Power Shortages From A.I. Data Centers: The Gas Power Comeback Has Already Begun
Trending Topics Jakob Steinschaden
A TU Wien-led study says AI data centers can’t always demand full power. Otherwise, more gas plants may come back just to feed the servers.
Based on reporting by Trending Topics, Jakob Steinschaden — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
The new pressure point in the AI boom isn’t chips. It’s electricity. A research team led by TU Wien says data centers have been treated as if the grid must always serve them at full tilt, and that model doesn’t fit a decarbonizing power system. Their paper in Joule argues for a different rule: make data centers flexible customers, not fixed loads.
The warning is not abstract. TU Wien points to xAI’s use of gas turbines mounted on trucks in Southaven, Mississippi, to power a data center in nearby Memphis. A stationary power plant would have needed environmental permits. The mobile turbines, the study notes, did not have the air quality approvals that would have been required for the same job in another form. The case is still disputed, but it shows how far operators are willing to go when grid access is tight.
The numbers in the paper are blunt. In the United States, 39 percent of the gas power plant capacity planned at the end of 2025 is tied to data centers, according to the study. And this is not just an American issue. Even Norway, with electricity that comes almost entirely from renewable sources, does not have a strong enough grid everywhere to meet the needs of large data centers.
Reda El Makroum of TU Wien says companies are often not choosing the cheapest electricity, but the place with the best grid connection and the fastest path to one. That leads the team to a set of practical fixes: shift some training runs to times when renewable power is abundant, sign contracts that guarantee a base level of supply with temporary cuts during grid stress, and use better operating data so planners can see how demand actually changes.
The model calculation in the study suggests something striking: at the U.S. sites examined, power draw would have needed to be reduced for at most 35 hours a year to stay within existing grid capacity. That makes the usual “we need gas now” argument look a lot less inevitable. Jensen Huang’s line about using more fossil fuel may be the industry mood, but the paper pushes back hard on the idea that fossil first is the only route to clean later.
My take — AI-written commentary, not fact-checked reporting
The industry keeps pretending flexibility is some niche technical detail, when it is really the bill getting handed to everyone else. If data centers want the grid to sprint for them, they should also accept a few red lights. Otherwise, this is just old fossil logic in a shinier box.
Read more about this at: Trending Topics