Image Source: Pexels
 

Tech companies are creating an energy Wild West grabbing land and turbines.

Desperate for Electricity

The Wall Street Journal reports AI Data Centers, Desperate for Electricity, Building Their Own Power Plants

In West Texas, natural-gas-fired power generation is under construction as part of the $500 billion Stargate project from OpenAI and Oracle (ORCL). Gas turbines are in use at Colossus 1 and 2, the massive data centers Elon Musk’s xAI is building in Memphis, Tenn. More than a dozen Equinix (EQIX) data centers across the country are using fuel cells for power.

With the push for AI dominance at warp speed, the “Bring Your Own Power” boom is a quick fix for the gridlock of trying to get on the grid. It’s driving an energy Wild West that is reshaping American power.

Most tech titans would be happy to trade their DIY sourcing for the ability to plug into the electric grid. But supply-chain snarls and permitting challenges are complicating everything, and the U.S. isn’t building transmission infrastructure or power plants fast enough to meet the sudden surge in demand for electricity.

Data centers have long taken power for granted, said KR Sridhar, founder and chief executive of Bloom Energy (BE) which provides fuel cells to companies that need on-site power, often in a hurry. “You build the data center. Well, you just plug it in.”

That isn’t possible anymore given the city-sized amounts of electricity needed to train AI models. One data center can devour as much electricity as 1,000 Walmart stores, and an AI search can use 10 times the amount of energy as a google search.

President Trump in January declared a national energy emergency, in part to keep the U.S. from falling behind China in the AI race. He has issued a series of related executive orders including one that aims to fast-track data-center construction and needed power infrastructure.

China will invest twice as much as the U.S. this year in power plants, storage and the grid, according to the International Energy Agency. It added about 429 gigawatts of new power generation last year, according to the think tank Climate Energy Finance, while the U.S. built about 50 gigawatts.

In some locations, data centers won’t be able to plug into the power grid until the 2030s because of the sheer backlog of projects and the fact that the nation’s high-voltage electric wires are running out of room.

Planning and building large-scale power plants or expanding grid infrastructure takes years. The process, normally gummed up, is even more difficult lately. Projects of all kinds face hurdles obtaining permits, equipment shortages, a labor crunch and rising costs, exacerbated by Trump’s tariffs on steel and aluminum, as well as some copper products.

Orders for transformers began climbing just as global supply chains became snarled at the start of the Covid-19 pandemic, according to data from energy consulting firm Wood Mackenzie. Data-center demand for the equipment is up 10-fold since then. It’s expected to quintuple next year. New factories and utilities’ efforts to replace aging or damaged equipment have added to the order backlog.

Massive turbines for large power plants have a yearslong backlog. But smaller turbines, reciprocating engines or fuel cells that also can use natural gas remain available—for now. Companies are snatching them up, adding them to data-center sites like Legos. Enough of them equal the output of utility-sized power plants or nuclear reactors.

One project opting to forgo a grid connection altogether is a Meta Platforms (META) data-center campus in Ohio. Regulators in July approved a plan by pipeline company Williams to build on-site natural-gas power. Williams has said it would spend about $1.6 billion on power and pipeline infrastructure for the Columbus-area site as part of the 10-year deal.

In Oklahoma, Gov. Kevin Stitt is clamoring for a piece of the action. His state advertises cheap electricity and, as needed, abundant natural-gas supplies for those with the money for a DIY power plant. 

“I don’t want to play Mother, May I? Can you generate power for me? Is it going to take seven years to connect to the grid? I need a gigawatt of power for AI,” Stitt mused in an interview.

Complicating the picture further, most of the recent investment nationally has focused on renewable energy. About 214 gigawatts of large-scale solar, wind and battery projects are under construction or in various stages of planning, about two-thirds of what is currently operating in the U.S. for those technologies, according to government data.

Analysts, however, expect spending on wind and solar to drop and project cancellations to rise because they are set to lose key federal tax benefits under Trump’s tax-and-spending law. The president and his team argue clean-energy projects don’t provide the round-the-clock power generation needed to meet AI demand and have promised to make permitting more difficult for wind and solar. Wind and solar developers say every available electron will be needed to help meet demand, and they can deliver quickly.

Already, at least $22 billion in new factories and electricity projects have been canceled or scaled back this year, according to data tracked by advocacy group E2, including everything from offshore wind to battery factories. The Energy Department is slashing another nearly $24 billion of funding for early-stage climate projects.

Build First, Figure out the Need Later

Here’s my favorite quote from the article.

“We don’t know exactly where we’re going to use it, but we know we have a multigigawatt planned development for the coming handful of years,” said Raouf Abdel, executive vice president of global operations at Equinix. “We want as much flexibility in our power supply as we can get.”

Caterpillar Connection

“Customers are saying, ‘Hey, can you help us bridge the two to three years until we can get a utility connection?’” said Jason Kaiser, group president of energy and transportation at Caterpillar. “That’s a new and growing opportunity for us.”

The engines have traditionally been used for backup or emergency power, while the smaller turbines have been used at power plants, for pumps and compressors in the oil field and as jet, marine or train engines.

Mark McDougal, co-founder of Joule Capital Partners, plans to use Caterpillar equipment with battery storage at a huge data-center project in Utah located on part of a commercial farm that his family has owned for decades. 

It’s designed so that it can connect to the grid later, McDougal said, “but we’re not relying on that.”

Does AI justify This Spending?

Color me skeptical.

On October 14, I addressed the question Is AI a Magic Bullet or a a Doomsday Machine?

But if the cost is totally borne by companies building the data centers, no one should care.

Unfortunately, that’s not the case.

As noted on September 7, Electricity Costs Are Soaring and AI Will Make Matters Worse

Electricity demand for AI data centers is soaring. The result won’t be pretty.

The data center companies are mostly acting like turbines and power plants are a temporary solution until the grid is ready.

But who is going to pay for the grid updates?

The answer is you and it’s happening already. Moreover, don’t think it will be over in two to three years, because it won’t.

More By This Author: