Amazon, Google, Meta and Microsoft are turning to natural gas to power the rapid expansion of AI data centers. The strategy offers something renewable projects cannot always deliver on demand: large quantities of dispatchable electricity near facilities that need to run continuously.
A forecast from energy research firm Noreva argues that today’s cheap fuel may give hyperscalers a false sense of security. The firm expects data-center demand, slower supply growth and rising liquefied-natural-gas exports to tighten parts of the US market. In its scenario, gas could exceed $10 per million British thermal units at some delivery hubs, compared with roughly $2 to $4.50 today.
That is a projection, not the present market consensus. Futures prices currently indicate relative stability. Yet the scale of the infrastructure now being planned means even a low-probability price shock deserves attention: the companies are committing capital to assets whose operating economics will remain exposed to fuel markets for years.
Hyperscalers are becoming energy developers
Meta said in March that it would build a 7.5-gigawatt gas plant in Louisiana for its Hyperion data center. Microsoft and Google announced gigawatt-scale projects in Texas days later, while Amazon plans a 7.6-gigawatt facility in the same state.
These projects pull technology companies further into physical infrastructure and commodity risk. Hyperscalers have historically preferred business models built around software and leased capacity. AI demand is now pushing them toward power-generation decisions more familiar to utilities, energy traders and industrial manufacturers.
Noreva CEO Peter Gardett told TechCrunch that at least one investor was surprised by the amount of gas-price exposure hyperscalers appeared willing to accept. His concern is not that building gas generation is irrational at current prices. It is that contracts and facilities designed around today’s conditions may look very different if fuel doubles or triples.
Fuel accounts for about half the cost of electricity from a large gas plant, according to figures cited by TechCrunch. A sustained price increase could therefore raise the cost of producing AI tokens, compress cloud margins or encourage operators to shift demand back to public grids.
Cheap regional gas may not stay isolated
Natural-gas prices have remained subdued as new supply offset declining production from older wells. Noreva expects producers to keep adding supply, but at a slower pace and at greater expense. At the same time, new pipelines are connecting previously discounted regions to national and international markets.
West Texas illustrates the change. Gas there has often been an unwanted by-product of oil production, and limited pipeline capacity left local producers with few buyers. Data centers could consume that inexpensive fuel close to its source. New export routes, however, allow the same supply to reach customers willing to pay more.
Gardett argues that expanding LNG exports and AI demand will tighten the market together. Regional bottlenecks could create sharp differences between hubs, with some locations experiencing extended prices above Noreva’s $10 forecast even if national averages remain lower.
The limitation is clear: the scenario depends on assumptions about demand, drilling, pipeline capacity and exports. Current futures traders are not pricing in the same disruption. Noreva’s forecast should therefore be treated as a risk case rather than a settled outcome.
Consumers could inherit part of the volatility
If private plants become expensive, hyperscalers may reconnect more of their demand to utility grids. That could spread the cost beyond cloud customers. TechCrunch cites research finding that 80 percent of consumers already worry about data centers raising their electricity bills; pressure on gas supplies could add household heating costs to that debate.
The exposure also complicates technology companies’ public climate narratives. After years of signing wind and solar agreements, hyperscalers are embedding themselves in fossil-fuel infrastructure because AI capacity is growing faster than cleaner power and grid connections can be delivered.
Natural gas may still provide a workable bridge. It can be built near data centers and dispatched regardless of weather, and today’s prices make the economics understandable. The unresolved issue is who absorbs volatility over the plant’s life: the operator, cloud customer, utility ratepayer or household gas consumer.
AI infrastructure is making energy procurement a material part of technology strategy. If Noreva’s tighter-market scenario arrives, earnings calls may begin to discuss gas hubs alongside GPU supply—and a shortcut to power could become a long-term cost constraint.
