By the middle of the next decade, the electricity demand of American data centers will require enough natural gas to make the sector, on its own, a bigger gas consumer than all but a handful of the world’s nations — not because of some exotic new technology, but because the industry chose the fastest, most available fuel to keep pace with the computing boom.
Key Points
- BloombergNEF’s revised 2035 outlook projects U.S. data centers will add 15 billion cubic feet per day (Bcf/d) of natural gas demand for electricity generation.
- That growth would make AI infrastructure the second-largest driver of U.S. gas-demand growth through 2035, trailing only new LNG export terminals on the Gulf Coast.
- Natural gas already supplies over 40% of the electricity powering U.S. data centers, giving the sector a well-established pathway to expand gas use further.
- Independent forecasts from Goldman Sachs, PwC, RBC, and EPRI vary widely — from roughly 3.3 Bcf/d to over 11 Bcf/d by 2030-2035 — underscoring how sensitive these projections are to assumptions about AI growth and grid buildout.
- Gas infrastructure firms are already committing billions in capital — Williams Company alone has targeted 6 gigawatts of power projects and about $10 billion in investment tied partly to hyperscale data centers.
Why Gas Won the Race to Power AI
The mechanism behind this shift is straightforward once you understand how the U.S. grid actually operates. Data centers, especially the hyperscale campuses built to train and run large AI models, need power that is available around the clock and can be delivered on a construction timeline measured in months, not decades. Natural gas turbines can be sited, permitted, and connected far faster than nuclear reactors or even most large-scale renewable-plus-storage projects, and gas plants can run at high utilization regardless of weather. The International Energy Agency’s own accounting shows natural gas already supplies more than 40% of the electricity consumed by U.S. data centers today, the largest single source by a wide margin, which means new AI load is landing on a grid that is already gas-dependent by default rather than by ideology.
That existing reliance is what makes the forward-looking numbers so consequential. BloombergNEF’s revised 2035 outlook, published in September 2026, puts incremental gas demand from data centers at 15 Bcf/d — a volume large enough that reporting built on the forecast places U.S. AI infrastructure as the second-largest source of American gas-demand growth through 2035, behind only new LNG export terminals along the Gulf Coast. To put that scale in perspective, a country burning 15 Bcf/d of gas would rank among the world’s larger national consumers outright, which is the basis for characterizing data centers as poised to out-consume most nations on this single fuel.
How the Forecasts Stack Up Against Each Other
No single institution owns this number, and the spread across forecasters is itself instructive. Goldman Sachs, working from a projected 15% compound annual growth rate in data center power demand between 2023 and 2030, estimates about 3.3 Bcf/d of new gas demand by the end of the decade. RBC’s capital markets research places 2030 gas consumption from data centers closer to 6.1 billion cubic feet daily, a figure the firm says is substantially higher than its own prior estimate. PwC’s scenario work, built around bear, base, and bull cases for data center buildout, puts 2035 incremental gas demand in a range of roughly 7.6 to 11.5 Bcf/d. BloombergNEF’s 15 Bcf/d sits at the top of that range, reflecting either a more aggressive AI growth assumption, a longer forecast horizon, or both.
This dispersion is not a weakness in the underlying story so much as a reflection of how young and fast-moving the AI buildout still is. Every one of these estimates depends on assumptions about how quickly hyperscalers keep building, how efficient future AI chips become, and how much of the new load gets served by gas turbines versus nuclear power-purchase agreements, on-site generation, or demand-response programs that shave peak usage. EPRI’s own modeling captures this uncertainty directly, estimating that data centers could consume anywhere from 9% to 17% of total U.S. electricity by 2030 — a range wide enough to accommodate both a modest and an extraordinary gas-demand outcome.
The Capital Already Committed
What separates this forecast from a purely theoretical exercise is the money already moving. Williams Company, one of the largest natural gas pipeline and infrastructure operators in the country, has targeted 6 gigawatts of new power projects and announced roughly $10 billion in related investment, with company leadership explicitly tying that buildout to hyperscale data center customers, including facilities serving Meta. That kind of capital commitment does not happen on a whim inside a regulated, capital-intensive industry; pipeline and turbine orders lock in years ahead of construction, which means the gas industry itself is already underwriting a meaningful share of the demand growth that BloombergNEF, Goldman Sachs, and PwC are separately modeling from the electricity-consumption side.
15 Bcf/d of new data-center gas demand by 2035 is the quiet AI story. Compute needs power. Power needs gas.
— Ericnomics (@_ericnomics) September 15, 2026
What This Means for the Grid, Emissions, and Energy Policy
The practical stakes extend well beyond a single statistic. A gas-demand increase of this magnitude implies new pipeline capacity, new compressor stations, and new turbine orders concentrated in the regions where data centers are clustering — Virginia, Texas, Ohio, and the broader Southeast among them. It also raises the emissions question that accompanies any large expansion of fossil generation, even as some hyperscalers pursue nuclear and renewable offsets in parallel. And it puts pressure on regulators and grid operators to plan interconnection queues and reliability standards around a demand source that barely registered in load forecasts a decade ago. Because the entire projection rests on continued AI-driven growth, any meaningful slowdown in hyperscale construction or leap in chip efficiency would pull the gas-demand curve down as quickly as it has risen — a reminder that this is a live, evolving buildout rather than a fixed destination.
Sources:
feedpress.me, iea.org, yahoo.com, kapsarc.org, cryptobriefing.com, thetradable.pro
© featurednews.com 2026. All rights reserved.
