For decades, the term infrastructure has called to mind toll roads, electricity grids, airports, power and transportation networks – the backbone of an industrial economy.

Today, however, the term also includes new kinds of digital infrastructure, such as data centers needed for artificial intelligence (AI) and assets linked to the energy transition. As this chart shows, more than 65% of infrastructure is now exposed to these high-growth themes because they require so much power. (We will address the recent DeepSeek news, and its implications for power demand, later in this article.)

Additionally, the cost of financing these assets is decreasing. The Federal Reserve cut interest rates by 100 basis points between September and December 2024, and we expect it to continue gradually normalizing rates going forward. This would reduce the discount rates applied to future cash flows, potentially increasing their present-day value.

What has held back the total return for infrastructure?

Consider an individual infrastructure project such as a wind farm. Builders typically finance these projects by borrowing money, which means taking on debt. For this reason, infrastructure companies have more than double the debt relative to their profits (Debt/EBITDA) of a typical U.S. company, as shown in the chart below.⁵

But once the wind farm is built, it generates electrical power that the owner sells to the operator of the grid. This income tends to be stable and can be sustained for a long time. As the chart below illustrates, infrastructure investments are about 24 years old on average, and their incomes are triple the dividend yield of the average S&P 500 company.

That reliance on debt becomes a bigger issue when the cost of financing rises. This was easy to see, as the Fed quickly raised interest rates from almost zero to over 5% in 2022–23. U.S. wind power additions fell 30% in 2022 compared to 2021.⁶

Today, the pricing of this asset class reflects the increased cost of borrowing. However, we believe it overlooks the future values of infrastructure investments based on their expected profits and incomes, and our expectations for further declines in interest rates. This has the potential to drive stronger performance.

We see this as a suitable time to consider allocating to the space because we believe demand for infrastructure will continue to grow at a rapid clip. We think durable and high-growth themes in AI and the energy transition will support that growth.

How does DeepSeek impact the AI infrastructure buildout?

In January 2025, DeepSeek, a two-year-old AI startup based in China, unveiled a new AI model that reportedly performed on par with leading U.S. models. What was most interesting is that the DeepSeek model is 10 times more power efficient than one such model, Meta's Llama 3.1. We believe DeepSeek will help ease one of the obstacles to broad AI adoption: constraints on power generation. These constraints are driving up energy prices and making AI more expensive to use.

More power-efficient AI will be less costly to use, and we expect this to help to spread its productivity gains throughout the global economy. While estimates for AI’s eventual power needs vary widely, even the lowest estimates would require substantial growth in power generation. In December 2024, the Department of Energy projected that U.S. data center electricity demand could increase by 16% to 26% by 2028. The efficiency of DeepSeek makes it less likely that demand will reach the higher end of that range, but a 16% increase is very significant.

Public companies recognize this investment opportunity, and they continue to surprise the market with their investment in AI infrastructure. In fact, after the DeepSeek announcement and the earnings report from Google parent Alphabet in February, the leading AI infrastructure providers are expected to grow capital expenditures by 29% year over year, up from prior forecast of 24%.⁷

What does it mean for investors?

We feel infrastructure is uniquely underpriced at present. As noted, we believe the market is underappreciating the strength and durability of the profits infrastructure assets can generate. We continue to believe AI is a powerful tailwind for these assets and believe the new administration is committed to accelerating this build-out.

To that end, President Donald Trump announced a joint venture between Open AI, SoftBank and Oracle called “Stargate” to fund artificial intelligence infrastructure. While further details are needed, the initial investment is said to total $100 billion, with an aim to increase to at least $500 billion over the next four years. This would equal the size of public and private investment from the CHIPS Act.⁸

In light of recent market turbulence, we believe investors can be underappreciating the value of infrastructure assets today. Infrastructure assets tend to offer stable cash flows, making infrastructure attractive as a long-term position that can provide potential diversification and consistent income.

Investors can gain exposure to infrastructure through both listed (public) and non-listed (private) investments. Historically, there has been little difference in the long-term median performance between the two. However, we believe non-listed infrastructure assets may offer slightly more attractive opportunities today due to discounted deal activity, as shown in the chart below. For new capital allocation, non-listed infrastructure provides the potential to deploy fresh capital at favorable valuations, enhancing the overall return potential.

We can help

Consult with your J.P. Morgan advisor to explore how you can incorporate infrastructure investments into your portfolio today.