AI data center surge pushes utilities toward new gas as Big Tech touts climate progress

Big Tech’s rapid buildout of AI data centers is beginning to reset U.S. power planning. Utilities in key cloud regions are proposing new gas capacity to meet hyperscale load, a shift that sits uneasily with corporate climate pledges and hourly clean energy targets.

The documents do not name customers. The load characteristics point to large cloud campuses operated by Microsoft, Amazon and Google, which report record capital expenditures and expanding AI infrastructure in recent securities filings.

What the filings show

Dominion Energy’s 2023 Integrated Resource Plan filed at the Virginia State Corporation Commission projects sustained load growth in Northern Virginia tied to data centers. The filing describes a step change in demand and evaluates multiple portfolios with additional gas-fired capacity to maintain reliability in the near term.

PJM Interconnection’s 2023 Load Forecast Supplement also cites data center growth in the Dominion zone as a primary driver of upward revisions. Georgia Power’s 2024 integrated resource plan update submitted to the Georgia Public Service Commission attributes accelerating load to large data centers and requests authority for additional firm capacity, including new natural gas resources and purchased power.

These utility assessments mirror risks the platforms disclose to investors. Microsoft’s 2024 Form 10 K warns that growth in its cloud business depends on adequate supplies of electricity and water for data centers and that constraints could affect Azure capacity. Alphabet’s 2024 Form 10 K notes rising electricity use and variability in grid carbon intensity as the company deploys more compute for AI. Amazon’s 2024 Form 10 K discusses exposure to energy price and availability for AWS operations as a material factor.

Published research has begun to quantify the resource intensity. A 2023 peer reviewed study led by Shaolei Ren at the University of California Riverside estimated that training a large language model on the scale of GPT 3 could consume hundreds of thousands of liters of water, with significant ongoing water and power needs for inference.

The stakes for ratepayers and emissions

If commissions approve new gas units and firm contracts, costs flow into customer rates for years. Dominion and Georgia Power both argue that traditional capacity is needed to firm variable renewables and to meet clustered, around the clock loads from data centers. Environmental groups counter in docketed comments that utilities are leaning too quickly on combustion resources and underinvesting in transmission, storage and demand response.

The emissions ledger is complex. Amazon, Microsoft and Google continue to announce new wind, solar and geothermal contracts and report annual matching of consumption with renewable purchases at a corporate level. Utility plans underscore a gap between annual certificates and real time supply, which can leave grids serving cloud clusters reliant on thermal generation during scarcity.

What happens next

Regulatory timelines are active. The Virginia commission is reviewing Dominion’s resource plan through ongoing proceedings that set the company’s planning direction and cost recovery. Georgia regulators have held hearings on the 2024 plan update and will rule on requested capacity additions and associated costs.

At the federal level, new U.S. power plant rules finalized in 2024 set carbon requirements for certain new gas units based on operating profiles, which could affect utility technology choices tied to data center loads. The next set of utility filings will show whether hyperscale customers are willing to sign firmer long term contracts for clean firm power, or whether ratepayers will continue to underwrite gas additions that make the AI boom possible.