On a napkin
The short version
Data center electricity demand — driven by cloud computing growth and accelerated by AI training and inference — is the dominant new load story in US electricity markets. After two decades of flat or declining US electricity demand, data centers (combined with electrification of transportation and buildings) are driving the first sustained load growth period since the late 1990s. The effects: capacity prices at record highs, transmission planning struggling to keep pace, hyperscaler procurement strategies pulling unconventional resources online, and a regulatory scramble across every major ISO to figure out who pays for what.
PJM capacity prices tell the story. The 2024/25 BRA cleared at $28.92/MW-day. The 2025/26 BRA cleared at $269.92 — a nearly 10× increase. The 2026/27 and 2027/28 BRAs both cleared at the FERC-approved price cap ($329.17 and $333.44 respectively). The 2027/28 auction was the first in PJM history where the entire RTO fell short of the reliability requirement. PJM's forecast peak load is growing by over 5,400 MW per delivery year — most of it data centers.
The scale of the demand growth
US data centers consumed approximately 4.4% of total US electricity in 2023, projected to grow to 6-12% by 2028 depending on AI adoption pace. The magnitude of individual data center loads has shifted dramatically: traditional cloud data centers drew 20-50 MW; hyperscale AI training campuses now draw 100-500 MW continuously. Reports of single proposed sites reaching 1,000-2,000 MW (1-2 GW) are increasingly common — each equivalent to multiple large nuclear units.
Major data center hubs include Northern Virginia's Loudoun County (the world's largest, over 4 GW operational), Phoenix-Mesa AZ, Dallas-Fort Worth TX, Columbus OH, Atlanta GA, Chicago IL, Quincy WA, and Reno NV. Newer development is shifting toward areas with available power capacity — the Texas Panhandle, the Mississippi Delta, Wyoming, and other regions with available transmission capacity and faster permitting. ERCOT large load interconnection requests quadrupled in 2025, with utilities and ERCOT itself proposing new approval frameworks to manage demand requests well in excess of available supply.
How capacity prices reflect the shock
PJM's capacity market is the most affected. The 2025/26 BRA cleared at $269.92/MW-day RTO-wide ($466.35 in BGE, $444.26 in Dominion) — a 10× increase driven primarily by data center load growth combined with generator retirements and a methodological change to marginal ELCC accreditation. The 2026/27 BRA cleared at $329.17/MW-day at the FERC-approved price cap, with all four major LDAs constrained. The 2027/28 BRA cleared at $333.44 (also capped) and was the first auction where the entire RTO fell short of its reliability requirement.
For commercial customers in PJM, this translates directly to higher capacity costs. A 10 MW industrial facility's annual capacity cost rose from roughly $105,000 at the 2024/25 price to over $1.2 million at the 2027/28 price — a roughly $1.1 million increase. Customers with high coincident peak (CP) tags face the largest exposures, making CP management more economically critical than ever.
The hyperscaler procurement playbook
Major data center operators have responded with aggressive procurement strategies that increasingly bypass conventional utility supply. Microsoft signed a 20-year PPA with Constellation Energy to restart Three Mile Island Unit 1 (the undamaged reactor). Amazon's AWS bought a data center campus co-located with Talen Energy's Susquehanna nuclear plant. Google, Meta, and other hyperscalers have signed massive solar, wind, and storage PPAs — collectively the largest corporate procurement portfolios ever assembled.
New approaches include nuclear co-location (data centers physically adjacent to nuclear plants, drawing dedicated output), behind-the-meter gas turbines (operators building their own generation on-site), and large-scale battery storage (providing backup and grid services). The regulatory and reliability implications are significant: co-located generation that doesn't deliver to the grid raises questions about cost allocation, stranded grid investment, and reliability contributions. FERC, PJM, and state regulators are actively developing new frameworks for how these arrangements should be treated.
What this means for everyone else
For commercial and industrial customers without unique data center load characteristics, the AI data center surge means significantly higher capacity costs, accelerated transmission rate cases, and intensifying competition for available generation capacity. PJM's market monitor and others have raised concerns about data center load growth crowding out other customers' supply or imposing reliability risks. Several regulatory reforms are in process — large load interconnection rules, demonstrations of supply availability, and special tariff structures — but the underlying tension between demand growth and supply additions is unlikely to resolve quickly. For procurement teams, this argues for locking in longer-term contracts, prioritizing CP/peak management, and pursuing on-site generation where feasible to reduce exposure to capacity price escalation.
Common questions
Related reading on The Outlet
Need help navigating this topic?
Pilot Energy’s advocacy team can help you make sense of the energy landscape and build a strategy that works for your organization.
Talk to an Advisor