The Not So Obvious Implications of Artificial Intelligence on the Energy Market

January 2025
In this article we discuss the exponential growth in the use of artificial intelligence worldwide, the associated proliferation of data centers and the energy needs and challenges created by these developments both in the U.S. and other countries.
A look at recent upward trends in the stock market reveals a notable increase in the stock prices of firms engaged in artificial intelligence (AI) in one way or another, ranging from chip producers to service providers. Of particular note are the stocks that make up the magnificent 7, defined as Nvidia, Apple, Microsoft, Amazon, Alphabet, Tesla and Meta.
AI is arguably one of the most consequential technologies of our time and, according to Menlo Ventures, business spending on generative AI surged 500% to $13.8 billion in 2024, up rather staggeringly from just $2.3 billion in 2023. Powering this AI boom are data centers- defined as a large group of networked computer servers typically used by organizations for the remote storage, processing and or distribution of large amounts of data. Data centers are essentially the back office of AI, cloud computing, online gaming, and just about everything else that takes place online these days. To illustrate the massive impact that AI growth is having on data centers, consider the fact that a single Chat GPT query made by a consumer on a laptop requires ten times more power than an equivalent search engine query.
In large economies, such as China, the European Union, the United Kingdom and the United States, data centers are estimated to account for 2-4% of electricity demand. According to Cloudscene, there are more than 11,800 operational data centers worldwide and about 45% are in the U.S. as compared to 17% in Europe and 16% in China. Large data centers today can consume as much electricity as an electric arc furnace steel mill, and tend to be concentrated in certain areas. In the U.S., Virginia, especially the Northern part of the state near the District of Columbia, is said to have 20% of operating date centers and has become known as Data Center Alley. Other leading areas include Santa Clara, CA and Phoenix, AZ. The pervasiveness of these facilities in those areas has caused concern among local residents who have seen land they expected to be used for housing repurposed for commercial use. Much of the reason for locating these facilities near residential areas and for clustering them together can be attributed to the fact that it is both easier and more economical to construct and operate facilities in geographic locations that already have well-developed and reliable power generation capabilities.
Given the huge increase in AI development, some have estimated that data centers could ultimately use as much electricity as a small country such as Holland or Ireland. Currently, in each of at least five states across the U.S., data centers have already surpassed 10% of electricity consumption. Meanwhile, in Ireland, the share is already over 20% of the country’s total consumption. With major data center campuses currently under development around the world, these high consumption levels could lead to considerable strain on local grids. Meeting this electricity demand with emissions-intensive sources of electricity could also throw regional energy transition targets off track. Given this concern, companies with large data center operations have become increasingly active in procuring low-emissions sources of electricity from renewables and investing in technologies to produce nuclear and geothermal power.
The old saying that timing is everything could not be more true when it comes to AI. There is currently a race to the top among those in the space for the development and implementation of advanced AI accelerators. AI accelerators are chip-containing hardware, or in some cases software, that speeds up the time needed for data processing. Importantly, as firms such as Nvidia, AMD and Intel bring these higher performing chips to market, they consume more power than previous generations. So, as greater numbers of these newer chips are shipped and installed more and more power is also required. Examples of this dynamic can be illustrated by looking at Nvidia’s A100 chip with a maximum power consumption of 400W as compared with the company’s next generation H100 at 700W (75% higher). Somewhat concerningly, this significant increase in power consumption occurred within just 2 years and one generation of GPU advancement. Similarly, AMD’s MI 250 accelerator draws 500W of power while its next generation MI 300 consumes 750W (a 50% increase). Given developments such as these, it is not surprising that Wells Fargo recently predicted that AI power requirements will surge by 550% from 8 TWh in 2024 to 52TWh in 2026, and even further by 1,150% to 652 TWh in 2030.
A recent study by Bain & Company indicates that the surge in data center power consumption could require more than $2 trillion in new energy generation resources worldwide to meet the growth demand. Additionally, the study indicates that power demand could outstrip supply within the next few years, requiring utilities to boost annual generation by up to 26% by 2028. This far exceeds the largest generation boost of about 5% that was achieved by U.S. utilities in the 18 years from 2005 to 2023. Funding these large capital investments to increases power generation to fuel this data center growth over the next 10 years would require utilities to generate 10%-19% in additional revenue yearly, causing incremental increases in customer bills estimated at 1% annually.
According to a September 2024 report by McKinsey & Company, the U.S. is expected to be the fastest growing market for data centers, growing from 25GW of demand in 2024 to more than 80GW of demand in 2030. Furthermore, data center demand may make up between 30 and 40% of all new demand added until 2030.
Beyond power generation, another facet of utilities is the demand for generation and transmission equipment. This includes but is not limited to generators, uninterruptable power supplies, transformers, switchgear and power distribution units. Among the items most in demand and short of supply are transformers, a large portion of which have been produced offshored over the years. As a way of addressing this transformer shortage, Cleveland Cliffs, North America’s largest producer of flat rolled steel, announced recently that they would open a manufacturing plant in Weirton WV at a cost of $150 million to make three phase electrical distribution transformers. Similarly, Revere Copper recently opened a new plant in Mebane, NC to make copper bar for electrical switchgear, citing significant demand increases from AI oriented data centers.
Looking towards the future, utilities are expressing both enthusiasm and caution for what lies ahead. Dominion Energy, one of Virginia’ s principal utilities, is temporarily rationing power to some new data centers, awaiting construction of new transmission lines. Meanwhile in the Midwest, large regional utility American Electric Power (AEP) has proposed a new, higher rate for data centers to avoid running out of transmission capability. Arizona, meanwhile, largely supported by Arizona Public Service, has proposed building 800 miles of new transmission lines since they have determined existing transmission capacity will be maxed out by 2030.
Among the more interesting solutions put forth to supply the increased power needed by data centers is the proposed restart of the Three Mile Island nuclear plant near Harrisburg, PA. This two reactor nuclear plant, infamous for a widely publicized accident in 1979, had been operating its sole remaining reactor up until 2019, when it was mothballed due to high operating expenses and the availability of power from lower cost natural gas fired plants. While the restart has been proposed by and agreed to via the consummation of a 20-year power purchase agreement between Microsoft and plant owner Constellation energy, the expenditures required to rehabilitate and reopen the plant are expected to exceed $1.6 billion.
Increasing power generation itself, however, is a far more complex matter than just increasing production of the equipment necessary. The better part of the last decade has seen an emphasis on renewable and/or green energy from both a government and industry perspective. Faced with a new administration in Washington in 2025 that will likely lessen green and renewable initiatives, the power and energy industry has a challenging road to navigate ahead in order to effectively accommodate and facilitate the transformative nature of AI and its associated data center energy needs.