Amazon, Google, Meta, and Microsoft are betting on natural gas to power their AI data centers, despite a new report warning that prices could triple in the US.
Noreva, an energy research firm, says hyperscaler demand is clashing with slower supply growth and rising liquefied natural gas exports. The result could be severe price shocks that these tech giants are not prepared to handle.
“I think everyone in the energy markets has been lulled into a sense that gas prices can’t go up,”
Peter Gardett, CEO of Noreva, told TechCrunch. “You just need simple arithmetic to get to a much tighter gas market than you were in just a few years ago.”
Hyperscalers’ big bets
Low gas costs have pushed hyperscalers to lock up significant market share. In March, Meta announced plans for a 7.5-gigawatt natural gas power plant in Louisiana to fuel its Hyperion data center. Microsoft and Google followed shortly after, each committing to build gigawatt-scale gas plants in Texas. Amazon intends to construct a 7.6-gigawatt facility in the same state.
Companies that usually avoid large capital expenditures are now investing heavily in physical infrastructure. They are also moving deeper into energy markets, an area where they have little familiarity.
Gardett noted that at least one investor was surprised by the amount of natural gas price risk hyperscalers are willing to accept. “They’re doing things that are not normal for an off-taker to do,” he said.
Noreva expects prices to soar above $10 per million BTUs in certain hubs. Currently, prices range from $2 to $4.50 per million BTUs. The Henry Hub in Louisiana trades at just under $3.
Fuel makes up about half the cost of electricity from a large power plant. A doubling or tripling of gas prices would make “bring your own power” AI data centers far more expensive to run. This could drive up token costs or force hyperscalers to connect to the grid, which would push electricity prices higher.
For now, natural gas prices appear stable. Futures contracts do not anticipate major changes. “It’s not an unreasonable bet,” Gardett said. But he is not convinced they are right.
Surging demand
Natural gas prices have remained stable due to flat demand and steady new supply additions, which have countered production declines at older wells. Gardett expects energy companies to add more supplies, but not at previous rates. New wells are also becoming more expensive.
“That alone wouldn’t change the economics here. What’s changing the number is that finally we’re connecting the domestic gas market to the global gas market,” he said. “And the second is the AI demand pull.”
Hyperscalers have been drawn to Texas and Louisiana by cheap natural gas. In West Texas, most wells focus on oil, making the natural gas a byproduct with few markets. A lack of large pipelines meant producers sold gas at a discount to anyone who could use it. That is changing.
“They’ve finally built some pipelines out there, and a lot of that is headed towards export markets,” Gardett said.
As West Texas connects to national and international markets, demand there will influence prices elsewhere. Even modest price swings near hyperscalers’ major data centers could be magnified in other regions.
“You will get places where you get a lot of gas next to someplace where there’s none, and so you’ll get those big differentials,” Gardett said. It is these differentials that will drive prices in some regions above $10 per million BTUs for extended periods.
Under this scenario, even if hyperscalers can stomach higher prices, their natural gas consumption could add a new dimension to the data center backlash. Already, 80% of consumers worry about data centers’ impact on utility bills, mostly related to electricity. That concern could spill over to natural gas bills.
Hyperscalers are quickly enmeshing themselves in the fossil fuel world. It is a space where they have relatively little experience, but one that could soon materially impact their businesses.
“On future Alphabet earning calls, you will hear them talk about the correlation between natural gas pricing and Google results, which is strange, but that’s where we are,” Gardett said.
What it means
For people making things, the cost of generating the electricity needed for AI models could rise sharply. If hyperscalers pass these costs on, the price of using their services will increase. Alternatively, connecting directly to the grid could raise rates for everyone else.
