Goldman Sachs: Even if projects are delayed or canceled, U.S. data center power demand will still double within two years.

```

The AI wave is reshaping the US electricity landscape—data centers may become the core driver of US electricity demand growth in the next two years.

According to news from "Chasing Wind Trading Desk", on May 5, Goldman Sachs Commodities Research Team released their latest power market research report. Based on detailed project development progress data from data tracking platform Aterio, they made a systematic forecast of US data center electricity demand.

The bank believes that even if many projects are delayed or even cancelled, US data center electricity demand will still double within two years.

Data Center Expansion Accelerates, Scale Far Exceeds History

Let’s look at the basic data first.

According to Aterio, the US will add 6.4GW and 8.5GW of new data center capacity in 2024 and 2025, respectively. Based on current development plans, the annual additions for 2026 and 2027 will jump to 19GW and 69GW respectively—a leap in magnitude.

Goldman Sachs points out that the growth story is not just a national one. In PJM (Mid-Atlantic), ERCOT (Texas), and MISO (Midcontinent), each region’s planned addition for 2027 exceeds the total nationwide addition in 2025.

Delays and Cancellations Are Reality, But Can't Stop the Big Trend

Plans are plans; whether they can land is another matter.

Goldman Sachs compared project plans as of December 2024 with subsequent actual commissioning and found: Of the data centers originally planned to be commissioned in the next four quarters, only 72% were actually launched on schedule.

The main reasons for delays fall into three categories:

“Site Selection Game”: Developers often file applications in multiple regions to hedge risks, and eventually only proceed with the optimal location;
Supply Chain and Labor Issues: Waiting for electrical equipment during construction may cause months-long shutdowns;
The Construction Cycle Itself: It usually takes 1.5 to 2 years from project approval to commissioning. The farther out the schedule, the lower the probability of realization.

Based on historical patterns, Goldman estimates: Of the current plans, about 60% of new capacity within the next year will land on time; over the next two years, this rate drops to about 50%.

Even With Discounts, Growth in Demand Is Still Astonishing

Plugging the above discount factors into the model, Goldman’s forecast remains quite strong.

Specifically:

New additions in 2025 will total 8.5GW; by Q1 2026, 2.2GW will already be realized;
Q2–Q4 2026 will add another 11.5GW;
Annual additions in 2027 will reach 36GW;
By the end of 2027, total US data center installed capacity will reach 95GW, more than double the level by the end of 2025.

Corresponding to electricity demand (assuming 70% capacity utilization, consistent with the five-year historical average):

2025: 31GW
2026: 41GW
2027: 66GW

It’s noteworthy that Goldman’s internal teams have different forecasts. The equity research team, based on S&P 451 Research data, forecasts 39GW and 50GW for 2026/2027; the commodities research team, based on Aterio, forecasts 41GW and 66GW. Goldman admits in the report that differences in data sources and methodologies are the primary reason for the divergence. The overall market forecast range is also broad—from about 30–55GW (US Department of Energy/Lawrence Berkeley National Laboratory), to McKinsey’s 42GW, to Boston Consulting’s 67GW. Estimates vary widely.

By 2027, data centers will account for 8.5% of US summer peak electricity demand, up from just 4.1% in 2025.

Regional Divergence in the Power Market: Some Thirsting for Electricity, Some Relatively Relaxed

The explosive growth of data centers will not be evenly distributed, and the shock to regional power markets is very different.

The report categorizes the regions into three types:

1. Rising Reliability Risks (Tightest): PJM (Mid-Atlantic, covering Virginia, Ohio, etc.), MISO (Midcontinent), NW (Northwest)—these markets have limited new power generation, but data center demand is surging. Goldman believes these markets may have to reject some data center interconnection applications in the future and recommends hedging against local power price spikes.

2. Marginal Tightening (Relatively Loose): ERCOT (Texas), SPP (South Central), SE (Georgia)—These regions are seeing faster growth in power generation, so the additional demand from data centers will only cause mild tightening.

3. Saturated, New Additions Restricted: TVA (Tennessee), ISONE (New England), FL (Florida)—These markets are already in a critically tight state, and space for new data centers is extremely limited.

Goldman notes that this regional divergence highly aligns with data center siting logic: availability of electricity is the primary consideration, which explains why Texas and Georgia are attracting numerous new projects. Although PJM is tight on power, it still draws investment due to its geographic advantage of "proximity to customers" (especially Virginia and Ohio).

Forecasts Have Two-Way Risks, Goldman Recommends Dual Hedging

Analysts also explicitly listed the sources of forecast uncertainty:

Upside risks: Over time, new projects will keep entering development plans, so actual new additions could exceed forecasts; high capital expenditures might compress construction cycles to under a year;
Downside risks: Supply chain and labor issues may lengthen cycles; the data records are short (only traced back to December 2024), and market conditions change rapidly, making forecasts uncertain; in addition, many data centers do not disclose information, so data gaps exist.

Given the above two-way risks, Goldman’s strategy recommendation is: hedge both upward and downward power price risks—guard against price spikes in supply-tight markets like PJM, and guard against price pressure from oversupply in markets with rapid supply growth such as ERCOT.

~~~~~~~~~~~~~~~~~~~~~~~~

The above highlights come from Chasing Wind Trading Desk.

For more detailed analysis, including real-time interpretation and frontline research, please join [Chasing Wind Trading Desk·Annual Member]

Risk Warning and DisclaimerThe market carries risks, investment requires caution. This article does not constitute individual investment advice, nor does it take into account the particular investment objectives, financial situation, or needs of any individual user. Users should consider whether any opinions, viewpoints, or conclusions in this article are appropriate for their specific situation. Invest accordingly, at your own risk. ```