Goldman Sachs: U.S. Data Center Power Demand Could Reach 66 GW by 2027
AI summary card
Goldman Sachs: U.S. Data Center Power Demand Could Reach 66 GW by 2027
U.S. data center power demand is projected to surge from 31 GW in 2025 to 66 GW in 2027, accounting for 8.5% of summer peak demand and significantly tightening power markets such as PJM.
- Total U.S. data center capacity is expected to reach 95 GW by end-2027, more than doubling from end-2025 levels
- Data center power demand is forecast to hit 66 GW in 2027, representing 8.5% of U.S. summer peak power demand
- PJM (Mid-Atlantic) and MISO (Midcontinent) markets face severe reliability risks with extremely tight power balances
- ERCOT (Texas) remains only marginally tight due to rapid supply growth
- Recommend hedging against both upside price risks in regions like PJM and downside price risks from potential oversupply in regions like ERCOT
Report interpretation
Overview
This report, published by Goldman Sachs, argues that U.S. data center power demand will experience substantial acceleration during 2026–2027. By comparing historical planned capacity versus actual activated capacity, the report forecasts that despite delays and cancellations, total U.S. data center capacity will reach 95 GW by end-2027. This growth will tighten the overall U.S. power market and amplify regional disparities: already constrained markets like PJM will face heightened reliability risks, while regions with faster supply growth like ERCOT will remain relatively loose.
Core views
Accelerating Demand: The report projects U.S. data center capacity will more than double from end-2025 levels to reach 95 GW by end-2027. Annual new capacity additions will accelerate from approximately 8.5 GW in 2026 to 36 GW in 2027. Assuming a 70% capacity utilization rate, data center power demand will rise from 31 GW in 2025 to 41 GW in 2026 and 66 GW in 2027. By 2027, data centers will account for 8.5% of U.S. summer peak power demand. Widening Regional Divergence: Data center growth will lead to stark differences in power market tightness across U.S. regions. In PJM (Mid-Atlantic), MISO (Midcontinent), and the Northwest, limited planned generation additions combined with rapid data center growth will further deteriorate power balances, potentially forcing some future data center projects to be denied interconnection. In contrast, ERCOT (Texas), SPP (South Central), and the Southeast will see only marginal tightening due to ample planned generation additions, maintaining relatively loose conditions overall. Forecast Discrepancies and Uncertainties: Goldman Sachs’ forecast (based on Aterio data) exceeds that of its equity research colleagues using S&P 451 data (39/50 GW for 2026/27). This gap reflects differences in data sources and methodologies for adjusting potential delays and cancellations. The report also notes high uncertainty due to the nascent stage of the AI wave, limited historical data, and variable construction timelines that could shorten with higher capex or lengthen due to supply chain issues.
Analysis framework
The report employs a 'planned-to-realized' conversion rate analysis. It first obtains a detailed timeline of U.S. data center development projects (from Aterio data) and then examines the historical relationship between 'planned activation capacity' and 'actual activation capacity.' The study finds that only about 72% of data centers planned for activation within the next four quarters come online on schedule, with activation likelihood declining as lead times increase. Applying this historical conversion pattern, the report discounts future planned capacity to generate more realistic forecasts for capacity additions and power demand. This approach aims to filter out overly optimistic planning noise and provide a more robust demand outlook.
Methodology notes
Regional Power Supply-Demand Balance Analysis
The report assesses power market tightness and price risk direction in different regions (e.g., PJM vs. ERCOT) by analyzing the alignment between new generation supply and new data center demand.
Adjustment for Deviation Between Planned and Actual Capacity
The firm notes that publicly announced 'planned capacity' often exceeds 'actual realized capacity.' It uses historical data to calculate conversion rates (e.g., 72% on-time activation rate) to adjust expectations and avoid overestimating future demand.
Key data
- Forecast Total Data Center Capacity by End-202795 GWMore than double the end-2025 level
- Forecast Data Center Power Demand in 202766 GWBased on a 70% capacity utilization assumption
- Share of Peak Demand Attributable to Data Centers in 20278.5%A significant increase from 4.1% in 2025
- Forecast Annual New Capacity Addition in 202736 GWA sharp acceleration from 13.6 GW in 2026
- On-Time Activation Rate for Planned Capacity72%For projects planned for activation within the next four quarters
Impact & implications
The report concludes that surging data center power demand will progressively tighten the U.S. power market and reinforce structural regional divergence. In already critically tight markets like PJM, reliability risks will rise significantly, potentially driving localized upward price pressure. Conversely, in rapidly expanding supply regions like ERCOT, temporary oversupply relative to load growth could create downward price pressure. Therefore, the report recommends investors hedge against these opposing directional risks rather than taking a unidirectional position.
Risks
- Limited historical data due to the early stage of the AI wave may cause future development patterns to deviate from historical trends
- Supply chain and labor constraints could delay construction, while high capital expenditure might shorten build cycles—both introducing forecast uncertainty
- As time progresses, newly announced projects could push actual realized capacity beyond original forecasts
What to watch
- Approval and construction progress of new generation capacity in tight regions like PJM and MISO
- Relocation trends of data center developers across regions (e.g., toward Texas and Georgia)
- Regional power price volatility and breaches of reliability thresholds (e.g., 15% reserve margin)