The End of Generally Flat Use: Berkeley Lab's 2024 Report on U.S. Data Center Energy

Congress asked for this report. The Energy Act of 2020, it says, calls for the U.S. Department of Energy to publish an update to the laboratory's 2016 study of data center energy use, and Lawrence Berkeley National Laboratory's 2024 United States Data Center Energy Usage Report, dated December 2024, is that update. It estimates historical use back to 2014, "relying on previous studies and historical shipment data," and it "provides a scenario range of future demand out to 2028." What follows comes from its executive summary, not the chapters behind it.

Flat, then not

For years the number barely moved. From 2014 to 2016, the summary says, annual use stayed "at about 60 TWh, continuing a minimal growth trend observed since about 2010." TWh is terawatt-hours. In the early and mid-2010s, the summary says, a shift from on-premise data centers to colocation or cloud facilities "helped enable efficiency improvements that allowed data center electricity use to remain nearly constant at a time when the data center industry grew significantly."

In 2017 that changed. The server base started growing, and GPU-accelerated servers for AI became a significant enough part of it that total use began to rise again, so that by 2018 data centers used about 76 TWh, or 1.9% of total annual U.S. electricity consumption. By 2023 their use had reached 176 TWh, or 4.4% of total U.S. electricity consumption. The summary gives the compound annual growth rate as approximately 7% from 2014 to 2018 and 18% between 2018 and 2023.

It puts the change in its own words. The expansion of data center services into areas that need new kinds of hardware "has ended the era of generally flat data center energy use," and "the rapid growth in accelerated servers has caused current total data center energy demand to more than double between 2017 and 2023."

A range after 2023

After 2023, total use "is presented as a range to reflect various scenarios," covering future equipment shipments, operational practices and cooling energy use. The equipment side rests on "the assumed number of GPUs shipped each year, which depends on the future GPU demand and the ability of manufacturers to meet those demands."

For 2028, the scenario variations "provide a range of total data center energy estimates, with the low and high end of roughly 325 and 580 TWh in 2028." At an assumed average capacity utilization rate of 50%, the summary says, that scenario range "would translate to a total power demand for data centers between 74 and 132 GW." GW is gigawatts. The same scenario range "represents 6.7% to 12.0% of total U.S. electricity consumption forecasted for 2028," and the scenarios put the compound annual growth rate between 2023 and 2028 at 13% to 27%.

The step from energy to gigawatts rests on one stated assumption, a utilization rate. A planner sizing supply for that demand would want to know how far the answer moves when that assumption does.

The summary also sets the growth beside a larger one. This surge, it says, "should be understood in the context of the much larger electricity demand that is expected to occur over the next few decades from a combination of electric vehicle adoption, onshoring of manufacturing, hydrogen utilization, and the electrification of industry and buildings."

What the method can see

The estimates come from what the report calls a bottom-up model, one "that calculates total electricity use from an installed base of data center equipment." That method, it says, "avoids overestimation that can be caused by tracking data center load for projects that" have not yet selected a power provider, but it "requires many inputs and assumptions developed from limited publicly available data, proprietary market analyst data, and review by industry representatives and stakeholders."

The summary states the limit in its own words. The areas of future research it identifies "would be furthered by greater transparency in data center energy use, as the lack of data availability significantly limits the analysis in this report." And "The lack of direct energy data available in a sector with rapidly evolving technologies limits the analysis in this report, especially when trying to understand and estimate future energy demand scenarios."

The report says its method avoids overestimation from counting projects that have not yet selected a power provider, and that the lack of data significantly limits its analysis. A lender leaning on the 2028 range is leaning on inputs and assumptions the report says were developed in part from limited publicly available data.