The bet hyperscalers just made
Over the past year, the largest cloud providers moved decisively toward natural gas as the fastest way to bring dedicated power online for AI data centers, sidestepping the years long queues now backing up at grid operators nationwide. Meta committed to a 7.5 gigawatt plant in Louisiana. Amazon locked in 7.6 gigawatts in Texas. Microsoft and Google each committed to gigawatt scale gas plants of their own in Texas. Combined, these commitments represent one of the largest simultaneous corporate bets on a single fuel source in recent memory, spanning four companies that between them account for the majority of global AI infrastructure spending this year. Each project was announced within months of the others, suggesting a coordinated read of the power market rather than four independent decisions arrived at separately.
The appeal is straightforward: gas plants can be built faster than nuclear and do not depend on the same grid interconnection queues currently backing up in Texas and across FERC jurisdiction territory. Speed of deployment and price stability are different questions, and a new forecast from energy research firm Noreva argues hyperscalers have answered the first question thoroughly while leaving the second largely unpriced. That gap between how much attention each question received is the core of Noreva's warning.
What the price forecast actually says
Noreva projects natural gas prices could triple in parts of the United States in the coming years, potentially exceeding $10 per million BTU in certain hubs. Current prices range from roughly $2 to $4.50 per million BTU, with Henry Hub in Louisiana sitting just under $3. A move to $10 would not be a modest cost increase, it would be a fundamental repricing of the fuel underpinning some of the largest data center buildouts in the industry's history.
Peter Gardett, Noreva's CEO, put it bluntly: 'I think everyone in the energy markets has been lulled into a sense that gas prices can't go up.' Gardett said hyperscalers are 'doing things that are not normal for an off-taker to do' when it comes to price risk exposure, a pointed critique from an industry veteran watching companies with no prior track record in energy markets sign long term fuel commitments at a scale utilities themselves would treat cautiously.
Why fuel price risk hits differently for compute buyers
Fuel represents roughly half the operating cost of a large power plant, which means a gas price spike does not stay contained to the energy sector. It flows directly into the cost structure of every data center that plant powers, and from there into the price of every AI token, training run, or inference call that data center supports. For an industry that has spent two years selling customers on falling per token costs, a structural gas price shock would be an unwelcome and largely unpriced variable running in the opposite direction.
The risk is compounded by scale. A hyperscaler locking in a single gigawatt scale gas plant is effectively making a multi decade bet on fuel economics through one contract, at a size where even routine market volatility becomes a material line item. If Noreva's forecast is even partially right, the companies most exposed are the ones that moved fastest and largest into dedicated gas generation, precisely the four hyperscalers now running billions of dollars in gas commitments at once.
How this compares to the alternative bets
Microsoft, Amazon, and Meta have each supplemented their gas commitments with nuclear power agreements, including Microsoft's 835 megawatt Three Mile Island restart deal and Amazon's 1,920 megawatts through Talen Energy, arrangements that typically carry longer term fixed pricing structures than spot exposed gas contracts. Gas plants offered speed that nuclear restarts could match only years later, and that speed advantage came bundled with a different risk profile, one Noreva's forecast is now putting a specific number on for the first time in public.
How that tradeoff plays out depends entirely on how the gas market actually moves over the next several years, something nobody, including Noreva, can guarantee with certainty. The hyperscalers made a concentrated bet on a single commodity at a scale few corporate buyers outside the utility sector have ever attempted, and that concentration is itself the risk worth watching regardless of which direction prices ultimately move. A diversified power portfolio across gas, nuclear, and renewables would have spread this exposure more evenly than the current mix appears to.
Why the risk is likely to surface gradually, not all at once
AI compute prices are unlikely to spike tomorrow on the back of a single forecast. Most current contracts probably carry some price protection already, and gas markets do not typically move from stable to tripled overnight without warning signs along the way. The more realistic path is a gradual repricing that shows up first in new contract terms and later in renewal quotes, giving attentive buyers a window to renegotiate or diversify before the full cost lands on their invoice. Watching quarterly gas price data alongside your provider's capacity announcements is a low effort way to catch that shift early.
That gradual timeline is exactly why this forecast deserves attention now rather than later. As contracts come up for renewal over the next two to three years, the underlying fuel economics behind a provider's power supply will matter more than at any point earlier in this AI buildout, and understanding the exposure today gives a buyer leverage that discovering it in a renewal quote does not. Buyers who raise the question early are the ones most likely to get favorable terms written into their next contract.
What buyers of AI compute should do with this
If you are negotiating a multi year AI compute contract, ask your provider directly what share of the power behind that capacity is fixed price versus exposed to spot natural gas markets. Vendors rarely volunteer this breakdown, but it is a legitimate diligence question now that a credible energy research firm has put a specific price scenario on the table. A provider heavily reliant on newly built gas plants in Texas or Louisiana carries a cost structure that could shift meaningfully if Noreva's forecast plays out even partially.
Push for contract language that caps or shares fuel price risk rather than passing it through entirely, the same protection utilities negotiate with their own suppliers as a matter of course. A vendor unwilling to discuss its fuel exposure at all is itself useful information, since it suggests the provider has not yet modeled this risk internally either, let alone built a plan to absorb it without passing the cost straight through to customers like you.
