Meta, Amazon, Microsoft, and Google are betting more than 17 gigawatts of new natural gas capacity on the assumption that fuel stays cheap, and a new forecast from energy research firm Noreva argues that assumption is already breaking. Noreva expects natural gas prices at some U.S. hubs to top $10 per million BTUs in the coming years, roughly three times today's range of $2 to $4.50. The Henry Hub benchmark in Louisiana currently trades just under $3.
The hyperscaler build-out is concentrated in gas country. Meta committed in March 2026 to a 7.5-gigawatt gas plant in Louisiana to power its Hyperion data center. Days later, Microsoft and Google each announced their own gigawatt-scale gas plants in Texas. Amazon followed with plans for a 7.6-gigawatt facility, also in Texas.
Fuel represents about half the cost of electricity from a large power plant. Under Noreva's scenario, a doubling or tripling of gas prices would push "bring your own power" AI data centers well above the economics hyperscalers modeled at signing, either raising token costs or pushing operators back onto the grid and dragging electricity prices up with them.
Key facts
- 01Meta is building a 7.5-gigawatt natural gas plant in Louisiana to power its Hyperion data center; Amazon plans a 7.6-gigawatt plant in Texas.
- 02Microsoft and Google each announced gigawatt-scale gas plants in Texas within days of Meta's March 2026 announcement.
- 03Noreva forecasts prices at some hubs exceeding $10 per million BTUs, up from today's $2 to $4.50 range and Henry Hub's sub-$3 mark.
- 04Fuel represents roughly 50% of the cost of electricity from a large power plant, so a doubling of gas prices flows straight into token economics.
- 0580% of consumers are already worried about data centers' impact on their utility bills, per figures cited by Noreva.
Peter Gardett, CEO of Noreva, told TechCrunch that at least one investor he spoke with was surprised by how much price risk hyperscalers are absorbing directly rather than pushing onto suppliers. "They're doing things that are not normal for an off-taker to do," he said. Futures markets aren't pricing in a shock, and Gardett concedes the bet is defensible on today's data.
But the supply picture is tightening. Gas prices have held flat for years because new wells offset declines at old ones, but new wells are getting more expensive and production growth is slowing. Two structural shifts change the math from here: rising liquefied natural gas exports and the AI demand pull layered on top.
West Texas is the pivot. Most wells there target oil, with natural gas coming out as a byproduct that historically had nowhere to go, so producers sold it at deep discounts to anyone local who could burn it — exactly the arbitrage that made Texas attractive for AI data centers. New pipelines are now moving that gas toward export terminals, connecting a previously stranded regional market to global pricing.
The result is wider regional price differentials, not narrower ones. "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. Those spikes are what push individual hubs above $10 per million BTUs for extended stretches, even if the national average looks calmer.
There is a political dimension too. Roughly 80% of consumers are already worried about data centers' impact on their utility bills, a figure that so far has centered on electricity rates. If AI-linked gas demand starts moving residential heating bills in the same states, the backlash gets a second front — one where the hyperscalers have no operating history and little goodwill.
“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.”— Peter Gardett, CEO of Noreva
The hyperscalers do have levers. Long-dated hedges, on-site storage, dual-fuel plants, and eventually behind-the-meter nuclear or geothermal all sit in the roadmap. The near-term problem is timing: the gas plants are being built now, on decade-plus paybacks, against a fuel curve that the operators themselves do not control and cannot fully hedge.
The strategic risk isn't that gas becomes uneconomic — it's that AI infrastructure economics now depend on a commodity market with its own cycles, geopolitics, and export dynamics. Every incremental dollar per million BTUs flows into inference pricing, which flows into whether Gemini, Copilot, and the rest can hold margin as they scale. Alphabet's next capex-heavy quarter and a Henry Hub print are, for the first time, the same story.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
