AI Infrastructure

Frontier AI data centers are scaling to gigawatt size fast.

Some frontier AI facilities are projected to reach 1 GW of total power capacity within 1 to 3.6 years of starting construction, compressing infrastructure buildout into a timeline that local governments and communities may struggle to match with oversight, water planning, and power planning.

Tracked with satellite imagery, permits, company disclosures, and other public documents.

Scroll

How quickly are frontier AI data centers scaling in power?

Among recently launched frontier projects tracked by Epoch AI, the path from groundbreaking to gigawatt-scale power can be measured in years, not decades.

Operational

Under construction

Planned

Total new power capacity by year (MW)

1 GW threshold

Where local conflicts emerge.

As facilities grow larger, disputes increasingly center on resource demand, siting rules, and who gets to approve the terms.

Water demand

Evaporative cooling systems at large facilities can consume millions of gallons per day. As projects scale toward gigawatt capacity, water-use studies and disclosure requirements have become a primary flashpoint with local governments.

Electricity supply

A gigawatt-scale facility draws more continuous power than many mid-size cities. That load accelerates pressure for grid upgrades, raises questions about carbon commitments, and in some cases has prompted deals with nuclear and natural gas operators.

Zoning and land use

Industrial footprints spanning hundreds of acres frequently conflict with existing land-use plans. Developers have increasingly pursued municipal annexation into more permissive jurisdictions to sidestep county-level restrictions.

Taxation and incentives

Many jurisdictions offer substantial property-tax abatements to attract data centers. As facility size and resource demand grow, so does the public debate over whether the fiscal tradeoffs deliver meaningful community benefit.

Transparency and oversight

Facility plans often advance through non-disclosure agreements and permit exemptions, limiting public ability to assess water, grid, and environmental impacts before construction begins.

Case study

Linn County and Palo, Iowa

A local dispute shows how fast-moving infrastructure can outpace oversight frameworks when approval pathways differ across neighboring jurisdictions.

2023

Linn County adopts a stricter data center ordinance with water-use study requirements, environmental assessment thresholds, and mandatory public disclosure before approval.

2024

Google shifts its proposed project toward annexation into the city of Palo rather than proceeding under the county's tighter rules, bypassing the disclosure requirements.

Ongoing

The jurisdictional mismatch persists, illustrating how projects can exploit differing approval pathways between neighboring governments — a pattern likely to repeat as buildout accelerates.

How these centers get power

Power source matters alongside capacity scale. In Iowa, Google is linked to a long-term arrangement tied to the restart of the Duane Arnold nuclear plant. Nationally, many frontier AI data center projects are also turning to natural gas for faster deployment, raising questions about the sector's long-run carbon trajectory and grid resilience.