running for months. Inference – serving those models to billions of users – requires continuous compute capacity that scales with every query, every interaction, every automated workflow.

This report examines the global data center market as of early 2026. It covers market sizing, hyperscaler investment patterns, the colocation and multi-tenant segment, power and sustainability challenges, regional construction pipelines, technology trends including cooling and edge computing, and the principal risks facing the industry. All statistics cited are drawn from publicly available company filings, government agency reports, industry associations, and credible research firms.

Market Overview

Market Size and Growth

Grand View Research valued the global data center market at $383.82 billion in 2025 and projects $425.28 billion for 2026. Precedence Research provides a similar estimate of $386.71 billion for 2025, forecasting growth to approximately $430 billion in 2026 and $1.1 trillion by 2035 at a compound annual growth rate (CAGR) of 11.06%. Gartner, using a broader definition that includes all data center systems spending, reported $489.5 billion in 2025, a 46.8% increase from $333.4 billion in 2024.

Discrepancies between these figures reflect methodological differences. Narrower definitions focus on physical infrastructure (real estate, power, cooling, connectivity). Broader definitions fold in server hardware, storage, networking equipment, and software. What all estimates agree on is the direction and magnitude of growth: the market is expanding at double-digit rates, accelerating from pre-2023 trends, and AI is the primary catalyst.

Demand Drivers

AI workloads have reshaped what “demand” means in the data center industry. A single NVIDIA Blackwell GB300 rack draws 163 kilowatts. The next-generation Vera Rubin NVL144 system, expected in 2026, may require more than 300 kW per rack. Rubin Ultra NVL576 configurations, projected for 2027, could exceed 600 kW per rack. These power densities are 10-30 times higher than traditional enterprise compute racks, which typically draw 7-15 kW.

McKinsey estimates that demand for AI-ready data center capacity will rise at an average rate of 33% per year between 2023 and 2030. By 2030, approximately 70% of total demand for data center capacity will be for facilities equipped to host advanced AI workloads. This represents a structural shift: most new data center builds are now designed around AI workloads rather than general-purpose compute.

Figure 1: Combined capital expenditure of the five largest hyperscalers (Amazon, Alphabet, Microsoft, Meta, Oracle) from 2022 to 2026. The 2026 figure represents the midpoint of current guidance ranges. Spending has increased roughly fivefold in four years, driven overwhelmingly by AI infrastructure investment. Sources: Company earnings reports; IEEE ComSoc Technology Blog; Futurum Group.

Hyperscaler Investment

Amazon Web Services

$200 billion. That is what Amazon plans to spend on capital expenditure in 2026, the majority directed toward AWS data centers. CEO Andy Jassy told investors that “the capital we’re spending and intend to spend this year is predominantly in AWS.” The figure represents a 52% increase from $131.8 billion in 2025, which itself was up from approximately $83 billion in 2024 – a roughly 140% increase over three years.

AWS disclosed a backlog of $244 billion at year-end 2025, growing materially faster than reported revenue. That backlog represents multi-year contractual commitments from enterprises and AI-native firms and effectively underwrites the 2026 infrastructure buildout. In 2025, $96.5 billion of Amazon’s total capex was directed specifically at data center infrastructure, covering facilities, servers, networking equipment, and custom AI chips including the Trainium series.

Alphabet/Google

Alphabet’s 2026 capital expenditure guidance of $175-185 billion represents what may be the most aggressive year-over-year increase among the hyperscalers. The company spent $91-93 billion in 2025, after raising its guidance three times during the year – from an initial $75 billion estimate in Q1 to $85 billion in Q2 to the final $91-93 billion range. Quarterly spending accelerated throughout 2025: $17.2 billion in Q1, $22.4 billion in Q2, $24 billion in Q3, and $27.9 billion in Q4.

Two-thirds of Alphabet’s capex goes toward GPUs, TPUs, and servers. The remaining third funds new data center campuses, high-voltage substations, and optical fiber networks. Google Cloud’s backlog surged 55% sequentially in Q4 2025, doubling year-over-year to $240 billion. The company reported signing more deals above $1 billion through Q3 2025 than in the previous two years combined.

Microsoft

Microsoft spent $64.6 billion on capital expenditure in fiscal year 2025 (ending June 2025) and is on track to spend $120 billion or more in fiscal year 2026. Q1 FY2026 capex was $34.9 billion, including $11.1 billion on data center leases alone. Q2 FY2026 capex reached $37.5 billion, with roughly two-thirds directed at short-lived assets, primarily GPUs and CPUs. On an annualized basis, that quarterly rate implies approximately $150 billion.

By the end of fiscal year 2025, Microsoft operated more than 400 data centers and had commissioned approximately 2 GW of new capacity during the year. CEO Satya Nadella stated that the company expects to increase AI capacity by more than 80% throughout fiscal year 2026 and to “roughly double” its data center footprint over the following two years.

Meta

Meta’s 2026 capex guidance of $115-135 billion represents a near-doubling from $72.22 billion in 2025. CEO Mark Zuckerberg has stated that the company expects to spend at least $600 billion on U.S. data centers and related infrastructure by 2028. The 2026 spending would push Meta’s capex intensity to 55-67% of projected revenue, a ratio without precedent among profitable technology companies of this scale.

Meta’s infrastructure strategy extends beyond traditional capex. The company carries approximately $30 billion in off-balance-sheet commitments through AI-related operating leases structured via special purpose vehicles. A $27 billion deal with Nebius for additional AI compute capacity further expands the company’s infrastructure footprint beyond its own facilities.

Financing the Buildout

The scale of hyperscaler investment now exceeds what internal cash generation can finance. Bank of America estimates that the largest cloud companies will spend approximately 90% of their operating cash flow on capital expenditure in 2026, up from 65% in 2025. Hyperscalers raised $108 billion in debt during 2025, and Morgan Stanley projects that hyperscaler borrowing will top $400 billion in 2026, more than double the $165 billion borrowed in 2025. Some forecasts project $1.5 trillion in cumulative hyperscaler debt issuance over the coming years.

This financial dynamic introduces a new dimension of risk. Unlike software development, which scales efficiently once built, physical infrastructure requires continuous spending that cannot be easily reversed. If AI revenue growth disappoints or enterprise adoption slows, these companies will be carrying massive fixed-cost structures and debt loads.

Colocation and Multi-Tenant Data Centers

Market Size and Growth

$104.2 billion. That was the global data center colocation market in 2025, according to MarketsandMarkets. The firm projects this figure will reach $204.4 billion by 2030, reflecting a 14.4% CAGR. In the United States specifically, the colocation market generated $38.80 billion in 2025 and is expected to reach $65.44 billion by 2030 at an 11.0% CAGR. The gap between U.S. and global growth rates reflects the faster expansion of colocation markets in Asia-Pacific, the Middle East, and Latin America.

Major Operators

Equinix remains the largest colocation provider globally, operating more than 260 data centers across 70+ metros in 33 countries. Full-year 2025 revenue was $9.217 billion, a 5.36% increase from 2024. Q4 2025 revenue reached $2.420 billion, a 7.03% year-over-year increase. Annualized gross bookings hit $1.6 billion, with interconnection revenue growing 9% year-over-year on a normalized and constant currency basis. For 2026, Equinix guides total revenue of $10.123-10.223 billion, representing 10-11% growth.

Digital Realty posted $6.112 billion in 2025 revenue with adjusted EBITDA of $3.34 billion. Q4 2025 revenue was $1.6 billion, a 14% year-over-year increase. Hyperscale leasing exceeded $800 million across the year, with the Americas producing 65% of Q4 bookings. The company guides 2026 revenue of $6.6-6.7 billion and adjusted EBITDA of $3.6-3.7 billion.

Iron Mountain has aggressively pivoted toward data centers, with 2025 revenue of $6.90 billion (up 12.23% year-over-year). Data center revenue surged 39% in Q4 2025, and the company’s 2026 guidance of $7.625-7.775 billion includes plans to lease more than 100 MW of new data center capacity.

American Tower generated $10.645 billion in 2025 revenue (up 5.1%) with net income of $2.629 billion (up 15.3%). While its core business remains cell towers, the company is expanding its data center portfolio to capture AI-driven demand.

Vacancy and Pricing

North American data center vacancy rates have reached historic lows. JLL’s year-end 2025 report recorded vacancy at 1% for the second consecutive year. CBRE’s H1 2025 data showed 1.6% across North America, with Northern Virginia – the world’s largest data center market – at just 0.76% in Q1 2025.

Pricing has responded accordingly. Colocation rates have doubled from $70-80 per kilowatt to $150-160 per kilowatt in major markets. Some markets have seen pricing reach $200+ per kilowatt per month, comparable to 2011-2012 peak levels. Lease terms have extended from the traditional five-to-seven-year range to more than ten years, as operators lock in long-term commitments from hyperscale tenants. Eight gigawatts of colocation capacity is currently under construction across North America, and 73% of that capacity is already preleased.

Figure 2: Full-year 2025 revenue for the four largest publicly traded data center operators. American Tower’s revenue includes its broader tower portfolio. Iron Mountain’s data center segment is growing fastest (39% in Q4 2025), reflecting its aggressive pivot toward digital infrastructure. Sources: Company earnings reports (Q4 2025).

Power and Sustainability

Electricity Consumption

415 TWh. That was global data center electricity consumption in 2024, according to the International Energy Agency, representing approximately 1.5% of total global electricity use. Data center power consumption has grown at 12% per year over the past five years. The IEA projects this figure will reach approximately 945 TWh by 2030, nearly doubling from 2024 levels and representing close to 3% of global electricity consumption. From 2024 to 2030, data center electricity consumption is expected to grow by about 15% per year – more than four times faster than total electricity consumption growth from all other sectors combined.

In the United States, where the majority of the world’s hyperscale data centers are concentrated, the numbers are starker. U.S. data centers consumed 183 TWh in 2024, more than 4% of national electricity. The IEA projects U.S. data center demand will exceed 250 TWh by 2026. The Department of Energy estimates that by 2030, U.S. data center consumption could reach 426 TWh, a 133% increase from 2024. Gartner’s November 2025 forecast projects global data center electricity demand growing 16% in 2025 alone, then doubling by 2030.

Grid Strain and the Transformer Crisis

Power procurement has overtaken real estate and IT equipment as the primary bottleneck for data center construction. In a 2025 industry survey, 72% of respondents identified power and grid capacity as “very or extremely challenging.” The constraint has migrated from the server rack to the electrical substation: U.S. interconnection queues now delay projects by years, and utility providers in multiple regions have warned of capacity shortages beginning in 2026.

Large power transformer lead times have stretched to 128 weeks on average, with generation step-up units (GSUs) averaging 144 weeks as of Q2 2025. Three years ago, these lead times were under 50 weeks. Only 20% of large transformers used in the United States are manufactured domestically. Prices have surged accordingly: power transformer unit prices are up 77% since 2019, with some classes of distribution transformers rising as much as 95%. Wood Mackenzie expects the pad-mount three-phase transformer shortage to worsen further as data centers, manufacturing facilities, and EV charging infrastructure all compete for the same equipment. Nearly $1.8 billion in announced North American transformer manufacturing expansions are underway, but new capacity takes years to commission.

Nuclear Energy

Big tech companies signed more than 10 GW of new U.S. nuclear energy capacity commitments in 2025, making nuclear power the most consequential new energy source for the data center industry. Microsoft’s 20-year agreement to restart Three Mile Island’s Unit 1 reactor targets 835 MW of generation capacity, with the $1.6 billion revamp originally scheduled for 2028 but accelerated to 2027. Google signed the first U.S. corporate small modular reactor (SMR) fleet deal with Kairos Power, covering 500 MW of capacity targeted for the early 2030s, and separately agreed to a power purchase agreement for 50 MW from Kairos’s Hermes 2 plant.

Amazon has invested more than $20 billion to convert its Susquehanna site into an AI data center campus and signed an agreement with Energy Northwest to develop four advanced SMRs, generating roughly 320 MW in the first phase with the option to scale to 960 MW total. Meta issued a request for proposals for 1-4 GW of new nuclear generation and signed a deal in 2025 to source energy from an existing Illinois nuclear plant. Deloitte projects that nuclear energy could meet up to 10% of data center electricity demand by 2035.

Renewable Energy

Data center operators accounted for 43% of all clean energy power purchase agreements (PPAs) signed globally in 2024. Amazon, Microsoft, Google, Meta, and Apple have collectively contracted approximately 50 GW of clean energy capacity to date, making them the largest corporate buyers of renewable energy in the world.

Several deals signed in 2025 stand out for their scale. Google announced a $4.75 billion acquisition of Intersect Power in December 2025, moving beyond power procurement toward direct control of energy generation and development. Meta and NextEra Energy Resources signed approximately 2.5 GW of clean energy contracts across 13 projects (11 PPAs and 2 energy storage agreements), scheduled to come online between 2026 and 2028. Microsoft signed an eight-year deal with Qcells to deploy 12 GW of solar energy, with the first phase expected to come online between 2026 and 2027.

In Europe, approximately one-third of the PPA market is now dedicated to data center operators. The largest European data center PPAs in 2025 were signed in Italy (568 MW), Finland (472 MW), and Spain (314 MW). A growing number of hyperscalers are shifting from annual renewable energy credit matching to 24/7 or time-matched clean energy procurement, where clean power is sourced to match actual energy use every hour of the day.

Water Consumption

Water use is emerging as a second front in the environmental debate around data centers. Data centers typically lose about 80% of the water they withdraw to evaporation during cooling. Annual U.S. data center water consumption could double or quadruple by 2028 to roughly 150-280 billion liters per year compared with 2023 levels. An assessment of 9,055 facilities worldwide indicates that by the 2050s, nearly 45% may face high exposure to water stress.

States are responding with legislation. California, Iowa, and Michigan are considering bills requiring operators to submit regular reports on water use, while South Carolina and Kansas are exploring mandates for closed-loop cooling systems that recirculate water rather than consuming it.

Figure 3: Global data center electricity consumption in terawatt-hours, with IEA Base Case projections through 2030. Consumption is growing at approximately 15% per year, more than four times faster than total electricity consumption growth from all other sectors. The 2030 figure of 945 TWh would represent roughly 3% of global electricity use. Sources: International Energy Agency (IEA), “Energy and AI” (2025); IEA, “Electricity 2025.”

Regional Analysis

North America

Roughly 50% of global data center capacity is located in the Americas, which and are growing at a 17% CAGR, faster than any other major region. The United States represents roughly 90% of the Americas’ capacity. At the end of H2 2025, capacity under construction in the Americas reached 25.3 GW, with pre-commitment rates approaching 89%.

Northern Virginia remains the world’s single largest data center market, accounting for 6.3 GW of capacity under construction, about 25% of the Americas pipeline. However, supply constraints are forcing geographic diversification. Dallas-Fort Worth is emerging as the second-largest market, with recent completions and ongoing projects expected to double the market’s size by year-end 2026. Charlotte-Raleigh, Austin, and San Antonio are attracting development with cheaper power and faster delivery timelines.

U.S. primary market construction capacity fell to 5,994.4 MW at year-end 2025, down from a record 6,350 MW at the end of 2024. Wood Mackenzie attributes the decline to permitting, zoning, and power procurement delays rather than any reduction in demand.

Europe

EMEA plans to add 13 GW of new data center supply, growing at a 10% CAGR. Inventory across the four largest European data center markets – London, Frankfurt, Paris, and Amsterdam (the “FLAP” markets) – increased by 7.2% over the past year. Frankfurt and Paris led with annual inventory growth of 13.7% and 11.2%, respectively.

European data center development faces a distinctive set of constraints. The Netherlands imposed a moratorium on new data center construction in the Amsterdam region in 2022, though it has since been partially lifted. Ireland has introduced planning conditions requiring on-site power generation for new data centers. These regulatory actions reflect growing public concern about the impact of data centers on housing availability, electricity prices, and sustainability targets.

Sovereign data requirements under the EU’s General Data Protection Regulation (GDPR) and the proposed European Data Act continue to drive demand for local capacity. A third of the European PPA market is now dedicated to data center operators, with the largest 2025 deals signed in Italy, Finland, and Spain.

Asia-Pacific

Asia-Pacific data center capacity is projected to grow from 32 GW to 57 GW by 2030, a 12% CAGR. In the second half of 2025, the regional development pipeline expanded by 2,751 MW to reach 19,371 MW total, comprising 3,677 MW under construction and 15,694 MW in various planning stages. Operational capacity stood at 13,763 MW.

Southeast Asia has become the fastest-growing subregion. The Southeast Asia data center construction market was valued at $5.42 billion in 2024 and is expected to reach $11.80 billion by 2030 at a 13.82% CAGR. Malaysia’s Johor Bahru corridor is absorbing enormous hyperscale investment because it offers Singapore-adjacent connectivity at significantly lower land and power costs. Microsoft, Google, and ByteDance have all committed substantial capacity in Malaysia.

Japan and South Korea remain significant markets, driven by both domestic enterprise demand and the strategic importance of low-latency connectivity to China and Southeast Asia. India represents the fastest-growing APAC cloud market, with a 19.1% CAGR fueled by government Digital India programs and an expanding startup ecosystem.

Middle East

The Middle East is emerging as a significant data center market, with capacity projected to triple from 1 GW in 2025 to 3.3 GW over the next five years. The Middle East data center colocation market is expected to attract roughly $33.79 billion in investment from 2025 to 2030. AWS announced plans to invest more than $5.3 billion in Saudi Arabia for data center infrastructure, with completion expected by 2026. The region’s appeal centers on its proximity to both European and Asian markets, government investment incentives, and abundant solar energy resources for powering facilities.

Figure 4: Data center capacity under construction or in pipeline by region as of H2 2025. The Americas dominate with 25.3 GW, driven primarily by the United States (90% of regional capacity). Asia-Pacific’s 19.4 GW pipeline includes strong growth in Malaysia, India, and Japan. EMEA’s 13 GW includes both the established FLAP markets and expansion into the Nordics and Iberian Peninsula. Middle East and Latin America figures are estimates based on announced project pipelines. Sources: CBRE, “Global Data Center Trends 2025”; DC Byte, “2026 Data Centre Outlook.”

Technology Trends

Liquid Cooling

Rising thermal demands from AI accelerators have made liquid cooling a necessity rather than an option for new high-density deployments. NVIDIA’s current-generation Blackwell GB300 racks draw 163 kW. The next-generation Vera Rubin NVL144, expected in 2026, may require more than 300 kW per rack. At these power densities, traditional air cooling simply cannot remove heat fast enough.

The global data center liquid cooling market was estimated at $6.65 billion in 2025, expected to reach $8.17 billion in 2026, and is growing at a 20.1% CAGR through 2033. The liquid immersion cooling submarket was valued at $2.64 billion in 2025 and is projected to grow to $16.55 billion by 2034 at a 22.62% CAGR. Dell’Oro Group projects the total liquid cooling market will approach $7 billion by 2029.

Three primary liquid cooling approaches are competing for dominance. Direct-to-chip cooling, which routes liquid through cold plates attached directly to processors, is the most widely adopted today. Single-phase immersion cooling, where servers are submerged in a non-conductive fluid, is gaining traction for ultra-dense deployments. Two-phase immersion cooling, which uses a fluid that boils on contact with hot components, offers the highest cooling efficiency but remains more complex to deploy. In January 2026, Submer and Inspur deepened their collaboration to deploy two-phase immersion systems for hyperscale racks in China, targeting power densities exceeding 100 kW per rack.

AI-based thermal management algorithms are increasingly used to optimize cooling in real time, adjusting coolant flow rates and temperatures based on workload patterns. Waste heat recovery is also gaining attention: several operators in Northern Europe are piping waste heat from data centers into district heating systems, turning a cost center into a revenue stream.

Edge Data Centers

Edge computing represents a smaller but growing segment of the data center market. Precedence Research values the edge data center market at $18.32 billion in 2025 and $21.71 billion in 2026, growing at an 18.50% CAGR through 2034. Fortune Business Insights provides a lower estimate of $16.11 billion in 2025.

Edge deployments are driven by use cases requiring low latency: autonomous vehicles, industrial IoT, real-time AI inference, content delivery, and gaming. The relationship between edge and centralized data centers is complementary rather than competitive. AI model training will remain concentrated in hyperscale facilities, while inference workloads that require millisecond-level response times will increasingly migrate to the edge.

Modular and Prefabricated Construction

Construction timelines have become a bottleneck as significant as power procurement. Traditional data center construction takes 18-24 months or longer. Prefabricated modular designs – where data center components are manufactured off-site and assembled on location – can reduce deployment times to 6-12 months. Microsoft, Google, and Meta have all invested in modular construction approaches to accelerate their buildout schedules.

Challenges and Risks

Overinvestment Risk

The most consequential risk facing the data center market is that the current investment cycle overshoots demand. The hyperscaler capex commitment for 2026 alone – $602-690 billion across the top five – is predicated on the assumption that AI workloads will continue growing at rates that justify these expenditures. If enterprise AI adoption proves slower than projected, if AI model efficiency improvements reduce compute requirements, or if the economic returns from AI applications disappoint, the industry could face a period of overcapacity.

Financial structure amplifies this risk. With hyperscalers projected to spend 90% of their operating cash flow on capex in 2026 and borrowing expected to top $400 billion, there is limited flexibility to scale back. Physical infrastructure cannot be paused mid-construction without significant writedowns.

Power Availability

16 GW of new data center capacity is scheduled to come online globally in 2026 across approximately 140 projects. Sightline Climate estimates that 30-50% of that pipeline will not materialize on schedule, primarily due to power constraints. The gap between announced capacity and delivered capacity has been widening, creating a risk that supply shortfalls persist for years.

Transformer lead times of 128-144 weeks mean that power equipment ordered today will not arrive until 2028-2029. New natural gas generation takes 3-4 years to permit and build. SMR nuclear technology is promising but unproven at commercial scale, with the earliest deployments targeted for the late 2020s or early 2030s. The mismatch between the pace of AI demand growth and the pace of energy infrastructure deployment is the single largest constraint on the industry’s trajectory.

Regulatory and Community Opposition

Data centers face growing opposition in communities where they are concentrated. In Northern Virginia, which hosts the world’s largest cluster of data centers, local governments have tightened zoning requirements and imposed noise ordinances. The Netherlands imposed construction moratoriums. Ireland requires on-site power generation. Multiple U.S. states are introducing water usage reporting requirements.

The environmental footprint of data centers is increasingly subject to public scrutiny. AI-related carbon emissions could reach 32.6-79.7 million tons of CO2 in 2025 alone. Data center water consumption in the United States could quadruple by 2028. As these figures become more widely understood, regulatory constraints are likely to tighten further, particularly in water-stressed regions and markets where data centers compete with residential users for electricity.

Supply Chain Constraints

Beyond transformers, the data center supply chain faces shortages in multiple categories. High-voltage switchgear, backup generators, fiber optic cable, and specialized construction labor are all in tight supply. GPU availability remains constrained, with NVIDIA’s Blackwell architecture sold out through mid-2026 with a backlog of 3.6 million units. Cooling equipment lead times have extended as manufacturers struggle to keep pace with the shift from air to liquid cooling.

Geopolitical Risk

Data sovereignty requirements, export controls on AI chips, and geopolitical tensions create uncertainty about where data centers can be built and what hardware they can contain. U.S. export controls on advanced AI chips to China have reshaped the competitive dynamics in Asia. The EU’s data governance framework increasingly favors in-region processing. These regulations fragment the global market and increase costs for operators that serve customers in multiple jurisdictions.

Figure 5: 2026 capital expenditure guidance by company. Amazon leads at $200 billion, followed by Alphabet ($175-185 billion), Meta ($115-135 billion), and Microsoft ($120 billion+). Oracle targets $50 billion. Meta’s midpoint figure represents a capex intensity of 55-67% of projected revenue. Sources: Amazon Q4 2025 earnings; Alphabet Q4 2025 earnings; Meta Q4 2025 earnings; Microsoft Q2 FY2026 earnings; Oracle earnings guidance.

Future Outlook

The Supply-Demand Mismatch Will Persist Through 2028

16 GW of new data center capacity is scheduled for 2026, but only about 5 GW is currently under construction, and analysts at Sightline Climate estimate that 30-50% of the total pipeline will slip. At the same time, hyperscalers are committing to spend $600-690 billion in 2026 alone, with AI infrastructure accounting for three-quarters of that total. The gap between committed spending and deliverable capacity suggests that the power and construction bottleneck will not ease before 2028 at the earliest. Colocation vacancy rates will likely remain below 2% in Tier 1 markets through that period, and pricing above $150/kW will become the new baseline.

Nuclear Will Become a Meaningful But Not Dominant Power Source

The 10+ GW of nuclear commitments made in 2025 represent the most significant shift in data center energy procurement in years. However, nuclear’s actual contribution will take time to materialize. The Three Mile Island restart is targeted for 2027, Kairos Power’s SMR fleet for 2030+, and Amazon’s Energy Northwest SMRs for the early 2030s. Deloitte’s projection that nuclear could meet 10% of data center electricity demand by 2035 appears reasonable. Natural gas and renewables with storage will remain the dominant energy sources for data center growth over the next five years.

The Colocation Market Will Consolidate

The gap between the largest colocation operators and smaller regional players is widening. Equinix, Digital Realty, Iron Mountain, and American Tower collectively generate more than $32 billion in annual revenue and have access to capital markets that smaller operators cannot match. The capital requirements for AI-ready facilities – high power density, liquid cooling, low-latency interconnection – raise the bar for new entrants. Expect continued M&A activity, with larger operators acquiring regional players for their land banks, utility interconnections, and permitting approvals.

AI Efficiency Gains Are the Wildcard

The strongest counterargument to the current investment thesis is that AI models are becoming more efficient. Techniques like quantization, distillation, and mixture-of-experts architectures can reduce the compute required for a given inference task by 2-10x. If these efficiency gains outpace demand growth – if, for example, a 2027 model delivers GPT-4-level performance at one-tenth the compute cost – then the economics of the current buildout could shift dramatically. The industry’s response has been that efficiency gains will be consumed by increased usage: every reduction in cost per inference opens new use cases and expands the addressable market. Whether that assumption holds will determine whether 2026’s investment looks prescient or excessive.

Geographic Diversification Will Accelerate

Power constraints in established markets are pushing development to new regions. In the United States, Dallas-Fort Worth, Charlotte-Raleigh, Austin, and San Antonio are growing rapidly. Internationally, Malaysia’s Johor Bahru corridor, the Middle East (Saudi Arabia and the UAE), India, and the Nordics are attracting disproportionate investment. This geographic diversification carries its own risks – less mature power grids, untested regulatory environments, thinner talent pools – but the mathematics of supply constraint leave operators little choice.

The Industry’s Carbon and Water Footprint Will Face Greater Scrutiny

The 2024 figure of 415 TWh of global data center electricity consumption will roughly double by 2030. Water consumption in the United States could quadruple by 2028. These numbers are becoming part of the public discourse, and legislators are responding. The data center industry’s sustainability narrative – anchored in renewable energy PPAs and carbon offset purchases – will face pressure to include absolute reduction targets, Scope 3 emissions accounting, and water usage caps. Operators that fail to address these concerns proactively will face permitting delays, higher regulatory costs, and community opposition that slows their buildout.