Cloud Giants Turn to Natural-Gas Centers as New Forecast Warns of Double-to-Triple Prices

After years of massive wind and solar purchases, the hyper-scalers—Amazon, Google, Meta, and Microsoft—have begun betting on natural gas to power the data centers that underpin their AI ambitions, which they may later regret. A research report from energy analyst Noreva warns that gas prices in parts of the United States could triple in the coming years as hyper-scaler demand collides with a declining supply rate and a surge in liquefied natural gas (LNG) exports.

The scale of the shift is unprecedented. In March, Meta announced a 7.5-gigawatt gas-fired power plant in Louisiana for its Hyperion data center. Days later, Microsoft and Google each declared plans for similarly sized plants in Texas, and Amazon followed with a 7.6-gigawatt project in the same state. Companies that historically avoided heavy CapEx are now deep in a foreign, costly energy market.

According to Peter Gardett, chief executive of Noreva, fuel accounts for just over half of the electricity cost at a large power station. Current prices range from $2 to $4.5 per million BTU (about 7 to 17 shekel), with the widely used Henry Hub benchmark in Louisiana hovering just under $3. Noreva projects prices above $10 per million BTU at certain supply points, a jump that would make the “bring-your-own-electricity” model dramatically more expensive, push token-level costs higher, or force firms to connect to the grid, raising tariffs across the board.

Peter Gardett explains that price stability so far has stemmed from relatively flat demand and a steady supply that has offset declines in older wells. The supply growth rate is expected to slow, and new wells are costlier. A key variable is the integration of the local market with the global one: new pipelines from West Texas, where gas was once a cheap oil by-product, are now channeling volumes to export. When the region links to the world market, local price swings are reflected internationally and vice-versa.

Futures contracts have not yet priced in such a surge, and Peter Gardett admits “it’s not an unreasonable bet,” but he remains unsure the market is correct. One investor he spoke with expressed surprise at the price-risk exposure the hyper-scalers are willing to shoulder: “They do things that are not typical for an off-taker to do.” If the forecast materializes, the construction of self-powered stations could become a financial burden precisely when the AI race demands ever more capacity.