Powering America: How Will the Generation Stack Evolve?

In a new Super-Sector Analysis report from Citi Research, a team of analysts led by Ryan Levine forecasts that U.S. power demand will grow at a 2.8% compound annual growth rate (CAGR) over the next 15 years, vs. ~0.5% in the previous decade. In terms of demand, data-center load growth is the largest contributor of power demand growth at 0.9%, followed by industrial load, electric-vehicle (EV) load, commercial load and residential load. In terms of the independent system operators (ISOs) and their multi-state grids, we see growth being led by the Electric Reliability Council of Texas (ERCOT) at 3.1%, followed by the Midcontinent Independent System Operator (MISO) at 2.8%, the Pennsylvania-New Jersey-Maryland (PJM) Interconnection at 2.8%, the California Independent System Operator (CAISO) at 2.7%, and Other at 2.6%.

We estimate that the U.S. generation mix will evolve from 41% gas in 2025 to 30% by 2040 and 12% solar to 33%, with nuclear remaining at 7% on a Megawatt (MW) basis. By 2040 we expect ~96 Gigawatts (GW) of gas, net of retirements; ~54 GW of nuclear; ~625 GW of solar; ~55 GW of wind; ~262 GW of battery storage; ~2 GW of geothermal; and ~7 GW of other resources to be online to meet the demand outlined above.

Our projection suggests that baseload-generation additions will roughly match incremental around-the-clock power demand from data centers, advanced manufacturing and a portion of commercial and industrial load. Accordingly, intermittent-renewables additions including solar, wind and storage will roughly match incremental residential and EV load.

The regional picture

With U.S. power demand at an inflection point given stagnant growth over the last two decades, we developed a systemic approach to forecast power demand, one that leverages cross-sector Citi Research projects, meetings with ISOs, those ISOs’ regional forecasts, correlation analysis of economic indicators and load drivers, government public disclosures and third-party data providers, and implications of the One Big Beautiful Bill and recent executive orders.

While the biggest drivers of power demand are data centers, advanced manufacturing and EVs, we also took a closer look at other commercial, industrial and residential growth, crypto load and other factors.

The topic that generated the most debate was the regional breakdown of power growth, as all states and regions are competing to capture this growth through new legislation, regulation, commercial negotiations and other policymaker actions.

The primary driver of growth in U.S. power demand is the rapid growth of artificial intelligence (AI) and the related data-center buildout (See Heath Terry’s AI study.) In 2023 and 2024, we estimated historical data-center load by regions and markets; our projection shows that PJM had the largest share of data-center load at 45% or 12.7 GW in 2024, largely driven by northern Virginia, Chicago and Columbus, Ohio, followed by Other regions at 32% or 8.9 GW. Our 2024 data-center load estimates for ERCOT, CAISO and MISO were relatively small compared with these two areas.

Looking ahead, among the different regions we think ERCOT is likely to gain the most shares (~9 GW from 2025 through 2030) given additional clarity after the passage of Texas’s Senate Bill 6 and the potential availability of new gas generation. We expect PJM to keep growing at a steady pace (~8.4 GW from 2025 through 2030), primarily driven by established data-center clusters in northern Virginia and Ohio. We think MISO could capture ~4.8 GW load by 2030 as requests for new power are increasing and its Expedited Source Addition Study was approved, which should speed growth in select pockets of the Midwest. We project ~1.8 GW incremental load in California by 2030, as we see near-term growth as constrained by fires, rate making, local opposition and questions about CAISO reliability.

To reflect development risk, we took the official data-center load forecasts published by ISOs and applied different haircut rates — in effect, “trimming” the published forecasts to match potential utility load with our projected capacity ceiling for chip supply. 

Another potential long-term driver of growth in power demand is advanced manufacturing, driven by reshoring trends and the proliferation of advanced technology. We forecast between ~1.5 GW and ~3.0 GW per year of incremental advanced-manufacturing load through 2030, largely driven by Other regions, ERCOT and MISO. 

We built a model to estimate total U.S. power demand from EVs. Our estimate is that total EV demand may reach ~75 TeraWatt hours (TWh) in 2030, growing to ~434 TWh in 2040, implying a 27% CAGR from 2020 to 2040. 

An important note: While data centers, advanced manufacturing and EVs are long-term potential drivers of power demand, most of this demand is still expected to be driven by residential, commercial and industrial customers. 

Where is the power going to come from?

From a power-generation perspective, the U.S. market is constrained by the availability of resources and while keeping a manageable level of reserve margins at each ISO. Each power-generation resource is governed by utility commissions, integrated resource plan processes, state politics, merchant power price economics, federal policy and the supply chain. 

In creating a forecast of the U.S. power generation mix from 2025 through 2040, we offer several key takeaways.

The buildout of natural-gas-fired generation should pick up. Gas-fired generation is the only baseload resource that can be built a few years, it’s relatively affordable, and it has lower construction risks. As demand for gas generation picks up, the supply chain for gas turbines is becoming one of the main constraints. At its Development Day this year, NextEra indicated that the power supply chain can only support 75 GW of gas through 2030. In response, utilities and power companies have formed strategic partnerships with gas-turbine suppliers to minimize execution risk.

On-site generation may fill a near-term gap. Over the past decade, oilfield-services companies have deployed small-scale mobile turbines and gas reciprocating engines in order to convert natural gas into power to run hydraulic fracking pumps. This was done to save fuel and maintenance costs compared to legacy diesel engines. Now, these companies are looking to deploy these technologies at data centers and other industrial applications as a bridge solution, with them being seen as part of a redundant power-supply base longer-term. Oilfield-services companies are looking to sign multiyear power-purchase agreements, during which time they’ll provide primary power for either a flat rate or fixed spark spread. After grid connect, the equipment can be used as an additional, localized source of power, as backup power, transferred to another data center awaiting grid connect, or even operate as a “peaker” plant selling power to the grid during elevated demand periods.

Midstream companies look to capture value. Midstream companies have also moved down the value chain to capture what appear to be superior economics and new avenues for growth, though approaches have been quite varied. 

Coal retirements are set to accelerate. When we talk to utility management teams, the universal feedback is that they’d prefer to replace their old, inefficient coal power plants with new resources, but are getting requests from politicians and regulators to extend these plants’ lives, with factors at work including jobs and resource adequacy. Our assumption is that coal retirements will occur according to plan in the next few years, and accelerate starting in 2030.

What’s in store for offshore wind? Offshore wind development in the U.S. is mainly concentrated in PJM and Other regions, but is facing challenges from permitting and policy changes arriving with the Trump administration, as well as supply-chain issues, cost inflation and overruns, and uncertainties related to commercial agreements. Our forecast assumes the majority of ~3 GW projects under construction will come online, albeit with some timeline shifts, but we’re more conservative about other proposed projects.

What’s the solar generation outlook? While the One Big Beautiful Bill made changes to key incentives related to utility-scale solar, levelized cost-of-energy and capital-cost estimates show solar remains competitive with traditional energy-generation technologies even without incentives. This should mean the technology remains attractive. We model an acceleration in near-term solar installation growth given incentives to put projects in service by 2027. Data-center demand and quick deployment times are additional tailwinds for this growth. Looking beyond 2027, we model a decline followed by gradual growth, with another leg down after 2030.