Natural Gas Data Center Demand: The AI Power Squeeze Explained

Gas becomes the default fuel for AI

The AI infrastructure buildout is increasingly a natural gas story. Data center developers have announced roughly 101 gigawatts of on-site gas generation. The main reason is that they want to avoid long waits to connect to the grid and to lock in steady baseload power. 1 Big tech names are part of the shift. Microsoft and Meta have both been building gas-fired plants to run their facilities. 2

The scale of that demand is now showing up in fuel forecasts. BloombergNEF expects gas burned to generate electricity for US data centers to rise by 15 billion cubic feet per day over the decade to 2035. That forecast already assumes many announced projects will never be built. 4 The increase alone would exceed the current gas consumption of every country except China, Russia, Iran and the US. It is also more than double the 6.9 bcf/d BloombergNEF projected as recently as December. 4

Why the pipeline keeps growing

The project queue explains the revised outlook. Department of Energy figures show about 36.6 GW of data center capacity under construction and 201.5 GW in planning across the US. 1 The first quarter of 2026 added another 16 GW. That brought the planned gas project pipeline to 64 GW by 2030, and individual projects are also getting larger: new ones announced in 2025 averaged around 590 megawatts. 1

Global academic estimates point the same way, though they measure electricity rather than gas. A study in Nature projects that AI-focused data centers could use between 239 and 295 terawatt-hours by 2030, up from about 118 TWh in 2024. That would be roughly 1% of global electricity demand. 5 The study also finds the buildout heavily concentrated. North America, Western Europe and Asia-Pacific hold more than 90% of projected computing capacity. Smaller systems such as Oregon, Ireland and Iowa feel more strain than large ones like Texas, which can absorb new load more easily. 5 That geographic unevenness helps explain why developers are reaching for self-supplied gas rather than relying on local grids.

The cost of building gas plants

The fuel itself is not the expensive part. US natural gas prices have stayed low despite the war in Iran. 2 Building the plants is a different matter. BloombergNEF data shows the cost of a new combined-cycle gas turbine plant rising from under $1,500 per kilowatt in 2023 to $2,157 last year. That is a 66% jump, and construction timelines have stretched about 23%. 2 One summary of the report described costs as having "nearly doubled" in two years, which overstates the 66% figure. Either way, the direction is clear. 2

Equipment scarcity is part of the picture. GE Vernova has sold out heavy-duty turbine deliveries through 2028, with only about 10 GW of slots left for 2029 and 2030. 3 Its backlog has reached 100 GW of contracted capacity: 44 GW in firm orders and 56 GW in slot reservations. Lead times now exceed three years. 3

Where the real bottleneck sits

Accounts differ on what is actually slowing projects down. RBC Capital Markets lists turbine supply constraints among several structural headwinds, along with regulatory changes and local opposition. 1 GE Vernova's leadership disagrees that turbines are the limiting factor. The company says permitting is what pushes projects to different sites, and that it is not "gating" buildouts. 3 A turbine maker has an obvious interest in that view, but the point should not be dismissed. RBC notes that community opposition blocked or delayed about $98 billion in projects in 2025 alone. 1

The most reasonable reading is that no single factor is the constraint. Turbine order books, permits, construction costs and local resistance together set the pace. Gas supply is not the problem. Gas plants are, along with the approvals and equipment needed to build them. When equipment is booked for years and plant costs climb by two-thirds, the economics of a 590-MW behind-the-meter project change significantly.

What it means

Three implications stand out. First, AI's energy footprint is becoming a measurable factor in US gas demand, not just a grid planning issue. BloombergNEF's doubled forecast shows how fast expectations are moving. 4 Second, rising capital costs and multi-year lead times favor developers with deep pockets and early turbine reservations, which could concentrate the buildout among the largest players. 23 Third, resistance from communities and regulators may decide where facilities end up as much as power availability does. 13

If the forecasts hold, the cheap-gas assumption behind many AI plans will be tested. The risk is not necessarily higher fuel prices but higher costs to build the machinery that burns the fuel. Investors and utilities should watch turbine slot availability and permitting outcomes as closely as gas prices.