A boom in artificial intelligence (AI) investments now drives the United States electricity market. New data centers and upgrades to existing data centers are creating a vast demand for power. But big AI firms are choosing natural gas plants to provide electrical power, not renewable energy.
Since ChatGPT released their conversational AI chatbot in November of 2022, the world of artificial intelligence has exploded. ChatGPT showed that machines can learn the complexities of human language and interaction. Within five days, the chatbot had attracted over one million users. ChatGPT ushered in the modern AI revolution, producing trillions of dollars of investment in the US and around the world.
Artificial intelligence uses vast amounts of computing power and databases with trillions of parameters to solve problems and make predictions. Generational AI can create text, images, music, and videos. These capabilities require enormous data centers filled with thousands of servers using high-speed graphics processing units (GPUs).
Amazon, Google, Meta, Microsoft, Tesla, and other firms are investing billions to build new data centers and to upgrade existing facilities. According to USA Datacenters, America had less than 2,700 data centers at the start of 2024. Now about 3,900 data centers are operating, an increase of more than 40 percent in 18 months.
Data centers traditionally support the internet and cloud storage. At the start of 2024, data centers consumed about 4% of US electricity. But when servers are upgraded with thousands of high-speed GPUs to deliver generational AI, they use six to ten times as much power. It is estimated that by 2030 US data centers will consume 20% of the nation’s power. And the overwhelming majority of electricity for these AI facilities will come from new natural gas power plants.
Green energy advocates want renewables to power the AI data centers, but wind and solar systems fall far short. AI computers operate 24-hours a day, which can’t be supported by the 30% capacity factor of wind or the 15% capacity factor of solar. Gas plants are designed to deliver their rated output over 90% of the time.
Nuclear plants may be able to power AI in some cases. Retired nuclear facilities in Palisades, Michigan and Three Mile Island, Pennsylvania are being restarted to drive AI. But a start-up nuclear plant costs up to 10 times as much as a gas-fired power plant. It can take more than five years to bring a nuclear plant on-line, compared to 1-2 years for a gas plant. Small modular nuclear reactors promise cost and cycle time breakthroughs, but these are still unproven technology.
Wind and solar also suffer from transmission costs and delays. These systems are scattered over wide areas, requiring new transmission towers and lines to be built across the countryside to aggregate enough power for a large data center. In contrast, gas plants provide concentrated generating power that can be built near the data center and existing transmission lines.
Elon Musk’s Tesla is a leading provider of solar systems and grid-scale batteries. But Tesla chose 35 on-site gas turbines to supply electricity for its Colossus xAI supercomputer facility in Memphis, Tennessee. Colossus was built and expanded to 200,000 GPUs in less than six months in 2024 and claims to be “the most powerful AI training system yet.”
The Colossus build is an example of BYOP, or “bring your own power.” Rather than using electricity from the grid, leading technology firms are building their own natural gas plants on-site to power data centers.
AI data centers are huge facilities. EdgeConneX plans to build two data centers in New Albany, Ohio, to be completed by the fall of 2027, using existing warehouse space and new construction. The facilities will total 1.2 million square feet, covering an area larger than 20 football fields. These data centers will be powered by on-site gas plants.
More than 100 gas-fired power plants are planned for deployment in Texas, including 108 new plants and 17 expansions. The projects will provide 58 GW of electrical power. Sixty percent of the projects aim for completion by the end of 2028. These new facilities, if completed, will deliver more than three times as much electricity as all of the wind systems currently operating in Texas, the leading US wind state.
Earlier in July, President Trump joined state senator Dave McCormick at the Energy Innovation Summit in Pennsylvania. At the conference, they announced private investments of more than $90 billion in AI data centers and new or expanded power generating systems. Amazon, Blackstone, CoreWeave, Pennsylvania Data Center Partners, PowerHouse Data Centers, and other firms will build AI infrastructure in Carlisle, Lancaster, Pittsburgh, and other Pennsylvania locations.
Pennsylvania is one of the largest producers of natural gas. Recently, the price of gas in Pennsylvania has been too low for local firms to make a profit. But most new AI installations will be powered by gas plants, to be constructed by PPL Corporation, Frontier Group, and others, providing new high-demand customers for gas producers. Westinghouse Electric also plans to build ten new nuclear reactors in the state, to be on-line by 2030.
Meta claims to be building the largest data center in the Western Hemisphere. The facility will be located in Richland Parish in northeast Louisiana, 250 miles north of New Orleans, and will cover 3.5 square miles or 2,250 acres. When completed in 2030, the installation will use twice as much electricity as the city of New Orleans. Entergy Louisiana is building three large gas-fired power plants at the Meta site at a cost of more than $3 billion.
The artificial intelligence revolution now drives the US electricity market. Most new AI data centers will be powered by newly constructed natural gas facilities. The Net Zero electricity transition is being left in the rear-view mirror.
Originally published in the Tennessee TriStar Daily.
