The United States runs more data centers than any other country, at about 5,426 as of March 2025. This number is growing exponentially, and bringing immense power demands along with it.
The U.S. Department of Energy says data-center electricity use has tripled since 2014 and is “expected to double or triple by 2028,” with load rising from 4.4% of U.S. electricity in 2023 to 6.7-12% by 2028.
Deloitte projects AI-oriented facilities alone could reach 123 GW by 2035 (30× 2024 levels) and most operators expect AI to push demand higher through 2035. As Microsoft’s Brad Smith told the Senate, “We need half a million electricians that we do not have now,” pointing to a human-capital bottleneck as real as any transformer.
Local economic results are uneven. A proposed campus in Jasper Parish, Louisiana touts $7.5B in investment and ~280permanent jobs; Bethlehem, Pennsylvania weighed $1.39B for 48 jobs. Meanwhile, Northern Virginia’s world-scale cluster faces mounting water and grid strain. Virginia alone is estimated at ~1.8B gallons/year of data-center water use, forcing tougher trade-offs on siting and mitigation.
The construction and operation of data centres from OpenAI, Meta, xAI, among others, sits inside that reality: accelerating AI loads, tight interconnection queues, and communities negotiating jobs and tax base against land, power, and water.
1. OpenAI – Stargate (Abilene, Texas)
Location: Abilene, Texas
Status: Operational + expansion
Stakeholders: OpenAI, SoftBank, Oracle, MGX, Crusoe Energy Systems, Blue Owl Capital, JPMorgan (debt financing)
Cost to build: Approx. $15 billion for Abilene site; Oracle chip purchase $40 billion; overall Stargate investment could reach $100–500 billion
“Each building is a half a million square feet. There are 10 buildings currently being built, but that will expand to 20…” – Larry Ellison (Oracle)
Stargate is a massive AI infrastructure venture. The Abilene campus, spanning over 900 acres, is backed by OpenAI (operational lead), SoftBank (financial lead), Oracle, and MGX as equity stakeholders. Crusoe and Blue Owl own and develop the site, with JPMorgan providing ~$9.6 billion in loans. Initial build (~8 buildings, ~200 MW, ~980,000 sq ft) costs about $15 billion, with Oracle purchasing $40 billion in Nvidia GB200 chips. The project will rise to ~1.2 GW by mid-2026. Depending on funding, Stargate’s full U.S. buildout could cost between $100–500 billion.
2. xAI – Colossus (Memphis, Tennessee)
Location: Memphis, Tennessee
Status: Operational + forthcoming expansion
Stakeholders: xAI (Elon Musk), Dell Technologies, Supermicro, MLGW, TVA, City of Memphis
Cost to build: Approx. $7 billion in hardware; $80 million graywater facility
Colossus, xAI’s Memphis supercomputer, was built in record speed: 122 days, reusing a 785,000 sq ft former Electrolux plant. Key stakeholders include xAI, Dell, Supermicro, plus local utility partners MLGW and TVA. Hardware costs alone are estimated at $7 billion for ~200,000 GPUs; additional infrastructure includes an $80 million graywater recycling facility for cooling. Colossus draws ~150 MW, supplemented by portable gas generators, raising environmental concerns.
3. Meta – Prometheus (New Albany, Ohio)
Location: New Albany, Ohio
Status: Under construction
Stakeholders: Meta
Cost to build: Multi-billion dollar investment (specific figure undisclosed)
“We’re calling the first one Prometheus and it’s coming online in ’26.” – Mark Zuckerberg
Prometheus, Meta’s first titan-scale AI data center, is under construction and targeted for launch in 2026. It uses rapid-deployment, tent-style modular structures to accelerate build speed. While Meta hasn’t disclosed precise costs, this multi-gigawatt campus is part of a broader, multi-billion-dollar AI infrastructure strategy across its Prometheus and Hyperion projects.
4. Meta – Hyperion (Richland Parish, Louisiana)
Location: Richland Parish, Louisiana
Status: Under construction
Stakeholders: Meta
Cost to build: Estimated up to $50 billion
Hyperion is “almost the size of Manhattan.” – Mark Zuckerberg
Meta’s Hyperion AI campus is under construction, expected to draw up to 5 GW. “almost the size of Manhattan.” Estimated build cost may reach $50 billion, exceeding initial projections. The project includes new gas generation and renewable energy infrastructure approved by Louisiana regulators for power sustainability.
5. AWS – Project Rainier (St. Joseph County, Indiana & others)
Location: St. Joseph County, Indiana (multi-site)
Status: Under construction
Stakeholders: Amazon Web Services, Anthropic
Cost to build: $11 billion initial investment announced (Indiana site)
Project Rainier is AWS’s few-gigawatt AI cluster for Anthropic’s Claude, spanning multiple U.S. data centers including one in Indiana. The $11 billion initial investment covers dozens of building pads and infrastructure. AWS claims Rainier will deliver 5× the compute of Anthropic’s previous biggest cluster, possibly totaling up to 2.2 GW across modules.
6. CoreWeave – Lancaster AI Campus (Lancaster, Pennsylvania)
Location: Lancaster, Pennsylvania
Status: Under construction
Stakeholders: CoreWeave, Blue Owl , Turner/Wohlsen Construction, Pennsylvania utilities
Cost to build: $6 billion committed
Governor Shapiro noted Lancaster’s power abundance is “critical to the region’s energy stability amid AI demand.”
CoreWeave is investing $6 billion into its Lancaster AI campus, starting at 100 MW IT load and scalable to 300 MW. Turner/Wohlsen Construction is the builder. The project includes a $200 million power upgrade and is expected to create ~600 construction and 175 operational jobs. Pennsylvania officials consider the site vital to managing AI-related grid load.
Quote: Governor Shapiro noted Lancaster’s power abundance is “critical to the region’s energy stability amid AI demand.”
7. Google – Virginia AI Campus (Chesterfield County, Virginia)
Location: Chesterfield County, Virginia
Status: Under construction
Stakeholders: Google, State of Virginia
Cost to build: ~$9 billion regional investment (not all for this campus):
Governor Youngkin: This project will bolster “long-term infrastructure and economic development.”
Google’s AI expansion in Virginia includes a new Chesterfield County campus, part of a $9 billion investment across the region by 2026. This site is tailored for Tensor/TPU-powered AI services, with emphasis on sustainable energy partnerships. Governor Youngkin emphasized its significance for Virginia’s economic and infrastructure development.
8. Microsoft – San Antonio AI Campus (Castroville, Texas)
Location: Castroville / Medina County (near San Antonio), Texas
Status: Under construction
Stakeholders: Microsoft, local governments, utility providers
Cost to build: $1.5 billion expansion
Local leaders suggest Microsoft’s expansion is reviving “Silicon Valley of Texas” discussions.
Microsoft is spending $1.5 billion to construct ten AI-enabled data centers in and around Bexar and Medina counties. Examples include SAT 93 and 94 (each ~250,000 sq ft, ~$350 million) and SAT 89/90 (~489,400 sq ft, $765 million). Officials note the project’s scale may redefine the region as a tech hub, but also raises infrastructure concerns.
9. QTS – Atlanta-Metro Data Center (Atlanta, Georgia)
Location: Atlanta, Georgia
Status: Operational (expanding)
Stakeholders: QTS Realty Trust, Georgia Power
Cost to build: Not disclosed; estimated several hundred million dollars
Baxtel calls it “one of the largest data centers in the world.”
QTS’s flagship Atlanta-Metro campus totals ~970,000 sq ft on 95+ acres, with over 200 MW capacity (planned 275 MW). The facility serves hyperscale and enterprise AI clients. Engineers note its design efficiency and LEED Gold certification as key differentiators.
10. Switch – Tahoe Reno 1 (Tahoe Reno, Nevada)
Location: Tahoe Reno Industrial Center, Nevada
Status: Operational (Phase 1 open) with expansions underway
Stakeholders: Switch, NCR, AT&T, local officials
Cost to build: Phase 1 estimated ~$800 million–$1 billion (industry estimate)
Switch EVP Adam Kramer: facility “fulfills the vision… to create the largest data center ecosystem in the world.”
Tahoe Reno 1 is Switch’s first large-scale colocation data center in Nevada: 1.3 million sq ft, ~130 MW of Tier IV space. Operated by Switch, the campus hosts hyperscale clientele like NCR and AT&T. Switch touts renewable power and security-backed infrastructure.
America’s largest data centers have become tenants on the grid and in local tax bases. Echoing projections from the U.S. Department of Energy, Goldman Sachs Research projects that power demand from data centers will more than double by 2030. Interconnection requests for new campuses now dwarf historic norms; utilities report waves of “phantom” proposals that complicate planning and raise the risk of passing upgrade costs to ratepayers. Even optimists concede the pace collides with physical limits. “Wanting to grow your AI at that speed and at the same time meet your climate goals are not compatible.” says Quentin Good, policy analyst at Frontier Group to CNET.
Financial analysts too are skeptical. Praetorian Capital’s Harris Kupperman warns that new data centers could generate $40 billion in annual depreciation while pulling in only $15–20 billion in revenue, a mismatch that suggests long-term economic risks.
Companies like Microsoft, Google, and Meta are spending tens of billions quarterly to keep pace, while communities face mounting questions about land use, aquifer depletion, and grid stability. Corporate promises of “water positivity” and net-zero operations by 2030 look increasingly difficult to reconcile with reality.
