This is an extract from the report “Electricity Demand Growth and Data Centers: A Guide for the Perplexed” prepared by Bipartisan Policy Center.

Understanding US electricity sector load growth: A look at the data

Changes in the composition and geographic distribution of different industries and economic activities have always influenced U.S. electricity consumption, as have a host of other hard-to-predict factors, from technological developments and behavioral changes to short-term weather variations. Analysis of national U.S. electricity demand shows that there is no evidence for explosive demand growth in recent years, in the annual, monthly, or sectoral data, although some states have seen demand growth. 

Figure 3 shows Georgia Power’s forecasts of future demand, compared with actual net generation going back to 2006, using data from the utility’s Form 714 filings. The figure shows a large increase in Georgia Power’s 2023 forecast (indicated by a red line) relative to its pre-2022 forecasts (indicated by blue lines) and its 2022 forecast (indicated by a green line). The 2023 projection shows growth of about 75% in total electricity generation by 2033.

A similar pattern emerges from forecasts for the PJM Interconnection, which includes the state of Virginia, as shown in Figure 4. The 2023 forecast represents a substantial departure from past forecasts, showing a roughly 30% increase in demand by 2033 in the 2023 forecast. 

The 2023 forecast for the eastern Virginia portion of PJM anticipates one of the largest peak load increases for any utility in that region. Northern Virginia has the largest data center market in the world, with 70% of the world’s internet traffic originating in or passing through Loudoun County.

Load growth from data centers in context

Load growth due to data centers in a specific region can be difficult to predict. Data center developers consider multiple states as possible locations for data centers, and they query multiple utilities simultaneously for electricity rates and incentives prior to making a final selection. Therefore, counting data center project proposals to forecast load growth can result in the overestimation of data centers likely to be built in a specific service territory. Only national or regional level tracking of these projects can give an accurate picture, but such tracking currently does not exist, at least in a publicly available form.

Figure 6 shows the sources of new load growth in Dominion Energy Virginia. In this case, the majority of projected load growth is driven by the commercial sector, and specifically by anticipated growth in data center demand. Industrial and residential electricity sales, in contrast, are expected to be flat.

Another contributing factor to the uncertainty of projecting data center demand growth is the availability of land and new transmission capacity needed to support new data centers. Easy expansion of data centers in Virginia has depended on cheap land, low power prices, and public support—conditions that prevailed until relatively recently. There is evidence that data center buildout is entering a new and more contentious phase with siting issues and growing public opposition. Thus, expectations that growth will continue for the next 14 years as rapidly as it has in the past are uncertain. Currently, utilities are collecting better data, tightening criteria about how to “count” projects in the pipeline, and assigning probabilities to projects at different stages of development. These changes are welcome and should help reduce uncertainty in forecasts going forward.

Electrification is also important to load growth

Data centers are only one potential source of load growth in coming decades. Others include shifts in industrial investments and electrification of vehicles, heat, and industrial processes. Demand for air conditioning is also increasing as the world warms. Due to the rapidly changing nature of the U.S. commercial sector, policymakers must consider these factors when deciding how to address electricity load growth.

Figure 7 shows U.S. electricity use in 2020 and projected use in 2030, highlighting projected load growth by sector for 2020-2030. The projections for buildings, industry, and transportation are part of a high electrification scenario compiled in the Fifth National Climate Assessment for the United States. This scenario exercise was completed before recent concerns about electricity growth from data centers became salient—and before the onshoring effects from the Inflation Reduction Act became evident—so it is fair to infer that AI-driven scenarios of data centers’ load growth were not included in the projections. On top of that bar (in red) the highest projections is added, by the Electric Power Research Institute (EPRI), of data center load growth to 2030 for comparison. This comparison shows that data centers would still only account for 25% of expected growth over 2020-2030.

This comparison is admittedly a bounding one, given that the EPRI projections are at the high end of recent forecasts, but it does demonstrate that other sources of potential load growth are likely to be substantial compared with growth in data centers’ electricity use. Total data center electricity use would be about 8% of U.S. electricity use in 2030 based on EPRI’s projection, up from about 2% in 2020. The EPRI estimate for 2020 is comparable in percentage terms to estimates in but higher than estimates for the world, which are around 1%.

Understanding the potential for load growth from computing

The growth of two major factors will determine the impact of new computing applications on electricity demand: 

• Service demand, as determined by the quantity of computations being performed. 

• Computing efficiency, as determined by the amount of energy required to perform each computation.

As segments of the computing industry grow rapidly, an additional third factor may come into play in the form of limits on industrial production capacity to supply AI chips and servers, to expand the infrastructure needed to build new data centers (such as backup power generators), and to modernize the electricity grid. These three drivers are interdependent, and their trajectory in coming years is difficult to predict. Consider future demand for AI compute: The industry’s current growth projections are aggressive, but whether they materialize depends on businesses realizing positive economic returns from AI investments and on whether users’ concerns about accuracy and reliability can be adequately addressed. The industry’s growth trajectory could also be affected by new technologies that can deliver similar services but without some of the risks and liabilities associated with current AI models.

Historically, as demand for computing services has increased, the efficiency of meeting that demand has also increased rapidly, offsetting some or all of the growth in demand for computing services. In the early stage of the AI boom, efficiency was not top of mind, and companies bought all the AI equipment that was available, regardless of efficiency. As constraints in deploying AI manifested, the industry rightly began to focus on efficiency as one path to alleviating those constraints. This pattern matches what happened from 2000-2005 when electricity used by data centers in the United States and globally roughly doubled, and the industry then focused on improving efficiency. This effort led to slower growth in data centers’ electricity use in 2010 and little growth from 2010-2018.

Growth in the ability to meet service demand can also be uncertain because of supply chain constraints in producing and deploying AI servers and supporting equipment. In the first phase of the recent AI boom, people bought as many AI servers as the industry could produce, causing shortages that could persist if demand continues to grow rapidly. There are also constraints on the physical systems—such as backup power generators for data centers—that may hamper the speed of AI deployment. If service demand growth moderates, these issues become less pressing. When growth in new technologies is rapid, it can affect the rate at which these technologies can be deployed.

Conclusion

It is incumbent on utilities, regulators, policymakers, and investors to investigate claims of rapid new electricity demand growth and to ensure that expectations are based on the latest and most accurate data and models. Although data centers’ electricity use appears to be growing again, exactly how that growth will play out in coming years is deeply uncertain, both because growth in the use of AI is uncertain and because progress in efficiency is uncertain. It is likely, however, that other sources of demand growth, such as the onshoring of industry and the electrification of transportation, heating, and industry, will be bigger drivers of total demand growth than data centers in the medium to longer term.

Access the report here