Powering AI: Investment set to transform U.S. energy infrastructure

The electrical value chain can be broken into generation, transmission and distribution. Due to steep construction costs, electricity transmission and distribution are typically considered natural monopolies. This means that only an entity large enough to control an entire market will generally be able to afford necessary capital expenditures. As a result, most utilities are granted monopoly control over a local market with a mandate to provide low-cost, reliable energy as a public good.

Source: Council of Foreign Relations¹

Construction of the current electric transmission system began shortly after World War II to help link the more than 4,000 electric utilities that operated in isolation from one another. These connections enabled utilities to share the economic benefits of building large, often jointly owned power plants to serve their combined electricity demand at the lowest possible cost. The grid was built to accommodate coal and natural gas plants, which is why clusters of transmission lines can be found in Appalachia and Texas.

Energy-intensive AI collides with an aging grid

The American Society of Civil Engineers recently found that 70% of transmission and distribution lines are well into the second half of their expected 50-year lifespans.² To make matters worse, the physical buildout and maintenance of electrical grids has ground to a halt in the past decade due to project delays, rising costs and political concerns.

As a result of aging infrastructure and chronic underinvestment, the U.S. electrical grid is becoming increasingly unreliable. It is vulnerable to both cyberattacks and extreme weather events, with outages costing more than $150 billion annually.³

The U.S. has been extremely lucky over the past 20 years with essentially flat electricity demand. This primarily stemmed from energy efficiency gains including better insulation and the increased prevalence of LED light bulbs. Efficiency gains were notable in data centers, which serve as the backbone for internet infrastructure. They benefited from advancements in server efficiency and a shift in workload from traditional to hyperscale data centers.

However, electricity demand is expected to increase nearly 2.5% annually through 2030, with roughly 40% of this growth stemming from data centers alone due to the increasing use of artificial intelligence (AI) models.⁴ These models are more energy intensive than the simple data retrieval, streaming and communication applications that drove the technology sector’s growth over the past decade.

Exhibit 2: Electricity consumption per request

Source: Electric Power Research Institute (EPRI)⁵

Electricity demand heading “up and to the right”

As the pace of efficiency gains slows and the AI revolution gathers steam, utilities are expecting significant increases in electricity load growth. There are numerous variables to consider, including the development of business models, rates of increase in mature applications, and computational efficiency gains. As a result, the range of estimates from both utilities and investment firms is extremely wide.

Drawing on government information about existing data centers and electricity demand forecasts by industry experts, the Electric Power Research Institute (EPRI) prepared four scenarios of potential electricity consumption in U.S. data centers through 2030. Even in a moderate growth scenario where AI is a bust and not widely adopted, data center electricity demand was projected to grow 5% annually through 2030. If AI lives up to its hype, expanding rapidly, and there are limited energy efficiency gains, annual growth could be as high as 15%.

We believe that conservative growth scenarios should serve as a base case when considering potential risks and opportunities. Innovation can occur at both the generation and transmission levels with less slippage, as well as the product level with more efficient hardware. Regardless, in each forecast scenario electricity growth is “up and to the right,” which represents a material shift from the past two decades.

Exhibit 3: Composition of growth scenarios

Source: EPRI⁶

Meeting the demand

An important and frequently debated question is how this demand will be met. Predictions vary widely, but most acknowledge that an “all hands on deck” approach will be required. To date, utilities have mostly relied on natural gas to meet the electricity demands of data centers. This is not surprising given 24/7 baseload power requirements and the extensive pipeline infrastructure in the U.S.

We believe it is crucial to determine where future demand will come from. The four largest hyperscale platforms account for 80% of global data center capacity.⁷ Each of these firms has been instrumental in driving the shift towards contracting for renewable energy and all have committed to power their operations with 100% renewable energy by 2025.⁸ Recent “behind the meter” deals between hyperscalers and independent power producers (IPP) highlight this dynamic.

Integrated resource plans (IRPs) recently filed by utilities also provide insight into how increasing load growth will be met in the coming years. Plan details vary by utility and geography, but it is clear to us that renewables and natural gas are poised to account for the overwhelming majority of additional generation capacity.

Investment in the next chapter of U.S. grid infrastructure

There are two conflicting dynamics at play: (1) an aging U.S. electrical grid and (2) surging electricity demand. As a result, we believe there will be unprecedented capital expenditures from companies in the technology and utilities sectors.

• Infrastructure construction services: Upgrading transmission lines built to accommodate legacy coal and gas plants only add limited value, so there will likely need to be significant construction of new infrastructure. Construction and engineering contractors that specialize in building electrical grids and pipeline networks stand to benefit.
• Digital infrastructure: With data center supply already struggling to keep pace with surging demand in metropolitan areas, digital infrastructure operators could be huge beneficiaries. We believe firms with a dominant presence in Northern Virginia’s “Data Center Alley” and Silicon Valley may be best positioned to capitalize on secular trends.
• Electrical grid equipment: Upgrading aging infrastructure means businesses that provide electrical grid equipment such as circuit breakers, low-voltage components and overhead cables appear well-positioned. Firms that specialize in smart grid technologies may also play a key role as infrastructure becomes more distributed.
• IPP and utilities: Power generation traditionally has been a stagnant pocket of the market, but it has experienced a rejuvenation of growth from renewables capital investment. However, an uncertain regulatory environment and geographical considerations may lead to both winners and losers.
• Water infrastructure: Data centers require significant volumes of water for their liquid cooling systems, putting stress on underground aquifers, many of which are already being depleted at unsustainable rates. Businesses such as water utilities and equipment manufacturers stand to benefit from the commodity’s scarcity value. The latter includes companies that produce wastewater treatment systems, water pumps, fluid control products and related analytical systems.

Our Sustainable Investing team has tracked this theme over time, researching and writing extensively about our aging grid, the energy transition and the eventual electrification of (almost) everything. The major challenge of upgrading U.S. energy infrastructure presents investors with a diverse set of opportunities, and we believe investment managers that specialize in these markets will have the greatest likelihood of successful outcomes.