Loading...
Trading ideas and investment strategies discussed herein may give rise to significant risk and are not suitable for all investors. Investors should have experience in relevant markets and the financial resources to absorb any losses arising from applying these ideas or strategies.
BofA Securities does and seeks to do business with issuers covered in its research reports. As a result, investors should be aware that the firm may have a conflict of interest that could affect the objectivity of this report. Investors should consider this report as only a single factor in making their investment decision.
Key takeaways
- US power demand was nearly flat for more than a decade, with consumption rising 2.9% between 2010 and 2023 or 0.2% annually
- Now, data centers, EVs, energy intensive industry, and building electrification are set to boost demand by 70GW from 2023-30
- Rising renewable capacity, more coal retirements, and soaring demand could test regional grid stability in the medium term
After a decade of stagnation, US power demand is rising…
US electricity demand has been nearly flat for more than a decade, with total consumption rising just 2.9% between 2010 and 2023 or 0.2% YoY on average. Power demand stalled partly because of sluggish economic activity in the early 2010s and efficiency gains at the residential, commercial, and industrial levels, which were facilitated by LED lighting, more efficient appliances, and improved insulation among other factors. But power demand is also affected by weather, which was milder over the past year and masked underlying demand strength. In fact, on a weather adjusted basis, US electricity demand increased YoY in 2023, and growth should continue into 2030, driven by data centers, EVs, energy intensive industry, and other dynamics.
…and should increase by 70GW or 15% from 2023-30…
After rising 13 GW in 13 years, total US electricity demand averaged 457GW or 4000TWh in 2023, but we expect demand to accelerate, adding 70GW (~610TWh) between 2023-30. The commercial sector should drive demand growth (35GW or ~310TWh), led by a rapid expansion in data centers (25GW or ~220TWh), which have leapfrogged EVs to become the most immediate and meaningful source of growth. Even though EV adoption has faced headwinds recently, we still expect electrification to play a role in the medium term, with more than 10GW (~90TWh) of incremental charging demand across passenger and commercial vehicles. Other sources of demand growth are more obscure and difficult to assess, yet we see potential for more than 10GW of power demand growth across batteries, semiconductors, and other energy intensive sectors.
…while rising renewable capacity challenges grid stability
Grid stability concerns have been rising due to accelerating renewable capacity growth and more than 100GW of coal plant retirements in the past decade. New gas fired capacity concealed these changes until recently, but the pace of gas additions is slowing just as renewable capacity additions and power demand accelerate. Through 2030, another +50GW of coal could be retired, while gas additions remain below 10GW. Meanwhile, wind and solar should add 100GW and 240GW respectively, contributing to net capacity growth of about 290GW. However, after adjusting for the intermittency of renewables, potential incremental deliverable power looks more like 55-60GWa, but new behind the meter solar could deliver an another 15GWa. With power demand expected to rise about 70GW through 2030, grid stability could be tested regionally, and the call on thermal power may be insatiable at times.
Power sector loading pains
US power loads have finally started to rise after a decade of stagnation…
Electricity demand in the US has been nearly flat since the mid-2000s, as limited industrial growth and the rise of LED usage, improved appliance efficiencies, and other factors helped keep demand growth in check (Exhibit 3). After declining sharply during the pandemic, electricity demand recovered strongly in 2021-22 but struggled in 2023 as mild weather (Exhibit 4). Despite the short-term hiccup, several structural shifts in the US economy should propel demand higher into the end of the decade.
…and demand growth has occurred on a weather adjusted basis
Electricity demand, like thermal fuel consumption, is influenced by the weather. Warmer weather during summer months typically leads to higher AC usage and thus power demand. Similarly, during winter, electricity loads are rise as temperatures get colder and electric heating ramps up (Exhibit 5). This relationship should increase as more US building use electric heat and heat pumps, a shift that is being encouraged through federal, state, and local policies and subsidies. Although actual loads decreased YoY in 2023, weather adjusted loads have continued to increase YoY. Within the power stack, weather adjusted power sector gas burns have shown an even more pronounced increase YoY (Exhibit 6), which likely reflects coal retirements and the rapid deployment characteristics of combined cycle gas plants to supplement intermittent renewables generation (see Solar shines bright and More wind power won't be a breeze).
Several sectors should drive electricity demand growth into 2030
Policy changes and technological innovation have set the stage for strong US electricity consumption growth over the medium term. The IRA, the CHIPS and Science Act, rising fuel efficiency standards, and other dynamics are driving a massive increase in US industrial and manufacturing investment. Furthermore, the AI revolution is spurring organic investment in data centers, semiconductors, and other technology subsectors. Data center electricity demand growth dominates the landscape through 2030 and could add as much as 25GW or 220TWh of incremental power demand over the next seven years (Exhibit 7). Electrification of the vehicle fleet should also contribute more than 15GW or 130TWh through charging and from electricity used to produce new EV batteries. Other areas of the economy should also support power demand growth. In total, we think electricity consumption could rise as much as 70GW (~610TWh) through 2030, or a 2.1% CAGR (Exhibit 8).
AI catapulted data centers to the forefront of power demand growth…
Five years ago, data center power demand appeared to be a growing, yet predictable source of power demand. However, advances in Artificial Intelligence and the rapid increase in power usage to train and deploy those systems caught grid operators off guard. Indeed, load growth estimates have increased substantially across nearly every ISO. PJM's Dominion zone, which includes Loudoun County, the most concentrated data center hub in the world, has seen 10-year average load growth estimates rise from 0.5% in 2020 to 5.5% in 2023 (Exhibit 9). AI data center demand has sprouted up almost overnight and some are projecting a 4x increase in AI data center power demand at hyperscaler facilities by 2030 (Exhibit 10).
…and growth is expected to continue at a rapid pace into the 2030s
Data center power demand estimates put forth by consultants and industry participants point to global data center power demand nearing 60GW in 2023 or about 1-2% of global electricity demand. By 2028, one industry participant suggests data centers will require an additional 36GW or 62% growth (Exhibit 11). The subset serving AI is expected to rise even faster, growing nearly 300% over the same timeframe. Meanwhile, our equity research colleagues project very robust growth in AI server unit sales through 2027 (see Artificial Intelligence set be a key driver for server market growth) (Exhibit 12).
Yet, there is uncertainty around the evolution of AI power demand…
Different stages of the AI development process require substantially different amounts of energy, with training being the most energy intensive stage by a large margin (Exhibit 13). Once an AI model is trained and deployed, the energy cost of each inference, or using the model to make predictions or draw conclusions, is quite small. For many AI models, it takes millions of inferences to use the same amount of power as it takes to train the AI model. Some have argued that once all AI models are trained, energy use from the sector could fall dramatically. In theory, this may be true, but AI sector appears to still be in the early stages of development, with significant growth, and thus training, ahead. Furthermore, like software or any other type of technology, new versions of specific AI programs will likely roll out in the future, weakening the validity of that argument. AI is also expanding into different use cases, some of which are significantly more energy intensive than others. For example, image generation inferences use substantially more energy than image classification inferences (Exhibit 14). Myriad use cases and the rapid development of AI technology make it incredibly difficult to predict the pace of AI energy demand growth.
…and the location of new AI demand relative to surplus power supply
Power could eventually become a constraining factor in AI development. In fact, the concentration of data centers in Loudoun County has created challenges for the utility there (Exhibit 15), leading to delays as power generation is unable to keep pace with demand. The simple solution would be to spread out into other areas, but proximity and latency are important for many data center operators and contributed to the existing concentration in Loudoun County. Given the energy intensive nature of AI data centers, it would seem logical to establish new hubs in areas where low or negative electricity prices occur (Exhibit 16), but many of those pricing nodes don't have enough power to meet new data center needs. Instead, other new hubs are sprouting up in places like Columbus, Ohio, and some data centers are bringing power plants behind the meter to bypass grid connection issues.
