Over the past two years, much has been said about the scale of load growth associated with data centers and how electric utilities will serve it.
The mismatch at the heart of the AI data center market — digital infrastructure build times are often measured in months, while power systems take years — has worried investors and policymakers enough to call for everything from fast-tracking nuclear permitting to co-locating gas generation at data center sites built on top of the vast shale resources of the Permian Basin, Pennsylvania, or Wyoming. Utilities and the highly regulated structures they operate within seemed ill-fitted for the moment at hand.
But a clear shift is underway. Look across the third quarter’s utility earnings reports and EEI’s recent investor conference and you see the sector moving well beyond reporting simple CapEx expansions and massive data center pipelines. To their credit, utilities are responding with spending plans totaling more than $1 trillion over the next five years, developing large load tariffs that create a specific rate class for data center operators to better allocate costs and risk, and increasingly incorporating flexibility requirements in exchange for faster interconnection.
IPPs are faring similarly well. An investor note from Melius Research put it this way: “If it wasn’t clear the new golden age of power has arrived, it should be now. The consistent themes of every release and calls were 1) inbounds are at the highest levels ever, 2) there are now more growth opportunities than many have experienced in their careers, and 3) pricing is rapidly rising.”
New generation
Just in this past week we’ve seen important developments across each facet of the utility relationship with this market:
- AEP brought two gigawatts of data center load online in Q3, and reported 28 GW of large load contracts to be signed by 2030, 22 GW of which is for data centers. There are currently 190 GW worth of active projects in their interconnection queue. Their system peak is now projected to reach 65 GW by 2030, almost double their current 37 GW peak. AEP’s $72 billion five-year capital plan is up 33%, from $54 billion.
- Dominion Energy, a bellwether for the data center market, has 47.1 GW of data center contracted capacity, with 9.8 GW of that already in take-or-pay electric service agreements, and another 9 GW in construction letters of authorization, which require customers to reimburse Dominion Energy for all spent costs should they walk away.
- Duke Energy announced it plans to add 13 GW in new generation to its system over five years as part of a roughly $100 billion capital plan, up from its original CapEx guidance of $87 billion.
- Entergy, home to major Google and Meta planned data center builds, described its $41 billion capital plan through 2029 and said it secured an additional 4.5 GW of new generation to serve primarily its data center and industrial customers.
- NextEra Energy, via a 25-year PPA with Google, announced it would be restarting the 615 MW Duane Arnold nuclear plant in Iowa by early 2029. The agreement with Google also includes exploring the development of advanced nuclear generation across the U.S. NextEra added 3 GW of new renewables and storage to its backlog of generation, which stands at 29.6 GW today.
- DTE Energy is on a roll, having signed its first data center contract for 1.4 GW of capacity for a project by OpenAI, Oracle, and Related Digital, ata site colloquially referred to as “the Barn.” DTE Energy will be supplying all the power to the project from its existing resources, including energy storage systems. DTE currently has 3 GW of large loads in late-stage negotiations, with a plan to finalize in 2026 and commence construction in 2027.
- Southern Company is benefitting from both data centers and manufacturing, reporting 8 GW of large load contracts and a pipeline of over 50 GW, supporting a sales forecast of 8% annual growth through 2029. Seven GW, representing 23 projects, have already broken ground across their territories. Georgia Power recently filed an update to its load forecast, requesting “10 GW of capacity resources, which includes five natural gas combined cycle units and 11 battery energy storage facilities,” according to Southern Company CEO Christopher Womack.
- Xcel Energy detailed its $60 billion capital plan, which supports at least 3 GW of new data center capacity and includes 7.5 GW of renewable energy, 1.9 GW of energy storage, and 3 GW of natural gas generation. Xcel’s data center pipeline includes 1 GW contracted or under construction, an additional 2 GW of high-probability opportunities, and roughly 20 GW of pipeline in discussion.
Large load tariffs
The worry that data centers will add fuel to rate increases across the country has led to active proceedings in nearly every state. In recent research published by RMI that leveraged Halcyon’s Large Load Tariff Tracker, analysts found that when looking across 65 different state-level tariffs, the five ratepayer protections most commonly cited were: minimum contract term lengths (many over 15 years), minimum monthly billing demand (take-or-pay arrangements, of ten 75 to 90% of contract capacity), collateral requirements, exit fees, and capacity reassignment.
As recently as last week, regulators in Kansas and Michigan approved new large load tariffs for Evergy and Consumers Energy, respectively, which incorporate many of these protections, including minimum contracting terms, and requirements that these customers pay 80% of projected demand, regardless of usage.
Flexibility
In an effort to keep utilities and IPPs from relying solely on adding new generation to serve data centers, particularly gas-fired power plants, the industry has been looking at ways to leverage flexibility to increase grid utilization and “unlock” capacity.
Flexibility is a broad term, but it usually refers to one of three things: reducing or eliminating compute demand during peak times on a grid, moving that compute to other locations on a network to relieve stress at a given point, or keeping compute the same while drawing on collocated generation or storage to make up for curtailed grid power. The options all have their costs and limits to the data center operator, but utilities are increasingly demanding some level of flexibility in exchange for faster interconnection to their grids.
Important recent developments include:
- A partnership between Nvidia, Emerald AI, EPRI, Digital Realty, and PJM announced their Aurora AI Factory, the first data center built to a new reference design for power-flexible AI infrastructure. This represents the mode of flexibility where the data center can respond to grid constraints dynamically, drawing on local storage or managing workloads.
- In late October, the DOE directed the Federal Energy Regulatory Commission to expedite the interconnection of large loads that agree to be curtailable, as well as colocated facilities that are both curtailable and dispatchable. This comes as part of a broader ANOPR on potential reforms to standardize large load interconnection, especially for data centers to power artificial intelligence. Lots of work yet to be done here.
- Vistra’s CEO Jim Burke stated in a recent earnings call that he sees enough capacity in off “super peak” hours to meet demand from data center companies seeking to interconnect to ERCOT and PJM, if they employ onsite storage and demand response flexibility.
- While flexibility most often refers to the site of the data center and its interconnection point to the transmission grid, there has been increasing focus on leveraging the capacity of the distribution grid as well. WattCarbon has recently launched the Repowering California program to incentivize large energy users to invest in VPPs to unlock capacity or reduce peaks on grids where they plan to locate data centers. Voltus, more focused on the commercial and industrial market, recently launched its bring-your-own-capacity product with Cloverleaf Infrastructure. Both of these options are designed to speed interconnection by accrediting and certifying the capacity made available through VPPs.
These developments are an important signal that the entire AI-energy nexus is evolving to what is hopefully a stage of increased understanding and collaboration among participants. If the relationship between utilities and data center operators is moving to one where costs are fairly allocated, flexibility is non-negotiable, and investment is centered on expanding the grid’s capabilities rather than eschewing it entirely, then the AI boom could drive a level of grid modernization this industry’s been pushing for since 2005. This feels worth encouraging.
In contrast, an alternative model where hyperscalers build and own power plants, operating them on islanded microgrids because public infrastructure and regulations aren’t fit for purpose, would be a failure of will and purpose. If this is the next economic revolution, and it is built entirely on a marriage of bits and electrons to support infinite compute, then shouldn’t our power system be an equal partner?
A version of this story was published in the AI-Energy Nexus newsletter on November 12. Subscribe to get pieces like this — plus expert analysis, original reporting, and curated resources — in your inbox every Wednesday.
