AI’s Undeniable Downstream Impact

The artificial intelligence (AI) revolution has spawned dramatic innovation and spurred extraordinary spending as global industry attempts to realize its potential. But so far, much of the focus has centered on how AI might enhance business efficiency and improve the human experience.

Investors have steered toward select companies that primarily focus on the technology that generates AI’s expanding abilities. These include AI platforms and chatbots (think Open AI, 49% owned by Microsoft) and the computing and semiconductor chips powering them (think Nvidia). These companies have captured investors’ imaginations and helped drive the market higher, but we think the opportunity is broader than the most high-profile companies.

Behind the chatbots and chips triggering it all sit the large data centers that ostensibly serve as the refineries of the AI production process. These facilities require massive amounts of electricity. Cables and other connective equipment are also essential for this production network.

Data centers and their ongoing buildout face substantial challenges. Land limitations and AI’s increasing pressure on aging power grids threaten to shortchange data centers of the power they need. As a result, local opposition has led to moratoriums on data center construction in some areas. At the same time, prices for copper and other materials used for network connection have surged.

We’re finding that these challenges present potential investment opportunities beyond household-name stocks. We think the key to capitalizing on them is accurately identifying the difficulties and determining who can help resolve them.

Growing Demand for Data Centers Amid Supply Shortages

Technology usually gets cheaper over time. For example, on an inflation-adjusted basis TVs now cost a fraction of their price in the 1950s. Calculators, computers and cell phones all cost less than in the early years after their introduction. Their ubiquitous presence in today’s consumer landscape occurred alongside expanding affordability.

This isn’t yet the case with the large data centers necessary to bring AI to the masses because demand still overwhelms supply.

As of mid-2024, cloud service providers had announced plans to build 72 new data center campuses.¹ Many that don’t plan to build are choosing to lease instead, with data center leased capacity running two to three times the normal level. In the past two years since Open AI unveiled ChatGPT, rental rates for these facilities have surged 38%.² Simply put, spending on AI has exploded. Last year, large cloud service providers spent an estimated $154 billion on AI-related capital expenditures, and that figure will likely more than double by 2026.³

That increase reflects the scope with which providers want to expand their capacity. The 2024 Global Data Center Intentions Survey found that 97% planned to add data capacity this year, while none planned to decrease it.⁴

But even amid growing local resistance, the U.S. Department of Commerce projects demand for data centers will rise by 9% annually through the end of this decade.⁵ The reason? Again, there aren’t enough facilities to meet burgeoning needs.

Energy Consumption Challenges Facing Data Centers

Aside from land and regulatory concerns, some communities’ pushback against data centers has focused mainly on the facilities’ energy consumption. Data centers require an enormous amount of raw power that’s increasingly hard to find.

In a recent report, the consulting firm McKinsey outlined the dilemma, noting that by 2030, data center power consumption in the U.S. and Europe would likely triple, necessitating a substantial increase in electricity supply.⁶ Providing the estimated 50 gigawatts needed by 2030 in the U.S. would likely cost about $500 billion. Globally, power used specifically for AI could increase as much as fivefold by 2027.

With such significant growth, power grid operators expect electricity demand to rise substantially, potentially doubling over the next 15 years.⁷ As a result, prices for grid capacity have soared.

Overall, the U.S. grid system's operators can’t keep up with the large backlog of new facilities seeking to connect to it. Outside the U.S., the strain on power grids and generation capacity has led to moratoriums on new data centers in Amsterdam, Dublin and Singapore.

This creates investment opportunities in companies that supply power transmission and grid equipment and natural gas companies that supply the fuel for electrical plants.

At the same time, electrical utilities seeking to upgrade power generation capabilities have increased investment. Meanwhile, more so-called hyperscalers operating data centers have considered starting their own microgrids to generate power solely for their facilities. Earlier this year, for instance, Amazon Web Services paid Talen Energy $650 million for a data center adjacent to and powered by the Susquehanna nuclear power plant in Pennsylvania.⁸

The Role of Connectivity in AI Infrastructure

Enhancing the power grid creates investment opportunities in companies supplying power transmission and grid equipment. These opportunities extend to natural gas companies that supply the fuel for electrical plants, and the pipeline operators transporting that gas. Amazon’s acquisition of the Pennsylvania data center also signals support for alternative sources of power, potentially leading to the federal government bringing certain nuclear power plants back online.

Meanwhile, AI interconnection relies on relatively cheap fiber optic cables, and many have gone into the ground in the U.S. over the past two decades. AI-enriched network traffic between now and 2030 will expand at an estimated 120% compound annual growth rate.⁹ So, demand for fiber will likely remain strong, with fiber line installation virtually every time new road or highway construction occurs.

Within data centers, the copper coating of electrical systems enables cheaper liquid cooling of chips, using 25 times less energy than traditional designs.¹⁰ But copper primarily facilitates electrical wiring, components and data transmission to power the growing number of servers and computing systems. Not surprisingly, copper prices have risen 17% this year, and that trend could continue. Copper demand for North American data centers has been forecasted to rise by 20% to 25% by 2030 compared with a decade earlier.¹¹

In addition, recently introduced chips with more powerful capabilities contain as many as 5,000 copper connectors linking dozens of graphic processing units. Of course, semiconductor firms need to secure the products they use to build such chips. This should continue boosting demand for connectors, cable assemblies, sensors and circuit boards.

Addressing AI Infrastructure Needs

It’s clear that AI has created a new wave of technological transformation. Reaching its full potential will require power grid upgrades, more powerful computer chips and massive amounts of connective equipment.

Electrical utilities, power grid operators, hyperscalers and chip firms still need substantial investment. However, we believe these firms and the companies that supply them offer investors considerable opportunity.

Given the scope of AI’s expansion, we believe the related demand for the goods and services these companies provide appears promising. The downstream firms that can figure out the best way to deliver them will benefit the most. And they just may determine how close AI can come to fulfilling its bright future.