The $700 Billion Paradox: US Data Center Construction Drops for the First Time Since 2020

US data center capacity under construction fell 5.7% in 2025, dropping to 5.99 GW from 6.35 GW the prior year, according to CBRE Group.¹ The decline marks the first contraction in active US data center construction since 2020, even as hyperscaler capital expenditure approaches $700 billion for 2026 alone.² That disconnect between record investment and shrinking construction tells the most important story in infrastructure today: the bottleneck has shifted from capital to execution.

Demand has never been stronger. Vacancy sits at a historic 1% for the second consecutive year.³ Hyperscalers have pre-committed 92% of all capacity currently under construction.⁴ Microsoft disclosed an $80 billion backlog of Azure orders it cannot fulfill due to power constraints.⁵ The money exists. The customers exist. What does not exist is the physical capacity to turn dollars into operating data centers at the pace the market demands.

TL;DR: US data center construction contracted for the first time in five years despite record-breaking demand and nearly $700 billion in hyperscaler capex. The constraint has migrated from "can we fund it?" to "can we build it?" Power transformer lead times stretch to 128-144 weeks. Switchgear backlogs run 45-80 weeks. A wave of 54 local moratoriums has blocked or delayed $64 billion in projects. The construction industry faces a 439,000-worker shortage. Sightline Climate projects that 30-50% of the 2026 pipeline will not materialize on schedule. Meanwhile, Texas prepares to overtake Virginia as the world's largest data center market by 2030, as established hubs hit hard physical ceilings.

The numbers: what dropped and where

CBRE's year-end 2025 data reveals a market colliding with physical constraints at every level. The national decline masks sharper regional contractions in the markets that historically drove US data center growth.

Northern Virginia, the world's largest data center market, saw projects under construction fall 29%.¹⁶ Hillsboro, Oregon dropped 15%.⁶ Silicon Valley contracted 14%.⁶ These three markets represent the traditional backbone of US data center infrastructure, and all three hit walls simultaneously.

The pipeline tells an equally sobering story. Sightline Climate tracked approximately 140 projects slated for completion in 2026, representing roughly 16 GW of capacity.⁷ Only about 5 GW has active construction underway.⁷ The remaining 11 GW sits in "announced" status with no visible construction progress, despite typical build timelines of 12-18 months.⁷ Sightline projects that 30-50% of the 2026 pipeline will not come online before year-end, pushing delivery into 2027 or later.⁷

The pattern repeated from 2025, when 26% of expected capacity slipped its target date, and an additional 10% of projects quietly pushed back commercial operation dates without formal announcements.⁷ Project delays have become the default, not the exception.

The supply-chain siege: power equipment in crisis

The single largest constraint on data center construction sits in the electrical supply chain. Power transformers, switchgear, and generation equipment now dictate project timelines more than design, financing, or even site selection.

Transformers: 128-144 weeks and counting

Large power transformers average 128-week lead times, according to Wood Mackenzie's Q2 2025 survey.⁸ Generator step-up units (GSUs) stretch even further, averaging 144 weeks.⁸ Some shipments take longer still. Projects that failed to place transformer orders 18-24 months before groundbreaking now face multi-year waits.⁹

The pricing trajectory compounds the timeline problem. Since 2019, power transformer unit prices have increased 77%.¹⁰ GSU prices have climbed 45% over the same period.¹⁰ Distribution transformers in some classes have surged as much as 95%.¹⁰ A single hyperscale campus can require dozens of large transformers, each costing up to $250,000.¹⁰

Manufacturers have responded, committing nearly $2 billion to new or expanded transformer production capacity since 2023.¹⁰ Hitachi Energy alone announced a $1 billion investment in US grid infrastructure manufacturing to address data center power demand.¹¹ But capacity expansion carries its own lead time. New transformer manufacturing lines take 2-3 years to bring online, and the demand surge shows no signs of slowing. Wood Mackenzie projects that most transformer specifications will remain in marginal shortage through 2030.⁸

Switchgear: the second bottleneck

Switchgear lead times run 45-80 weeks, creating a parallel constraint that compounds transformer delays.¹² Medium-voltage switchgear, essential for data center power distribution, has edged down to an average of 44 weeks in some categories but remains above 60 weeks for custom configurations.¹²¹³ Some assessments place high-specification switchgear at nearly three years.¹²

The convergence of transformer and switchgear delays forces project managers to lock in electrical equipment orders before finalizing architectural designs. Utilities and EPC firms report redesigning project schedules around equipment availability rather than construction sequencing.¹³

Fiber: a quieter crisis

Optical fiber prices surged over 70% from December 2025 to January 2026 alone, driven by AI-optimized hyperscale facilities that consume 5-10x more fiber infrastructure than conventional cloud data centers.¹⁴ At least one major fiber manufacturer has sold all inventory through the end of 2026.¹⁴ The quotation validity period for optical fiber has collapsed to just seven days, reflecting continuous price fluctuations.¹⁴

Ribbon fiber, the high-density cabling preferred for backbone connections in hyperscale facilities, now accounts for over 30% of major telecom operators' procurement and reaches 80% penetration in new AI computing centers.¹⁵ Fiber preform manufacturing, the upstream bottleneck, requires 18-24 months for capacity expansion.¹⁴ Global data center fiber demand surged 75.9% year-over-year in 2025.¹⁴ Vendors like Corning and CommScope have invested in expanding production capacity, but relief remains 12-18 months away at best.¹⁴

The workforce wall: 439,000 workers short

The US construction industry faces a shortage of approximately 439,000 workers, with data center projects absorbing a disproportionate share of available skilled labor.¹⁶¹⁷ The Associated Builders and Contractors (ABC) identified the gap concentrated in electrical work, plumbing, and welding, the exact trades data centers demand most.¹⁷

An Uptime Institute survey found that 52% of construction firms reported staffing shortages causing project disruptions in 2025, up from 43% the prior year.¹⁸ The trajectory points in only one direction. Amazon, Microsoft, and Meta plan 127 new hyperscale facilities globally in 2026, with each project employing approximately 850 construction workers over 18 months.¹⁸

Scale amplifies the problem. Peak crew sizes that once reached 750 workers at major data center campuses will hit 4,000-5,000 workers by early 2026 at sites like DataBank's Red Oak campus.¹⁹ Finding 5,000 skilled electricians and mechanical technicians in a single labor market, when the entire national industry already runs 439,000 workers short, creates mathematical impossibility.

Wages rise, but workers stay scarce

Data center construction workers earn an average of $81,800 per year, or roughly $39.33 per hour, representing a 32% premium over non-data-center construction roles.²⁰ Workers switching into data center projects typically see 25-30% pay increases.²⁰ Despite wage inflation, 40% of construction firms increased base pay in 2025 by more than they did in 2024, signaling that money alone cannot close the gap.²⁰

The structural issue runs deeper than compensation. The construction workforce is aging. Apprenticeship programs take 3-5 years to produce journey-level electricians. Immigration policy limits the pipeline of skilled tradespeople. And data center construction competes for the same workers building semiconductor fabs, battery plants, and grid infrastructure under the CHIPS Act and Inflation Reduction Act.

The moratorium wave: $64 billion blocked

Community opposition has matured from scattered complaints into a coordinated national movement. According to Data Center Watch, $64 billion in US data center projects have been blocked or delayed by local opposition, with 142 activist groups coordinating efforts across 24 states.²¹

Local moratoriums have proliferated at a pace the industry failed to anticipate. New Orleans passed a moratorium in late 2025.²² Madison, Wisconsin followed in early January 2026.²² Denver moved to seek a moratorium in February 2026.²³ Communities throughout Georgia and Michigan have enacted similar measures, with at least 19 Michigan towns pausing data center development.²⁴

At the state level, eleven in-session states filed data center moratorium bills during the 2026 legislative cycle.²⁵ Georgia HB 1012 would bar counties and cities from issuing any data center construction permit until March 2027.²⁵ Vermont S.205 proposes a moratorium on "AI data centers" through July 2030.²⁵ Oklahoma SB 1488 would halt development until November 2029.²⁵

The opposition draws from both sides of the political spectrum. Water consumption, noise pollution, property tax abatements, and grid strain have united constituencies that rarely agree on anything else.²⁶ TechCrunch reported that public opposition to AI infrastructure "heated up" measurably in February 2026, driven by tangible local impacts rather than abstract concerns.²⁶

The financial impact compounds over time. A 12-month moratorium delays not just construction but cascading procurement timelines for equipment with multi-year lead times. A project paused for a year of zoning review may discover its transformer order has expired, adding another 128 weeks to the restart timeline.

The geographic shift: Texas rises, Virginia stumbles

JLL's North America Data Center Report declared that Texas will likely surpass Northern Virginia as the world's largest data center market by 2030.²⁷ The numbers support the projection: Texas has 6.5 GW currently under construction, compared to Virginia's contracting pipeline.²⁷

Three megaprojects anchor Texas's ascent. Oracle and OpenAI broke ground on the flagship Stargate data center in Abilene.²⁸ Google outlined plans for a $40 billion expansion in West Texas.²⁸ Meta began constructing a major campus in El Paso.²⁸ Together, Amazon, Microsoft, Google, and Meta committed over $600 billion in AI infrastructure spending for 2026, with Texas capturing an outsized share.²⁸

Virginia's challenges run deeper than any single constraint. Grid capacity in "Data Center Alley" faces saturation. Land scarcity in Loudoun County has driven site costs to record levels. The 29% construction decline reflects a market hitting physical ceilings that investment cannot easily overcome.¹⁶

The shift extends beyond a two-market comparison. JLL found that 64% of the 35 GW global construction pipeline now extends beyond traditional mature markets.²⁷ Secondary and emerging markets in the Midwest, Southeast, and Mountain West attract projects priced out of or blocked in established hubs. The geographic diversification reduces concentration risk but introduces new execution challenges in markets with less data center construction experience.

Demand vs. delivery: the widening gap

The paradox at the center of the construction decline becomes clearest when measured against demand. Vacancy holds at 1% for the second consecutive year.³ CBRE data shows that 92% of capacity currently under construction has already secured tenants through pre-commitment agreements.⁴ Operators deliver buildings into a market where every megawatt has a buyer waiting.

Hyperscaler capital expenditure underscores the demand signal. CNBC reported in February 2026 that Google, Microsoft, Meta, and Amazon collectively approach $700 billion in AI-related spending for the year.² CreditSights raised 2026 hyperscaler capex estimates to approximately $750 billion for the top five companies, with Amazon alone projected at $200 billion, Alphabet at $175-185 billion, Meta at $115-135 billion, and Microsoft tracking toward $120 billion or more.²⁹

Approximately 75% of aggregate hyperscaler capex funds AI-related infrastructure, representing roughly $450 billion in AI-specific capital deployment.²⁹ The spending reflects not aspiration but committed orders, signed leases, and equipment procurement already in motion.

The mismatch between available capital and deliverable capacity creates several downstream effects. Lease rates climb as tenants compete for scarce inventory. Developers pay premium prices for pre-ordered equipment. Construction wages inflate as projects bid against each other for the same skilled workers. The entire cost structure of data center deployment ratchets upward, not because the projects lack funding, but because every project competes for the same constrained pool of labor, equipment, and permitted sites.

For organizations deploying and maintaining critical infrastructure at scale, this execution environment demands partners with deep field capabilities and geographic reach. Introl operates 550 HPC-specialized field engineers across 257 locations, with demonstrated capacity to deploy 100,000 GPUs. When every week of delay compounds downstream, having embedded technical workforce at scale becomes the differentiator between projects that deliver and projects that slip.

The constraint has a name: execution capacity

The data tells a unified story. Capital no longer limits data center growth. The constraint has shifted to execution capacity, a compound bottleneck built from power equipment lead times, workforce shortages, regulatory resistance, and supply chain fragility.

Consider the timeline math for a hypothetical 100 MW data center starting from scratch today. Transformer procurement: 128 weeks. Switchgear: 45-80 weeks. Permitting in an increasingly hostile regulatory environment: 6-18 months, assuming no moratorium. Construction workforce recruitment in a market short 439,000 workers: uncertain. Fiber procurement with 70%+ price surges and sold-out inventory: add months of negotiation.

Even with unlimited capital, the fastest realistic path from decision to energization now stretches well beyond three years for a greenfield project in most US markets. Five years ago, that timeline ran 18-24 months.

The industry has started adapting. Modular construction techniques reduce on-site labor requirements.³⁰ Pre-fabricated power systems compress equipment installation timelines.³⁰ Some operators bypass grid constraints entirely through on-site generation using fuel cells or natural gas turbines. Others secure equipment years before breaking ground, warehousing transformers and switchgear against future demand.

But adaptation at the margins cannot overcome structural constraints at the center. The US needs more electricians, more transformer manufacturing capacity, more permitted sites, and more grid interconnection bandwidth. Each of those inputs operates on its own multi-year timeline for expansion. Until supply catches demand across all four dimensions simultaneously, the gap between capital committed and capacity delivered will persist.

Key takeaways

For infrastructure planners

• Lock in power equipment procurement 24-36 months before target energization dates. Transformer and switchgear lead times will not shorten meaningfully before 2028.⁸
• Evaluate secondary and emerging markets where permitting timelines remain shorter and community opposition has not yet organized. The 64% of pipeline activity outside mature markets reflects rational site selection.²⁷
• Build regulatory engagement into project timelines from day one. The moratorium wave targets projects that appear without prior community consultation.²¹

For operations teams

• Plan for workforce constraints at every phase of construction and commissioning. Peak crew sizes approaching 5,000 workers require labor market analysis as rigorous as power availability studies.¹⁹
• Develop relationships with multiple equipment suppliers and explore refurbished or reconditioned transformer markets to reduce single-source dependency.¹⁰
• Invest in modular and pre-fabricated construction approaches that reduce on-site labor intensity and compress delivery schedules.³⁰

For strategic planning

• Model 30-50% slippage into 2026 capacity projections. Sightline Climate's data suggests planned timelines systematically overstate delivery dates.⁷
• Monitor the Texas-Virginia market shift for portfolio diversification opportunities. Texas's 6.5 GW construction pipeline represents the largest concentration of new capacity in the US.²⁷
• Factor rising construction costs ($11.3M/MW average, +6% YoY) into project economics, with AI-ready facilities running $20M/MW or higher.³¹

References