Hyperscalers' gas pivot trades a hedged power bet for an unhedged fuel bill

After years of buying wind and solar developments, Amazon, Google, Meta and Microsoft are now betting that natural gas will power the data centers behind their AI plans ^[1]. A new forecast from the energy research firm Noreva argues that gas prices could triple in parts of the US as hyperscaler demand collides with slowing supply growth and rising LNG exports, which would convert a cheap-power decision into a recurring operating-cost exposure ^[2].

The build commitments are already large. In March, Meta said it would build a 7.5-gigawatt gas plant in Louisiana for its Hyperion data center; days later Microsoft and Google each announced gigawatt-scale gas plants in Texas, and Amazon said it would build a 7.6-gigawatt plant, also in Texas ^[4][5][6]. Taken at face value, that is at least 17.1 gigawatts of announced self-supply ^[7], from companies that historically avoided heavy capital expenditure and are now pushing into energy markets they know less well ^[18].

The pricing gap is the whole story. Hub prices today run from about $2 to $4.50 per million BTU, with Henry Hub just under $3; Noreva expects sustained prints above $10 at certain hubs ^[8]. That is more than a tripling from Henry Hub, an increase of roughly $7 per million BTU ^[9]. Because fuel is about half the delivered cost of electricity from a large plant, a tripling of gas would roughly double the cost of power from these units ^[10][11]. The pass-through options are unattractive: higher token costs, or a retreat to the grid that pushes retail electricity prices up ^[10].

Noreva's mechanism is not a demand story alone. Peter Gardett, Noreva's CEO, told TechCrunch that prices have been stable because demand was flat and new supply reliably offset decline at old wells, and that supply can still grow, just more slowly and at higher cost per well ^[13]. What changes the arithmetic, he said, is that the domestic market is finally being wired into the global one, plus the AI demand pull ^[14]. West Texas is the clearest case: gas there has largely been an oil byproduct sold at a discount for want of pipelines, and the pipelines now being built point at export markets ^[15]. Once regional supply is exportable, local scarcity next to local glut produces wide basis differentials, and it is those differentials that Gardett expects to hold some regions above $10 for extended stretches ^[16].

The counterargument sits in the forward curve. Futures are not pricing large moves, and Gardett concedes the hyperscalers' position is "not an unreasonable bet," though he is not convinced they are right ^[12]. One investor he spoke with was "surprised" by how much gas price risk the hyperscalers are absorbing, doing "things that are not normal for an off-taker to do" ^[19]. That is the operator-relevant point: the exposure is structural, not hedged away by the plants themselves.

Watch three things. Whether any of these four disclose long-dated fuel hedges or tolling structures alongside the plant announcements, rather than merely capacity ^[19]. Whether West Texas pipeline capacity into export routes actually lands, since that is the link that transmits global pricing into hyperscaler fuel bills ^[15]. And whether the political cost widens: 80% of consumers already worry about data centers' effect on their utility bills, mostly electricity, and gas bills are the obvious next channel ^[17].