Artificial intelligence is a power-hungry business. The AI servers Nvidia shipped just last year are estimated to have consumed roughly the same amount of electricity as 20 million U.S. homes, according to Bank of America. Today, data centers use between 1% and 2% of the electricity produced worldwide, according to BofA analyst Andrew Obin. However, he expects their power consumption will grow at an 11% compound annual growth rate through 2030 — an estimate that could be a bit conservative given the energy-intensive nature of AI chips. “If AI hadn’t come along, data centers would not be booming the way they are today — and they wouldn’t be using as much power. But they absolutely represent another source of demand and stress on the grid while we’re electrifying,” JPMorgan analyst Stephen Tusa said. This is a big opportunity for companies such as Eaton , General Electric , Hubbell Power Systems and Vertiv that specialize in catering to the needs of data centers and servicing the power grids that supply them. The physical infrastructure needed for the data center market is estimated to have been worth $37 billion at the end of 2023, but upgrades should bring in $4 billion of new revenue through 2025, Obin said. That market includes power management, thermal management, IT and edge equipment as well as services and software. “AI and electrification trends suggests demand for electricity — and key electrical equipment — is set to accelerate through the end of this decade,” Obin wrote in a Jan. 16 note. In fact, “data centers are taking a disproportionate amount of incremental capacity being added by electrical equipment manufacturers.” Much of the energy data centers consume is used to train AI models as well as to run them. U.S. data centers consume 10 to 50 times the energy per floor space of a typical commercial office building, and these spaces comprise about 2% of total domestic electricity use, according to the U.S. Department of Energy. But there are limitations inhibiting data center capacity growth. Natural resources, power supply, funding — the necessities that allow for the existence of the data centers — aren’t exactly infinite. (In terms of natural resources, one way to think about the magnitude is that OpenAI’s ChatGPT drinks a liter of water — a quickly diminishing global resource — per 40 commands, which is the limit for users every 3 hours, according to a November report from BofA.) “It’s not just the utility challenge. It’s also a capital challenge,” TD Cowen analyst Michael Elias said. “We’re talking easily over a trillion dollars of spend here … and that money needs to come from somewhere.” The problem is especially severe in larger markets favored by the big tech hyperscalers, he said. Think companies like Amazon , Google , Oracle and Microsoft . But despite these limitations — ranging from available utility power to supply chain issues and a mountain of required capital expenditure funding — increasingly expensive and power-consuming data centers are still being built, giving data center product and infrastructure companies promising runway. “When it comes to advancing technology, society finds a way,” Tusa said. A ‘diversified electrical grid play’ The data center industry has consolidated and become more cyclical rather than growth-oriented, Tusa said. But while its R & D is relatively low, its fundamentals are strong. “While it’s relatively low tech, they have a pretty commanding percentage market share of the supply,” the analyst said. Power management giant Eaton is one of the primary beneficiaries of the data center boom, analysts say. It’s a “diversified electrical grid play,” Tusa said, noting that the company provides electrical infrastructure and industrial products for data centers, such as transformers and switch gear, but is also strong in other areas, notably in aerospace. Eaton generates roughly 9% of its revenue from data centers and an additional 5% from distributed IT, according to BofA, which forecasted the company will have about $2.6 billion in grid-related revenue in 2023, or 11% of its total. “Eaton has 31% of the U.S. market in power distribution and control,” Obin wrote in his recent note. “We argue a premium valuation is warranted due to broad exposure to key growth end markets, expected upside from cyclical operating leverage, strong margin performance, and Eaton’s less cyclical portfolio mix.” Obin is notably bullish on the stock, assigning it a $275 price target. Tusa holds a $230 target, which is below the company’s latest closing price of $243.90. The stock hit a 52-week high intraday on Monday. It’s an expensive stock, Tusa said. He expects earnings, due on Feb. 1, will be better-than-expected, which should lead to a higher price target. The consensus rating on Eaton is overweight, according to FactSet. Analysts’ average $251.12 price target implies shares could gain 2.7% over the next 12 months. The ‘data center play’ When it comes to data center infrastructure itself, the top-of-mind name for JPMorgan’s Tusa is Vertiv, a provider of cooling equipment and power monitoring products for data centers. “Vertiv is the data center play,” Tusa said. The company’s business took off last year as data centers began to need more electrical equipment capacity and generated more heat. Shares are already up 11.7% this year, and have soared more than 285% over the past 12 months. “The capex and the growth in these data centers and the amount of power they’re going to be consuming, these guys are right in on the guts of that,” Tusa said, referring to both overweight-weighted Vertiv and Eaton. “What’s important about the AI data centers is that these GPUs [graphics processing units] that have much higher computing power, they require a multiple of the megawattage of the power to work,” Tusa said. “Per square foot, it’s a 3x increase in the power consumption compared to the old CPU.” With data center-related revenue accounting for about 85% of Vertiv’s total sales, the company is a substantial chunk of the entire data center infrastructure market, BofA research shows. It gets about a third of its revenue from hyperscale data-center customers, while this market accounts for a much lower revenue percentage at larger companies like Eaton, Bloomberg reported . Vertiv’s management forecasts between 9% and 12% CAGR for its data center business between 2023 and 2028, according to BofA. Obin holds a $60 price target on Vertiv, indicating more than 12% upside for the stock. Energy conglomerate GE could generate about $5.2 billion in grid-related revenue in 2023, or 9% of its total sales, according to Obin, who has a $135 price target on the stock. The company’s portfolio of energy services and operations, GE Vernova, is set to launch publicly in the beginning of the second quarter . GE Vernova’s renewable energy business has faced consistent struggles due to inflation and supply chain pressures, however, and hasn’t turned a profit in the past two years. GE reported a disappointing profit forecast on Tuesday that was dragged lower by GE Vernova’s losses, which it said could post an operating loss between $200 million and $600 million for 2023. Analysts hold a consensus overweight rating on GE and a $143.86 price target, suggesting about 10% potential upside from Monday’s close. The stock has gained 2.8% so far this year. JPMorgan has a neutral rating on GE with a $124 price target, 5.5% below the stock’s Monday close. The ‘ultimate bottleneck’ One region investors might want to take a closer look at is Northern Virginia, which has the highest concentration of data centers in the U.S., according to BofA. With over 275 sites and tons more under construction or planned, the region is generating at least a third of global online use. Dominion Energy is the largest utility in the region, the analyst pointed out. The electric services company forecasts electricity load growth at a 6.6% CAGR between 2023 and 2028, he said, adding that just five data center customers represent 80% of that forecast. Analysts surveyed by FactSet have a consensus hold rating on Dominion, which has suffered from a slump in earnings. With more companies and organizations planning to build data centers, analysts expect issues of utility power capacity and pricing to intensify. “We’ve already built so many data centers that essentially the utility providers are saying ‘we need to pump the brakes or we need more time in order to deliver you the traffic,'” Elias said. “So the ultimate bottleneck right now, based on the conversations I’ve had with executives, is ‘how can we deliver utility?'” Hubbell is a company that might be in the right spot to solve these issues. About 60% of its business comes from electrical grid-related products, according to Tusa, who holds an overweight rating and $335 price target on the stock. Tusa said the shares are expensive, but slightly less-so than Eaton. “It’s the grid, which we find to be a very attractive opportunity over the next few years,” Tusa said. Slightly more bullish, TD Cowen on Thursday increased its price target to $362 From $347. That’s considerably higher than the average price target of $343 from FactSet, which suggests only about than 4% upside from Monday’s close. Shares are up more than 47% over the past year. While Hubbell missed earnings and revenue expectations for the third quarter, its president said during the Oct. 31 earnings call that over the next several years, Hubbell is “uniquely positioned” to solve critical infrastructure needs as grid modernization and electrification to continue to drive GDP and market growth. To satisfy power demands, either the utility provider ramps up their ability to deliver power or companies find new ways to power data centers separate from the electrical grid, Elias said, perhaps by repurposing already built industrial facilities or using small nuclear reactors to generate power with clean energy. “Those are all ways to deliver power to the data center. They have their challenges, but there are ways to deliver it,” he said. A ‘scarcity of human capital’ 2023 saw an influx of data center leases led by hyperscalers bracing for AI-led demand, Elias said, adding that of these customers are planning for data center capacity increases of almost 24 gigawatts by 2027. Looking at the total picture, data center customers include large hyperscalers such as Amazon and Microsoft, AI-driven and machine learning companies such as CoreWeave, and big companies that have tons of data such as Bloomberg, as well as sovereign entities using AI for government and state operations. Beyond the power capacity, another limitation is the soaring cost of data centers, which is pushing companies to find new sources of debt and equity to fund investments. Noting the strong demand and “limited ability to provide incremental supply,” Elias added that he expects data center pricing to increase over the near- and medium-term at a mid- to high-single-digit pace each year. Just one hyperscaler that’s planning to build 24 gigawatts of incremental capacity within the next five years would need an extra $240 billion in data center capital expenditure funding, he said. With supply limitations seeming insurmountable at times, analysts believe companies must look to innovative solutions to generate resources. “It is not just natural resources that face limitations. We are also facing scarcity of human capital and technology in the short term,” BofA global research strategist Martyn Briggs said. “A transforming world needs transformative solutions — not just grow, mine or produce more, but rethink the whole process in which we do things. Technology can help, from AI and more powerful computing speeding up development of new more efficient alternatives.”
