Data Centers Driving U.S. Gas

A new Macro to Micro report from a team led by Scott Gruber looks at how U.S. power-load growth through 2030 will be driven by the power demands of new data centers, generating additional demand for natural gas. But determining how much added demand is key to separating potential from hype.

Anything AI-related is getting significant attention, with a spotlight on the theme of AI and data centers driving power demand. Any long-range forecast for a technology growth sector requires broad error bars, but leveraging recent modeling from our Technology and Telecom teams allows us to present a forecast for AI/data-center associated power demand based upon a chip-supply approach. Our base case is for U.S. power demand associated with data centers to jump from ~21.4 Gigawatts (GW) in 2023 to ~52.0 GW in 2030, or from 187 terawatt-hours (TWh) to 456 TWh. As a result, data centers would equate to 10.9% of total U.S. power demand in 2030, up from 4.5% today.

That assumes ~80% of global data-center growth is in North America and ~73% in the U.S. As we translate IT loads to power demand, our assumption is that power utilization efficiency gradually declines in line with historic relationships, and that data-center power capacity utilization levels are ~65%.

Our 2028-2030 base case is built on the implication that U.S. utilities will slow the retirement of coal, gas and nuclear-power plants and execute on plans to build power plants and transmission and distribution infrastructure to keep the lights on. This implies increases of ~4% to ~6% in U.S. utilities’ capex budgets on average from 2024 to 2030, with large geographic and company-specific diversions in outcomes. These capex increases are driven by ~26 GW of gas new-build plants, some other power resources, and associated infrastructure investments.

While data-center-driven gas demand is certainly positive for the domestic natural-gas market, at least through 2030, the overall market outlook appears more dependent on LNG export growth as well as the depth of low-cost gas inventory to meet rising calls on U.S. gas production. Our bull case for domestic gas assumes healthy demand growth from the power sector and full utilization of LNG plants under construction.

But in our note we go several steps further, including assessing how data-center demand could increase global demand for natural gas. We also investigate how the supply chain may respond to meet the incremental call on gas with respect to incremental gas rigs needed and pipeline requirements.

In a bull case for U.S. natural gas, the data-center-driven growth in power demand for gas is incremental to LNG export growth (~12.6 billion cubic feet per day [bcfd] under construction but with upside), and complemented by other growth drivers of as much as 3 bcfd. That means the call on U.S. gas could grow by ~20 bcfd or more by 2030. In this scenario, growth in low-cost gas appears insufficient to meet the call, requiring activation of Tier II Haynesville/Marcellus acreage as well as Eagle Ford dry gas, Mid-Con gas and S. Delaware gas. We think this would push the marginal cost of gas from ~$2.75 per million cubic feet (mcf) toward ~$3.75/mcf and require ~70 incremental gas rigs.

While the U.S. should remain the leader in data-center expansion, the build-out will likely be world-wide, driving power demand growth. This could in turn increase demand growth for gas and hence LNG. But the data-center expansion will likely coincide with a slowing demand growth profile for LNG given renewables growth in China and nuclear restarts in Japan, to name just two factors. At the same time, LNG production capacity is set to expand by ~24 bcfd based only upon facilities under construction, with a long list of projects in the works. Therefore, we see a risk of oversupply in the global LNG market driving the need to curtail shipments by an estimated ~2 bcfd in 2026, rising to ~9 bcfd in 2030. Moreover, the U.S. may need to shoulder a majority of these curtailments given relative costs; the curtailments could be larger than the incremental gas demand growth from data-center expansion.

A reduction in the potential call on U.S. gas from ~20 bcfd toward 10–14 bcfd would have a material implication for the marginal cost of domestic gas. We forecast another ~5.5 bcfd of growth from the Permian, ~2.5 bcfd of growth from Dorado and the potential for 3+ bcfd of growth from each of Appalachia and Haynesville. Thus, low-cost basins appear to have the potential to meet all the incremental call if material LNG curtailments arise. That said, we believe there could be modest upward pressure on the marginal cost of U.S. gas from ~$2.75/mcf to ~$3.25/mcf over the next several years based upon resource maturation in Appalachia and Haynesville. From a rig-count perspective, the lower call on U.S. gas would require an incremental ~40 rigs.