By 2035, American data centers could burn more natural gas than Germany and Japan combined. That’s the headline number from a new BloombergNEF report, and TechCrunch AI reports it’s nearly double what the same analysts predicted just nine months ago. Data centers are now expected to be the second-biggest driver of U.S. gas demand growth over the next decade, behind only LNG exports.
The forecast lands at roughly 18 billion cubic feet per day. And BloombergNEF already trimmed it for reality. The number assumes plenty of announced projects never get built.
Where the demand actually comes from
The onsite gas plants get all the attention. Meta, Microsoft, Google, and Amazon have each announced plans for natural gas generation that skips the grid entirely, according to TechCrunch AI. Those projects will burn 2.9 to 3.4 billion cubic feet per day by 2035. That alone matches what every U.S. data center consumes today, grid power included.
But here’s what stands out: the onsite plants are the small part. BloombergNEF expects grid-connected data centers to add another 15 billion cubic feet per day in power-sector gas demand by the mid-2030s. That’s five times more demand growth than every other grid-connected sector combined.
So the story isn’t Big Tech building private power plants. It’s Big Tech leaning on the public grid, and the grid answering with gas.
The price problem nobody’s pricing in
Most of today’s data center economics assume cheap, stable gas. Analysts at Noreva think that assumption is shaky. Data center demand plus surging LNG exports could push prices sharply higher.
Tech companies can absorb a gas spike. Their balance sheets are enormous. Utility ratepayers are a different story. If gas prices climb, households and small businesses pay more for electricity to subsidize an AI buildout they never signed up for. That’s a political problem waiting to happen, and it connects directly to the local backlash against data center projects we’ve been seeing across the country.
The climate math
TechCrunch AI cites IEA figures: one cubic foot of natural gas releases about 60 grams of CO2 equivalent, including extraction and distribution. Run that against the projected demand and data centers add roughly 1 million metric tons of greenhouse gas pollution every day. That’s about 12% of current total U.S. emissions.
A decade ago, hyperscalers were racing each other on net-zero pledges. Now they’re racing each other on gas turbines. The pledges haven’t been retracted, but the timelines have quietly stretched.
What this means over the next 1-3 years
This is significant because it reframes the AI infrastructure race as an energy commodities story. A few things to expect:
- Gas turbine shortages get worse. Lead times are already stretched. Companies that locked in supply early hold a real advantage.
- Ratepayer fights become a regulatory theme. Expect state utility commissions to start asking who pays for AI’s power demand, and tech companies to get pushed toward covering their own grid costs.
- Nuclear and geothermal deals accelerate. Gas is the bridge, but every hyperscaler knows the optics. Long-term PPAs for firm clean power become table stakes.
- Energy pricing enters the AI cost stack. If gas prices move, so does the cost of tokens. Inference pricing isn’t immune to commodity markets.
Practical takeaways
If you build on AI, watch the energy inputs, not just the model benchmarks. Cloud pricing over the next few years will reflect what power costs, and power will reflect gas. If you’re picking a region or a provider, ask where their electrons come from and what their long-term supply looks like.
If you invest, the thesis is straightforward: the AI boom is quietly a natural gas boom, and the grid infrastructure sitting between them is the bottleneck.
One caveat on the forecast itself. BloombergNEF doubled this number in nine months. That’s a sign of how fast the buildout is moving, but it’s also a reminder that projections in this space have a short shelf life. The 2035 number will change again. The direction probably won’t.
The full BloombergNEF breakdown, including the onsite versus grid split, is covered in detail at TechCrunch AI.