Hyperscalers double down on natural gas for AI data centers

Meta, Microsoft, Google and Amazon are moving beyond buying renewable energy credits and into building large natural-gas power plants to run the data centers behind their AI ambitions. A new forecast from the energy research firm Noreva, cited by TechCrunch, warns that the same gas these companies are treating as cheap and abundant could become much more expensive in parts of the US as demand rises and supply growth slows.

According to TechCrunch, hyperscalers have spent years acquiring wind and solar developments, but the near-term constraint for AI is not annual “clean energy” accounting—it is hourly electricity that arrives when servers need it. Gas plants solve that dispatch problem, and Texas and Louisiana have offered a combination of cheap fuel, permissive buildout conditions and proximity to existing energy infrastructure. Meta has said it plans a natural-gas plant in Louisiana to power its Hyperion data center, and Microsoft and Google have announced gigawatt-scale plans in Texas. Amazon, TechCrunch reports, is also planning a large gas plant in Texas.

Noreva’s forecast is blunt: natural-gas prices could triple in some US hubs in the coming years as hyperscaler demand collides with declining supply growth and rising exports of liquefied natural gas. Peter Gardett, Noreva’s chief executive, told TechCrunch the market is tighter than it was a few years ago. The mechanism is familiar in commodity markets: when local buyers assume a regional discount is permanent, new pipelines and export capacity can erase the discount by linking the region to national and global prices.

Fuel costs, TechCrunch notes, can account for roughly half the cost of electricity from a large power plant. If companies adopt “bring your own power” models—building dedicated generation for their own campuses—a doubling or tripling of gas prices does not just dent margins; it changes the unit economics of AI itself, raising the cost of running models and potentially pushing providers back toward grid power. That shift would move the price shock outward: data centers reconnecting to the grid at scale can tighten supply for everyone else, lifting wholesale electricity prices in the same regions.

One of the more telling details in the TechCrunch reporting is how little of this risk shows up in the market signals hyperscalers usually rely on. Gas futures currently imply stable prices, with no broad anticipation of a step-change. The forecast instead points to structural changes—more expensive new wells, slower supply additions, and a domestic market increasingly tied to export demand—that do not fit neatly into a recent history of flat demand and steady supply growth.

For companies that market net-zero timelines while signing up for decades of combustion-based power, the bet is that gas stays cheap long enough to bridge to something else. Noreva’s warning is that the bridge itself may be priced like a scarce resource.

TechCrunch describes a market where new pipelines are already redirecting West Texas gas toward export routes. The same infrastructure that unlocks supply also removes the local bargain these data centers were built around.