Data Center Expansion to Double U.S. Power Demand by 2027

Overview

Goldman's power team, based on Aterio project pipeline data, predicts a significant acceleration in U.S. data center construction from 2026 to 2027, leading to a structural rise in national power demand. Even considering delays and cancellations, capacity is expected to increase from approximately 40 GW at the end of 2025 to 95 GW by the end of 2027, with corresponding power demand rising from 31 GW to 66 GW, or from 4.1% to 8.5% of the national summer peak. The report emphasizes regional differences: regions like PJM (Mid-Atlantic) and MISO (Mid-Continent) that are already strained will face higher reliability risks, while supply growth regions like ERCOT (Texas) and SPP (Mid-South) will only marginally tighten.

Core views

Demand side: Data centers are the largest single source of U.S. power demand growth from 2026 to 2027. Supply side: Project pipelines show 19 GW and 69 GW of new capacity in 2026 and 2027, respectively, far exceeding the 6.4 GW and 8.5 GW achieved in 2024 and 2025. However, historical data suggests only 60-70% of planned capacity is likely to come online as scheduled. Regional differences: Regions like PJM, MISO, and NW are expected to have reserve margins below the 15% reliability threshold, while ERCOT and CAISO remain above it. Risk considerations: Construction timelines, supply chain issues, local approvals, and power access bottlenecks could lead to project delays or cancellations, but there is also a possibility of more projects being added than expected.

Analysis framework

The report uses a two-step 'pipeline-realization' model: first, it uses Aterio's database of data center project schedules to obtain 'raw pipeline capacity', then it adjusts for historical realization probabilities based on the lead time to project commissioning. Finally, it converts the capacity to power demand using a 70% historical average load factor and cross-checks it with regional reserve margin scenarios to assess regional supply-tightness.

Methodology notes

• Industry/Industry Analysis FrameworkSupply-demand framework

  Predicting future supply based on project pipelines and historical realization rates, then combining it with demand-side load rates to assess regional power balance

  The report uses 'planned capacity × realization probability × load factor' as the core formula, accounting for the real probability of project implementation while maintaining demand-side flexibility, thus closely aligning with actual power market tensions.

• Cycle and Scenario FrameworkCapacity/Equipment Cycle (Kuznets)

  Data center construction cycles take about 1.5-2 years, which is a typical capacity expansion cycle, helping to predict the peak of projects in 2026-2027

  Understanding the construction cycle aids in explaining why the investment peak in 2024-2025 will translate into actual power load in 2026-2027, allowing for early prediction of grid pressures.

Key data

• U.S. Data Center Capacity95 GWExpected by the end of 2027, double that of the end of 2025
• Data Center Power Demand66 GWExpected for 2027, accounting for 8.5% of the national summer peak
• 2026 New Capacity Prediction13.6 GWAfter considering a 60% realization rate
• 2027 New Capacity Prediction36 GWAfter considering a 50% realization rate
• PJM Reserve Margin 2027E1.2%Below the 15% reliability threshold
• ERCOT Reserve Margin 2027E26.9%Above the reliability threshold

Impact & implications

The report suggests that data center expansion will exacerbate regional power market differentials in the U.S.: in regions like PJM and MISO that are already strained, the risk of power price increases will rise, potentially leading to the rejection of new data center projects; in supply adequate regions like ERCOT, data centers will still have site selection advantages. Overall, the U.S. power system will face an additional 35 GW of base load demand increment in 2026-2027, equivalent to two and a half Three Gorges Dams. Key constraints will be grid investments, power resources, and transmission bottlenecks.

Risks

• Short historical data for project pipelines and fast AI demand changes, leading to high prediction uncertainty
• Supply chain delays, local approvals, and power access bottlenecks could result in lower capacity implementation than expected
• Conversely, continued capital investment could lead to more projects than anticipated, increasing power price risks

What to watch

• Whether regional grid reserve margins fall below the 15% reliability threshold
• The actual commissioning progress of data centers compared to pipelines
• Whether ERCOT and SPP can add enough new generation capacity to meet demand