Look Forward — 2 December 2025
Navigating the US data center power crunch: On-site solutions offer a faster path to power
Surging electricity loads from data centers, electrification and manufacturing are outpacing grid capacity, prompting a shift toward customer-sited energy resources and capabilities to address data center energy demand.
By Ben Levitt and Doug Giuffre
Highlights
Although pockets of excess generating capacity exist across the US, data centers' impact on load growth is expected to exceed grid-based power generation and transmission capacity.
Limits to grid-based power generation and transmission capacity do not mean data center power demand will go unmet. Instead, technology, market, and policy approaches that deliver power supply for the grid's most constrained hours and locations will command a premium, as will strategies that quickly provide power independent of the grid.
Gas-based power-generating resources, such as gas turbines, are well positioned to meet current needs and have seen renewed demand amid the data center boom. Battery energy storage solutions also offer scalable near-term capacity but depend on evolving market rules and requirements.
With concerns rising over the slow pace of grid-connected capacity expansion, customer-sited energy resources and capabilities offer a faster path to power. Expansion of these resources may continue long term, especially as data center campuses grow.
US electricity demand is set to surge, driven by data centers, electrification, and the reshoring of manufacturing. These factors are expected to lift US grid-based electricity consumption by 17% by 2030 compared to 2025 levels, and potentially by 23% to 25% if data center operators achieve their goals. Grid resources alone are unlikely to meet this pace or maintain the reliability consumers expect. For the first time in decades, the expansion of large-scale electricity consumers will be limited by grid power availability and deliverability. Stakeholders are working to stretch existing grid assets, but these efforts are not matching the speed and scale of AI-driven data center energy demand. In this tight market, companies, states, and grid operators have an opportunity to meet the needs of large energy users and regional grids. With grid-connected capacity slow to expand, customer-sited energy resources and capabilities are positioned to accelerate power delivery.
Despite apparent excess capacity, US power grids still face a near-term power crunch
US power grids have significant excess generating capacity to accommodate load growth throughout most of the year. In many industries, average capacity utilization is used as a measure of market tightness. In the power sector, coal and natural gas combined-cycle fleets averaged 50% to 55% utilization over the past decade. This suggests incremental new loads could be accommodated by increasing utilization of the existing fleet. However, power markets require supply and demand to stay balanced at all times, from peak summer afternoons to cold winter mornings. To estimate how much new load US power grids can integrate, it is essential to understand the supply-demand balance during the most constrained hours.
During the most constrained hours, excess generating capacity across the US totals about 70 GW, slightly greater than California’s peak electricity demand. Over the next few years, surplus generating capacity is expected to remain within this range, based on the current pipeline of new projects and the outlook for generation retirements and load growth outside of data centers.
At first glance, with a pipeline of about 85 GW of new data center capacity requests expected by 2030, US power markets face a seemingly modest generating capacity deficit of 15 GW. However, data center operators face a much higher level of grid constraints for several reasons. First, to reliably support an additional 85 GW of demand, approximately 100 GW of total capacity, including surplus, is required. Another reason is that surplus generating capacity is not well aligned with expected data center growth on a regional basis. A disproportionate share of expected data center capacity is planned for some of the most constrained grids, such as PJM Interconnection. PJM, named for the original three core states of Pennsylvania, New Jersey and Maryland, is the largest US Regional Transmission Organization; it operates the electric grid and competitive wholesale electricity market for all or parts of 13 Eastern US states and the District of Columbia. PJM surplus capacity has declined sharply in recent years, with recent record-high-capacity auction outcomes pointing to persistent near-term tightness. Upside risk to load growth from other industrial demand — particularly in the grid managed by the Electric Reliability Council of Texas (ERCOT) — threatens to exacerbate the capacity deficit in that region as well.
Even within a market, spare generating capacity may not be immediately deliverable to new data centers because of regional transmission grid constraints. As the size of new data center facilities or campuses grows — reaching hundreds of megawatts or even gigawatt scale within a few years — the concentration of power demand strains local transmission systems. This may require extensive, time-intensive upgrades, including installation of large transformers and breakers.
Limits to grid-based power generation and transmission capacity do not mean data center power demand will go unmet. Instead, technology, market, and policy approaches that deliver power for the grid's most constrained hours and locations will command a premium, as will solutions that quickly provide power independent of the grid. Innovative companies, utilities, and states that can deliver with speed — a key requirement for data center customers — will be best positioned in this constrained environment.
Gas turbines, battery storage and delayed generation retirements are important but only part of the near-term solution
Gas turbines are among the technologies well-positioned to meet current needs and a segment with renewed demand amid the data center boom. In 2024, the same year gas additions hit rock bottom, orders for large new gas turbines hit a 20-year high of 14 GW, driven by data center load growth. This trend accelerated into 2025, with 18 GW of US-bound gas turbine orders placed in the first half of the year.
Although orders for large turbines have surged over the past 18 months, supply constraints and lengthy project timelines will push much of this new capacity into production in the late 2020s and early 2030s. The power crunch appears most severe through 2028 and 2029, when supply begins to catch up.
For example, PJM’s fast-tracking initiative, completed earlier this year, shows that most new gas capacity will not be in service before 2029, with uprates to existing facilities providing the bulk of near-term additions. Similarly, state-led efforts in Texas to expedite new gas projects are expected to have limited impact until 2029.
Overall, we expect 55 GW to 65 GW of new grid-based gas capacity in the US from 2025 to 2030, roughly double our expectations prior to the data center boom, but still short of incremental needs.
Beyond large gas turbines, utilities and regional power markets have raced to shore up near-term capacity by extending existing resources. Since the start of 2023, plans to retire over 37 GW of coal capacity (one-third of all previously announced coal retirement capacity) have either been delayed or avoided through planned conversions to gas-fired generation — including at least 10 GW since the start of 2025. Deals to restart recently retired nuclear power plants and uprate operating ones are set to deliver around 3 GW of incremental nuclear capacity by 2030. Uprates at existing gas-fired plants will also contribute. Even the restart of recently retired coal plants — likely remade as gas-fired resources — is now under consideration, a scenario unimaginable just a few years ago but revived by tight market conditions and supportive federal actions.
Although grid-connected battery energy storage system (BESS) capacity is poised for significant growth, the data center boom has so far given this technology only a modest 8 GW or 15% uplift through 2030. Faster timelines give BESS an edge relative to other resources, but grid-connected BESS still faces long development queues, while tougher economics persist outside of Texas, California and the Southwest. Still, with scalable near-term capacity in short supply, BESS can enable data centers to capitalize on the potential emergence of new market processes to fast-track grid connections in exchange for self-supplied capacity.
Most incremental near-term capacity will come from new generation capacity and retirement delays, but demand-side strategies — such as deploying enhanced network-utilization technologies, repurposing existing crypto-currency mining facilities into traditional data centers, or adopting advanced chip-cooling techniques to reduce facility peak cooling demand and improve data center energy efficiency — may help avoid some incremental capacity needs. Expanding existing demand-response programs is another option, but it remains uncertain whether market prices can incentivize data center operators to curtail grid consumption, even temporarily, due to operational complexity and the enormous cost of potential service interruptions.
Customer-sited solutions expedite power access amid grid delays
Data center operators overwhelmingly prefer to rely on grid power, but heavily regulated electric utilities — burdened by rising affordability pressures — struggle to respond quickly to growth demands. Key competitive wholesale electricity markets also face challenges to attract sufficient near-term investment due to frequent rule changes and administrative price caps that hurt investor confidence.
As concerns grow over the slow pace of grid-connected power supply and transmission capacity expansion, interest has shifted to customer-sited energy resources and capabilities, where load flexibility or primary power supply can make capacity quicker to obtain.
Load flexibility, combined with supportive market rules, enables faster grid connections through demand management
For data centers, load flexibility means dynamically restraining electricity consumption through workload management — such as rescheduling, slowing or re-routing IT tasks to other data centers — or temporarily relying on an on-site power supply during periods of grid stress. If enabled by supportive market rules and processes, flexibility can speed load interconnection by reducing the need for new grid infrastructure.
A range of resources can enable data center load flexibility, from backup diesel generators to on-site battery energy storage solutions. These approaches can be tailored to site-specific factors such as commercial requirements, emissions limits or operator expertise. Instead of waiting for new grid upgrades, flexible data centers that assume delivery risk — the risk that electricity will not be deliverable during periods of grid stress — can position themselves for faster grid connection.
However, making IT workloads grid-responsive remains novel in data center operations and design. In practice, apart from Google, few IT companies have successfully integrated data center load flexibility into utility plans on a large scale. Microsoft’s decade-old collaboration with Black Hills Energy used backup diesel generators to help avoid major investment in new generating resources. Future use cases must continue to address emissions limits, refueling logistics, and potential community opposition.
Moreover, although flexibility addresses the key data center bottleneck of grid-based power generation and transmission capacity, most states and markets do not yet offer a faster path to grid connection for flexible loads. Many of the load flexibility options that states, utilities, and operators are developing for large energy consumers remain undefined and do not yet have regulatory approval.
It seems clear, though, that as new load flexibility rules evolve, especially if they impose more stringent availability requirements, they will favor on-site resources and configurations that rely more on gas-based technologies than on energy-limited resources like shorter-duration BESS.
On-site power expedites power access until a permanent grid connection is established
Because load flexibility does not yet offer a clear path to faster grid connections and requirements remain uncertain, many data center operators are looking to deploy more on-site resources to provide a fully independent power supply, at least temporarily until a permanent grid connection is established.
Since late last year, specialists like Crusoe, Frontier Infrastructure, and Enchanted Rock have unveiled plans to colocate over 26 GW of power supply with data centers by the early 2030s, with potential for more. Clusters have emerged in Texas and the Mid Atlantic (PJM), which are strong natural gas producing regions and have the greatest expected grid-capacity deficit.
On-site power often features smaller generators — such as small gas turbines, fuel cells and reciprocating engines — which are currently favored for their speed, scalability, modularity and availability. Many data center operators envision their on-site power assets shifting to backup roles after a grid connection is secured, while others are considering deeper integration of on-site resources with regional markets. For now, while most on-site generation projects remain in development, several gigawatts are already operating, under construction or backed by firm orders for gas resources.
In the longer term, the rising scale of data center campuses could redefine optimal power supply configurations for highly concentrated loads. If campuses reach gigawatt-plus scale, as many forecast, the total power supply needs of a single campus could rival the size of smaller North American grids. Fermi America’s proposed 11-GW Texas campus, for instance, is comparable to Alberta, Canada’s peak demand. Such massive and concentrated power demands will almost certainly require local power supply, similar to other large industrial configurations like oil sands operations with on-site cogeneration.
Looking forward
While pockets of excess generating capacity exist across the US, surging load growth is outpacing available grid infrastructure. However, this supply-demand imbalance won't leave data center operators without options.
Energy solutions that address peak-hour bottlenecks and serve transmission-limited areas will attract premium pricing. Similarly, autonomous power systems that bypass grid dependencies and accelerate deployment timelines will prove valuable in meeting escalating data center energy demand.
On-site resources and agile specialists can play a major role in meeting the needs of large new energy consumers and regional grids. Over the long term, this dynamic may lead to more integration of planning and operations among data centers, utilities and regional transmission organizations.
This article was authored by a cross-section of representatives from S&P Global and in certain circumstances external guest authors. The views expressed are those of the authors and do not necessarily reflect the views or positions of any entities they represent and are not necessarily reflected in the products and services those entities offer. This research is a publication of S&P Global and does not comment on current or future credit ratings or credit rating methodologies.
Content Type
Look Forward Council Theme