Al Root

Did you know that artificial intelligence works in "factories?" It does, but without an artificial lunchbox and an artificial hardhat.

Factories are what AI chip maker Nvidia and electrical infrastructure supplier Schneider Electric call the data centers where ChatGPT or Alphabet's Gemini learn to answer users' questions and then help teenagers summarize The Odyssey. The Tesla computers behind the company's planned robo-taxi service are another example.

Those factories will be using a lot more power for a long time

BofA Securities forecasts data-center spending will grow 12% a year on average for the coming few years, hitting more than $400 billion by 2028. That is money for servers, networking, and storage. Including money for electrical components, cooling, and engineering, the number jumps to more than half a trillion by 2028, for a gain of 16% a year on average.

"Inference is turning out to be surprisingly power-hungry," wrote BofA Securities analyst Andrew Obin in a Tuesday report. Inference is when a trained AI model, such as ChatGPT or Grok, answers questions or completes a task.

"Nvidia's drive to introduce new GPU evolutions every year [comes] with doubling power consumption every year," says Steve Carlini, Schneider's vice president for innovation and data centers. Nvidia's H200 chips draw about 70 kilowatts per server rack, while Blackwell chips use about 132 kilowatts, the executive said.

"The Rubins that are coming up next are 240 kilowatts per rack," Carlini added.

Schneider builds electrical infrastructure inside of data centers and is working on racks that provide 1,000 kilowatts of power. This past week, Schnieder and Nvidia announced an infrastructure partnership designed to help Europe catch up in AI.

The pair are supporting Europe's InvestAI initiative, which they say aims to build at least 13 AI "factories across Europe while establishing up to five AI gigafactories."

"The U.S. [electricity] grid is seeing gigawatt-scale growth," wrote Obin. That is a significant call because the prefix "giga" represents a billion. Total U.S. electricity generating capacity is about 1,200 gigawatts.

After growing at 0.5% between 2014 and 2024, U.S. electricity demand should grow at 2.5% a year on average between 2024 and 2035, according to BofA. That will require 800 gigawatts of new generating capacity, Obin said.

In terms of dollars, that investment would total about $1.4 trillion over 10 or 11 years. That is a positive for the likes of Schneider Electric, though Obin doesn't cover that stock. He likes GE Vernova shares, rating stock in the power-generation equipment company at Buy. He has a $550 price target for shares.

GE Vernova stock was up 0.4% in early trading at $489.67, while the S&P 500 and Dow Jones Industrial Average were down 0.2% and 0.1%, respectively. Schneider Electric stock was down 0.3%.

Over the past 12 months, GE Vernova shares were up about 190%, leaving them trading for about 43 times estimated 2026 earnings, according to FactSet. Schneider shares were flat, trading for about 22 times estimated 2026 earnings.

Write to Al Root at allen.root@dowjones.com

This content was created by Barron's, which is operated by Dow Jones & Co. Barron's is published independently from Dow Jones Newswires and The Wall Street Journal.

(END) Dow Jones Newswires

June 17, 2025 12:28 ET (16:28 GMT)

Copyright (c) 2025 Dow Jones & Company, Inc.