Hyperscalers spent years chasing wind and solar. Now Amazon, Google, Meta, and Microsoft are betting big on natural gas to power their AI data centers, and a new research report says that bet could burn them. According to TechCrunch AI, energy research firm Noreva projects that natural gas prices could triple in parts of the U.S. as surging hyperscaler demand collides with slowing supply growth and rising LNG exports.
That’s the trend worth watching: the biggest names in AI are wading into fossil fuel markets they barely understand, right as those markets get a lot less forgiving.
The bets already on the table
The spending spree is real, and it’s happening fast. TechCrunch AI reports the recent commitments:
- Meta: a 7.5-gigawatt gas plant in Louisiana for its Hyperion data center
- Microsoft and Google: gigawatt-scale gas plants, both in Texas
- Amazon: a 7.6-gigawatt gas plant in Texas
These are companies that historically avoided heavy capital spending. Now they’re pouring money into physical power infrastructure and, in the process, taking on price risk that would make a seasoned energy trader nervous. Noreva CEO Peter Gardett told TechCrunch that at least one investor was “surprised” by how much natural gas exposure hyperscalers are willing to carry. “They’re doing things that are not normal for an off-taker to do,” he said.
Why the math is turning
Gas has been cheap and stable for years. Flat demand plus steady new supply kept a lid on prices, with Louisiana’s Henry Hub sitting just under $3 per million BTUs today. Noreva thinks that calm is ending, and Gardett points to two forces.
First, the U.S. gas market is finally getting wired into the global one. In West Texas, gas was long a throwaway byproduct of oil drilling, sold at a discount because there was no way to move it. New pipelines changed that, and much of that gas now heads to export markets. Second, the AI demand pull. Stack both on top of aging wells and pricier new drilling, and Gardett’s conclusion is blunt: “You just need simple arithmetic to get to a much tighter gas market.”
Noreva expects prices above $10 per million BTUs at certain hubs for extended stretches. That matters because fuel is roughly half the cost of electricity from a large plant.
Why it matters now
Here’s the chain reaction. If gas prices double or triple, “bring your own power” data centers get far more expensive to run. That cost has to land somewhere. It could push token prices up. It could send hyperscalers back to the grid, which then drives up electricity prices for everyone else.
And there’s a public relations angle brewing. TechCrunch AI notes that 80% of consumers already worry about what data centers do to their utility bills. Right now that anxiety is about electricity. Spread it to natural gas bills, and the data center backlash gets a fresh target.
Gardett isn’t calling this a certainty. Futures markets aren’t pricing in big moves, and he admits the hyperscaler bet is “not an unreasonable” one. But his track record is in energy research, and his read is that the market has been “lulled into a sense that gas prices can’t go up.”
What to do about it
The practical takeaways depend on where you sit:
- If you buy AI compute: Ask whether your provider’s cost structure is exposed to fuel prices. Long-term token pricing may not stay flat if input costs spike.
- If you run infrastructure: Treat fuel price risk like any other. Hedging, diversified power sources, and grid fallbacks are worth pricing in now, not after the first shock.
- If you’re watching the industry: Energy is becoming an AI story. Power availability and fuel cost are turning into real competitive variables, not footnotes.
Gardett put the endgame in a line that sticks: “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.”
Over the next one to three years, expect power strategy to become a bigger differentiator in AI than model architecture. The companies that treat energy as a core competency, not a procurement afterthought, will be the ones that don’t get caught flat-footed. Full details are at the original TechCrunch AI report.