The AI boom has enough room for manufacturers

Many legacy manufacturers worldwide have been trying to rebrand themselves as technology companies for more than two decades, but try as they might, even their own investors still tend to associate these companies more with hulking chunks of metal whose sales are dictated by the twists and turns of national economies. That might be about to change for a certain corner of the industrial world, not thanks to a fancy slide-deck promising skyrocketing software sales but a simple truth: Artificial intelligence (AI) can't exist without electrical equipment and air conditioners.

Pioneered by the US, AI works best with graphics processing units, or GPUs, which are capable of processing huge streams of data in parallel. GPU-powerhouse Nvidia Corp has predicted that $1 trillion of data-centre infrastructure will need to be upgraded to handle accelerated computing that supports generative AI tools such as ChatGPT. A recent Bloomberg Businessweek profile of Nvidia compared these chips to buying groceries by using dozens of people armed with handbaskets, each on a mission to find a specific item, like fruit or cereal, whereas the central processing units, or CPUs, that have historically dominated data-centre infrastructure are more akin to using a single shopping cart that roams the aisles.

That image is a helpful one in understanding how the expected AI-charged growth in GPUs might affect the need for power-management and cooling equipment in data centres. GPUs are more efficient at sorting through vast amounts of data, but they require about twice the power of traditional CPUs, Bank of America Corp analyst Andrew Obin wrote in a 1 June report. A 10% increase in power use would broadly require a more than 10% increase in electrical-equipment capacity because of the need for redundant infrastructure to help guard against power failures, Obin said. Higher power use also means more heat is generated, so data centres will need to invest in technologies that can keep both the equipment itself and the ambient temperatures appropriately cool. One incremental megawatt of power would require about 285 tons of cooling, similar to the requirements for a 115,000-square-foot commercial building, Obin said.

"We love working with data centre customers," Trane Technologies Plc Chief Executive Officer Dave Regnery said earlier this month at a UBS Group AG conference. "They're very sophisticated buyers, and they really push technology with us. And they like working with us, because we have a lot of expertise. … We come up with solutions that we didn't even think were possible."

This potential opportunity from AI helps explain why shares of Vertiv Holdings Co, which specialises in power-management and cooling equipment for data centres, have climbed almost as much as Nvidia's on a percentage basis in the wake of the latter company's blockbuster revenue forecast in late May. Vertiv struggled with operational missteps and supply chain challenges through much of 2021 and 2022, so its stock had some significant ground to make up. Shares of electrical equipment manufacturers Eaton Corp and nVent Electric Plc are both up about 15% over the same period, while heating, air-conditioning and ventilation company Trane has outperformed the broader S&P 500 benchmark. Shares of Johnson Controls International Plc, which acquired hyperscale data-centre cooling expert Silent-Aire for an upfront payment of $630 million in 2021, have also rallied. 

The next tech bubble or a real shift?

Shares of anything related to AI have rallied after Nvidia's blockbuster sales forecast

Vertiv is the former network-power business of Emerson Electric Co, which the company built in part through the $1.5 billion acquisition of Chloride Group Plc in 2010. Emerson later took a $508 million writedown on that deal, and former CEO Dave Farr memorably said in 2015 that getting into the network-power business has "not been one of my better approaches in life, and I wouldn't say I would put that on my tombstone to say that was a strong accomplishment." Emerson struck a deal in 2016 to sell the business to Platinum Equity for $4 billion; the Vertiv business became publicly traded in 2020 via a merger with a special purpose acquisition company backed by Goldman Sachs Group Inc and led by Dave Cote, who was previously CEO of Honeywell International Inc. It now has a market value of $8.5 billion. 

That example and myriad others suggest some scepticism is warranted on the prospects for a great AI-enabled data-centre spending extravaganza on electrical equipment and temperature controls. Industrial companies have tried to wedge themselves into every technology narrative that's captured market attention, with varying degrees of credibility. Barclays Plc analyst Julian Mitchell recalled that many manufacturers claimed to have "internet exposure" in the early 2000s because they had a website. And who can forget General Electric Co's claims that it could generate software sales that would rival those of Oracle Corp and SAP SE. Its digital business still exists but isn't exactly a powerhouse. In more recent years, many manufacturing giants have used pricey acquisitions to enter the realm of industrial software. Investors typically aren't thrilled by these transactions and are still waiting for most of them to pay off. 

One of the main selling points of accelerated computing is that it's more efficient and cost effective and therefore does a better job of maximising available space in data centres. "While we think there is an argument for a rising tide to 'lift all boats' as long-term data centre demand grows, it is debatable if having fewer units as a result of greater efficiencies would result in a greater amount of content in the data centre for" electrical-equipment suppliers such as Eaton and nVent, Mitchell wrote in a 9 June report. He doesn't cover Vertiv. "Historically, electrical intensity in data centres has not seen linear growth," and this is a reason investors encouraged Emerson to sell the network-power business, he wrote. Facebook parent Meta Platforms Inc trimmed its capital expenditure outlook for 2023 to reflect plans to pivot to a new more cost-efficient data-centre architecture that can support both AI and non-AI workloads. "We're expecting that the new design will be cheaper and faster to build," Meta's Chief Financial Officer Susan Li said on the company's February earnings call. 

Data-heavy future

Meta curbed its 2023 capital spending plans, but it's still going to shell out more than $30 billion this year. The company has said the "main driver" of its increased expenditures has been investing in infrastructure to support AI.

The best way to think about the AI opportunity from an industrial-equipment perspective may be to view it as incremental support for the already robust demand trends in the data-centre market. Growth has downshifted recently, but no one thinks the world already has enough data centres to power where it wants to go. As the semiconductor shortage made painfully clear, every piece of equipment — from tractors to factory automation tools to electronic cigarettes — is jam-packed with data-harvesting sensors. Eaton and Vertiv are already struggling to meet demand for their data-centre products, so additional AI-related orders would further pad backlogs out to 2024 and 2025, Obin of Bank of America said. Vertiv generates about a third of its revenue from hyperscale data-centre customers, while this market likely accounts for more like a low-single digit percentage of revenue at larger electrical equipment companies Eaton and Schneider Electric SE and HVAC companies Trane and Johnson Controls, according to estimates from Melius Research analyst Scott Davis.

It's too early to say whether AI will be enough of a swing factor to drive outsize growth at these companies relative to the broader economy, but it should at the very least reinforce, rather than intermediate, demand for power-monitoring and temperature-control technologies. Amid widespread concerns that AI will steal jobs and render entire industries extraneous, the guaranteed continued relevancy of electrical and cooling equipment is nothing to sneeze at.

Disclaimer: This article first appeared on Bloomberg, and is published by special syndication arrangement.