Energy Transfer’s February announcement of a deal to supply natural gas to an AI data center marked a transition point.

Analysts and operators have speculated since 2023 that an exploding AI sector would result in a follow-on boom in natural gas demand. Executives from just about every major U.S. midstream and gas supply company discussed ongoing negotiations for energy-hungry data centers for much of 2024.

Until ET’s announcement, however, no concrete deals between the natural gas and AI sectors had been made public.

“Our team met with CloudBurst [Data Centers] last summer to understand what data center projects are looking for. As you’ll hear me say over and over, many of these proposed sites are very close to our pipeline systems,” said Mackie McCrea, Energy Transfer’s co-CEO, in an exclusive interview with Oil and Gas Investor.

Macie McCrea Energy Transfer
Mackie McCrea, Energy Transfer’s co-CEO

CloudBurst entered into a long-term agreement with Energy Transfer for a greenfield campus project in San Marcos, Texas, located between Austin and San Antonio. Through its Oasis Pipeline network, the company will provide up to 450,000 MMBtu/d after CloudBurst reaches a final investment decision with its unnamed customer on the project.

Oasis is a 36-inch diameter, 1.2 Bcf/d capacity transport network that connects the Waha Hub to market hubs in Katy, Texas. The natural gas supply will go behind-the-meter to a 1.2-gigawatt power plant built on-site once complete. CloudBurst expects the center to go online by third-quarter 2026.

The talks between Dallas-based Energy Transfer and Denver-based CloudBurst started about a year ago, McCrea said, with CloudBurst reaching out first.

AI, data centers and the developing industry’s needs are things the energy industry in general has become familiar with over the last year. Just like the energy industry, there are a lot of people looking to get into the AI business. McCrea said Energy Transfer had to spend some time trying to separate the serious from the non-serious clients.

“We are not experts in data centers,” McCrea said. “But we do have a team that is dedicated to this now, and we’ll probably be adding to that team. They’re on an incredible learning curve.”

A team leads CloudBurst with years of industry experience. The company’s CEO Tye Johnson also leads Evolve Holdings, which has built data centers in the U.S. and Europe. CloudBurst was formed to build the next generation of data centers for AI, and the company lists the San Marcos project as its flagship.

Energy Transfer and CloudBurst eventually became comfortable enough with each other to start working toward a deal.

Keeping the lights on, absolutely

AI data centers can’t just go up anywhere. The projects need land, not just for the initial project but for follow-on expansion.

They also need access to low-latency, fiber-optic, ultra-speedy connections. The buildings need to be close to a metro area, and the site needs to have access to plenty of water for cooling.

Most importantly, AI data centers must have access to power—as in, a lot of power—and an extremely reliable supply.

Anyone with a home computer knows that a sudden power cut can be a hassle, and it’s exponentially more so at a data center with tens of thousands of chips.

A lot of media has been dedicated to the amount of power a data center needs, Bill Schweber, an electronics engineer, wrote for EE Times, an industry publication for electrical engineers.

Much less focus has been placed on the reliability of the power supply, but it’s one of the major reasons data center companies have turned to natural gas instead of solar and wind.

“The less-known aspect is that for data centers in general and AI in particular, a power outage—even for a fraction of a second—is a very bad thing, period,” Schweber said.

“If local power goes out, sags, brown-outs, or glitches, the data center must go on as if nothing had happened. These centers are looking for 99.99999% uptime (seven 9s), which is much higher than the customary five or even six 9s of hi-uptime reliability systems.”

In general, most data-center operators would prefer to get their basic power for their new installations exclusively, or primarily, from solar and wind, Schweber said. “The huge power demands make that a very difficult option.”

In fact, technology companies continue to adjust to the reality of their power requirements on the U.S. grid.

On CloudBurst’s website, the company includes a picture captioned “An AI created image of typical warehouse Data Center.” The picture shows a building surrounded by fields of solar panels and wind turbines.

Data Centers Proximity to Energy Transfer Pipeline Map
(Source: Rextag)

Back-ups for the back-ups

As planned, the actual facility will be connected to its own natural gas power plant powered directly from gas shipped in by Energy Transfer for its primary power. After that, he redundant systems will be built in.

The San Marcos facility will have a connection to the ERCOT grid. On-site diesel generators will be connected in case the other sources simultaneously fail, and a large battery system will give operators a few hours of time if needed.

“They’re doing everything they can to be as redundant as possible,” McCrea said.

Another selling point for the data center is Energy Transfer’s ability to provide natural gas storage, so the flow of fuel isn’t completely dependent on a working pipeline.

“That’s why we feel very good about our position. Our big diameter pipes throughout the country are close to many of the proposed data centers, and our ability to support them with reliable storage separates us from the competition in many of these areas.”

McCrea said the agreement with CloudBurst is a lot like other on-site provision deals Energy Transfer has done.

“Every project we do is a little bit unique, but this is really not much different than our core structure on many of our contracts, whether we’re doing LNG, industrial businesses, gas utilities or power plants,” he said.

Lessons learned

Energy Transfer expects data center agreements to become a regular part of its business profile over the next couple of years.

The market currently favors large gas and midstream companies, which are most likely to have well-developed transport networks that connect the major gas-producing fields to the metro areas favored by AI operators.

Data center location development may seem random on a map, but the locations chosen follow a similar trend. Companies have selected spots in Northern Virginia; Abilene, Texas; and San Marcos—all close to a major pipeline corridor and source of natural gas.

“If you look at the data centers and how they’re placed throughout the country, it’s as if we picked those locations,” McCrae said. “They’re so close to our pipelines.”

McCrea said Energy Transfer has been in talks about sites in Texas, Louisiana, Oklahoma and some other outlying areas.

“We don’t have quite as much storage (in some other areas), but we still have the ability to perform, and we’re talking to a lot of those folks,” he said. “Even in the midcontinent, we’re close to signing another deal.”

The deals for the industry could come fast and furious. AI is a leading-edge sector of the tech industry, and the largest companies are in a race to build to keep up and get ahead, both with their competitors and an expanding market.

“What we’ve learned, even over the last several months … is that a lot of these developers are in a hurry,” McCrea said.