Goldman Sachs raises U.S. power demand CAGR, with data centers becoming the core incremental source
AI summary card
Goldman Sachs raises U.S. power demand CAGR, with data centers becoming the core incremental source
Goldman Sachs introduced a regional U.S. data center supply-demand model, raised its 2030 data center power demand forecast to about 108 GW/947 TWh, and increased U.S. total power demand CAGR from 3.2% to 3.5%.
- U.S. data center power demand is expected to reach about 108 GW/947 TWh by 2030, implying about 23% CAGR in 2025-2030.
- PJM remains the largest regional market; MISO is expected to surpass ERCOT by 2030 and become the second-largest ISO data center market.
- Behind-the-meter power supply assumptions were raised significantly, with about 31 GW of capacity and about 22 GW of power delivery available to serve data center load by 2030, expected to meet 20% of data center demand.
- Beneficiaries include regulated utilities FE, XEL, DUK, SRE, AEP, as well as independent power producers TLN, VST, and NRG.
Report interpretation
Overview
The report updates the U.S. power supply-demand framework, with the core change being the introduction of a regionally segmented data center supply-demand model and an upward revision to the outlook for U.S. data center and total power demand. Goldman Sachs expects U.S. data center IT Power average annual capacity to reach about 121 GW by 2030, and utility power capacity adjusted for PUE to reach about 159 GW; combining regional occupancy and utilization assumptions leads to a forecast of about 108 GW/947 TWh of data center power demand.
Core views
Data centers are the key driver of upward revisions to U.S. power demand. PJM will remain the largest market, with demand of about 38 GW by 2030; MISO is among the fastest-growing regions, with demand share expected to rise from 9% to 16% and surpass ERCOT; ERCOT will still rank among the top three markets, but progress on interconnection queue reforms such as Batch Zero needs to be monitored. Among non-ISO regions, the Southeast, Southwest, and Northwest will also contribute meaningful incremental demand. Behind-the-meter power supply is receiving more near-term attention due to interconnection bottlenecks and customer time-to-power needs, but over the long term customers still prefer grid-connected power solutions.
Analysis framework
The report uses a bottom-up facility-level model based on the S&P Global 451 Research Datacenter Knowledgebase to estimate regional data center IT Power capacity, then converts it into Utility Power capacity using facility PUE, and derives actual power demand based on regional occupancy, vacancy, and utilization rates. The macro power demand section incorporates updated GS Macro forecasts for variables including housing, GDP, unemployment, corporate profits, investment, and industrial production, as well as the autos team's latest EV forecasts.
Methodology notes
Segment U.S. data center capacity and demand by power market
The model is segmented across PJM, ERCOT, MISO, CAISO, SPP, NYISO, ISO-NE, and non-ISO regions, reflecting differences in PUE, occupancy, utilization, and planned capacity by region.
Convert from IT Power to Utility Power
PUE measures data center energy efficiency. The report assumes an average U.S. PUE of about 1.4 and adjusts IT Power capacity by facility or regional differences into total power capacity that utilities need to serve.
Derive actual electricity demand from available capacity
The report assumes average U.S. data center occupancy of about 97% and average utilization of about 70%, and uses different assumptions by region to derive actual data center power demand of about 108 GW in 2030.
Residential, commercial, and industrial power demand forecasting
Residential demand references housing stock, vacancy rates, housing starts, and domestic demand; commercial demand references GDP, unemployment, corporate profits, and construction investment; industrial demand references industrial production, housing starts, single-family completions, and manufacturing reshoring.
Asset mapping & comparison
Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).
- FEBuy-rated PJM beneficiary utility
- Strengths
- PJM currently accounts for about 34% of U.S. data center demand, and this region remains the largest data center power market.
- Weaknesses
- Constrained by local permitting, interconnection, and power affordability.
- Comparison
- Compared with other regulated utilities, its PJM exposure makes it a more direct beneficiary of the existing high-demand base.
- Risks
- Data center project delays, regulatory resistance, and weaker-than-expected capital recovery.
- XELBuy-rated beneficiary of MISO data center demand
- Strengths
- MISO demand share is expected to rise from 9% to 16% and surpass ERCOT by 2030.
- Weaknesses
- MISO's absolute demand base currently remains below PJM.
- Comparison
- Compared with mature markets, MISO's appeal lies in faster growth and rising regional share.
- Risks
- Uncertainty around regional load realization, interconnection progress, and approval of grid investment.
- DUKBuy-rated utility benefiting from non-ISO Southeast demand
- Strengths
- The Southeast is expected to become the largest non-ISO data center market, with demand CAGR of about 20%.
- Weaknesses
- Project deployment in non-ISO markets depends on planning and regulatory approval by vertically integrated utilities.
- Comparison
- Compared with ISO markets, the non-ISO Southeast benefits from in-state capacity planning and the vertically integrated model.
- Risks
- Risks related to project siting, regulatory recovery, labor, and local opposition.
- SREBuy-rated, benefiting from ERCOT and CAISO exposure
- Strengths
- ERCOT remains one of the top three data center markets, and data center opportunities in CAISO are also improving.
- Weaknesses
- There is debate over speculative assumptions in ERCOT load forecasts.
- Comparison
- Compared with PJM/MISO names, SRE's regional mix of ERCOT and CAISO is more important.
- Risks
- Batch Zero progress, interconnection queues, actual regional load realization, and regulatory uncertainty.
- AEPBuy-rated PJM beneficiary utility
- Strengths
- PJM and related states have a high concentration of data center capacity, with clear power demand growth.
- Weaknesses
- Demand growth may slow after 2028.
- Comparison
- Similar to FE, the core advantage comes from PJM market exposure.
- Risks
- Risks around the pace of load growth, infrastructure buildout, and regulatory recovery.
- TLNBuy-rated beneficiary among independent power producers
- Strengths
- Can benefit through higher power prices and incremental data center PPAs.
- Weaknesses
- Returns depend more on power prices and contract signing than on expansion of regulated asset base.
- Comparison
- Compared with regulated utilities, IPPs have greater sensitivity to wholesale power prices and PPAs.
- Risks
- Declining power prices, weaker-than-expected PPA signing, and fuel and regulatory risks.
- VSTBuy-rated beneficiary among independent power producers
- Strengths
- Benefits from demand growth in PJM and ERCOT power markets and from data center PPA opportunities.
- Weaknesses
- Earnings are more sensitive to power price cycles and market volatility.
- Comparison
- Like TLN and NRG, it is more of a power-price sensitivity story.
- Risks
- Power price volatility, changes in capacity market rules, and weaker-than-expected project contracting.
- NRGBuy-rated beneficiary among independent power producers
- Strengths
- May benefit from stronger power prices and PPAs with data center customers.
- Weaknesses
- The report does not provide detailed standalone company-level forecasts.
- Comparison
- Compared with other IPPs, it belongs to the same basket benefiting from data center power demand-driven pricing and contract opportunities.
- Risks
- Risks around wholesale power prices, contract execution, regulatory policy, and load realization.
- WECSell-rated but with exposure to MISO growth
- Strengths
- MISO data center demand is growing relatively quickly.
- Weaknesses
- Goldman Sachs believes the related growth is already reflected in valuation.
- Comparison
- Compared with Buy-rated beneficiary names, WEC has less attractive risk-reward.
- Risks
- Valuation pressure, weaker-than-expected demand realization, and regulatory return risk.
Key data
- U.S. total power demand CAGR3.5%Raised from the previous 3.2%, including behind-the-meter data center demand; grid-delivered power demand CAGR remains 2.8%.
- U.S. data center power demand in 2030about 108 GW / 947 TWhHigher than the previous forecast of 726 TWh, implying demand CAGR of about 23% in 2025-2030.
- Data center IT Power capacity in 2030about 121 GWYear-end figure is about 125 GW.
- Data center Utility Power capacity in 2030about 159 GWYear-end figure is about 165 GW, converted from IT Power based on PUE.
- Incremental U.S. data center supply and demand in 2026-2030Supply about 101 GW, demand about 69 GWThe difference is mainly driven by occupancy and utilization assumptions.
- Average PUEabout 1.4Regional climate, facility age, and cooling technology can affect PUE.
- Average occupancy and utilizationabout 97% / 70%Regional utilization assumptions are roughly in the 55%-80% range.
- Behind-the-meter power supply assumption in 2030Capacity about 31 GW, power delivery about 22 GWPreviously about 20 GW and about 14 GW, respectively; expected to meet 20% of data center demand.
- PJM demand in 2030about 38 GWIncremental data center Utility Power supply/demand in 2026-2030 is about 36 GW/25 GW.
- ERCOT demand in 2030about 15 GWIncremental supply/demand in 2026-2030 is about 14 GW/9 GW.
- MISO demand in 2030about 17 GWIncremental demand in 2026-2030 is about 14 GW, with demand share expected to rise to 16%.
Impact & implications
This report reinforces the case for data center power demand as a core investment theme for U.S. utilities. For regulated utilities, data center load growth may support capital expenditure, infrastructure investment, and long-term electricity volumes; for independent power producers, stronger power prices and data center PPA opportunities may drive earnings upside. At the same time, local opposition, labor constraints, power affordability, and interconnection queue reform will determine the pace at which demand materializes.
Risks
- Rising opposition from local communities to data center construction.
- Concerns about power affordability may affect regulation and project approvals.
- Labor shortages may slow the construction progress of grids and data centers.
- Delays in interconnection queues and regulatory processes may postpone load energization.
- Load forecasts in regions such as ERCOT may include speculative assumptions.
- Not all announced behind-the-meter power supply may convert into contracted, deliverable capacity.
- Over the long term, customers may prefer grid-connected power, making behind-the-meter supply more of a short-term bridge solution.
What to watch
- Whether incremental PJM data center demand peaks around 2028 as expected.
- Whether MISO surpasses ERCOT by 2030 as forecast in the report, becoming the second-largest ISO data center market.
- Whether ERCOT Batch Zero and related catalysts in July-August improve visibility on near-term load energization.
- Whether the SPP HILL framework effectively alleviates bottlenecks in large-load interconnection studies.
- The conversion rate of behind-the-meter power project announcements into actual capacity, power delivery, and PPAs.
- Progress by regulated utilities in rate base, capital expenditure, and regulatory recovery.
- Changes in how data center customers trade off time-to-power, cost, and power reliability.