If you want to see how much gas will be powering the growth of data centers, just take a look behind the meter.
In a recent note to investors, the investment bank Jefferies wrote that EPCs and IPPs are increasingly embracing large, gigawatt-scale behind-the-meter generation projects to avoid waiting in lengthy grid interconnection queues, while staying clear of accountability for impacting electricity rates.
They went on to say, “BTM deployment is gaining traction, especially under [Texas and PJM’s ‘bring your own generation’ frameworks], as a strategy to manage capacity prices and utility bills. Our recent call with McKinsey indicated that 25-33% of incremental data center demand through 2030 will be met by BTM solutions.”
That level of demand is extraordinary. Forecasts for demand vary widely, but average around 100 GW by 2030. If a quarter to a third of that is BTM, that suggests up to 33 GW of BTM generation will be deployed over the next five years. Some of that will be renewables and storage, and some may be nuclear — but gas is the clear beneficiary. In fact, states like Oklahoma and Wyoming are out explicitly advertising their gas resources for data center developers who want onsite generation.
A recent TechCrunch article asserted that “the AI era is giving fracking a second act.” It pointed to the recent announcement that coding startup Poolside and the neocloud company CoreWeave are building Project Horizon, a 2 GW Texas data center with access to local fossil gas, as well as to an old gas power plant built by Occidental Petroleum.
Meanwhile, last week, a Wall Street Journal feature noted that xAI, Meta, and OpenAI have already opted for BTM gas generation, while Equinix has committed to onsite fuel cells.
Major announcements of BTM generation include:
- The OpenAI’s Stargate initiative, which is without question the biggest driver of this trend, recently announced five new AI data center sites across the U.S., bringing its total commitments to date to 7 GW of planned capacity; the project has a goal of 10 GW of capacity committed by the end of 2025. Much of this capacity is gas-fired, and will operate behind the meter for both primary and backup power, with integration to grid power and battery storage where available.
- Just this week, Joule announced a data center development in Utah that will incorporate 1.3 GW of on-site BTM generation and operate fully islanded from the local Rocky Mountain Power grid.
- Texas gas solutions company VoltaGrid, along with midstream gas operator Energy Transfer, last week announced they will be supplying Oracle with 2.3 GW of off-grid fossil gas power for its data centers. Earlier this year, VoltaGrid announced it would deploy over 1 GW of gas power solutions to Vantage Data Centers.
- In a fact sheet posted on US Representative Jodey Arrington’s site, a new Stargate data center in Shackleford County, Texas, is described as being “powered by an onsite, behind-the-meter, gas-powered microgrid utilizing Jenbacher reciprocating engines.” One in Doña Ana County, New Mexico, is described as powered by “an onsite, behind-the-meter, gas-powered microgrid utilizing Siemens and GE gas turbines will power the campus independently from the local grid, ensuring there is no impact on local electric bills or supply reliability.”
- Fermi America’s recent S-1 filing to go public describes their Project Matador as a “multi-phased project designed to deliver up to 11 GW of behind-the-meter energy and support up to 15 million square feet of AI-ready hyperscale compute infrastructure by 2038.” Time will tell if it’s real.
- In Evanston, Wyoming, Prometheus Hyperscale and Engie are planning a 1.2 GW data center that will include onsite gas-fired generation, with a plan to incorporate Oklo’s small modular nuclear reactors behind the meter in the future.
- In the Texas Permian Basin, home to the original fracking boom, FO Permian Partners and Hivolt Energy recently announced a 5 GW off-grid gas power solution for Texas data centers.
- Not all gas will go to conventional power plants or reciprocating engines. For its part, Bloom Energy recently received a commitment from Brookfield to invest up to $5 billion into its onsite BTM solid-oxide fuel cell power systems. They use fossil gas as their fuel source, though without needing to combust the gas.
- Earlier this year, Prometheus Hyperscale announced it is partnering with Conduit Power to build up to 300 MW of BTM gas-fired generation and battery storage at two data centers it’s building at Engie battery sites in Texas.
- International Electric Power is planning a 944 MW gas plant in Pennsylvania to power a data center. It will integrate with battery storage and backup power, and avoid interconnection with PJM for the time being.
In a recent analysis, the real estate services company JLL sees momentum behind BTM as well. Feeling sanguine about the state of the AI infrastructure market as it heads into 2026, JLL sees 8 GW of colocation capacity under construction, 73% of which is preleased.
More importantly, JLL states that developers now view “power constraints not as temporary obstacles but as permanent market managers that will define development patterns for the next decade. The new reality requires developers to secure power commitments years in advance, fundamentally altering project timelines and capital deployment strategies.”
In this context, JLL sees utilities and regulators increasingly playing a stronger hand by requiring “million-dollar application deposits, letter of credit obligations and take-or-pay policies covering 85% of requested power consumption,” which has led to a surge in adoption of BTM gas generation systems, either as a bridge to grid power, or as long-term solution augmented by grid-connected renewables and storage.
But will the demand last?
A key question that remains is whether BTM will keep making up a third of the data center power market for decades — or just in this period where speed-to-power decides every debate.
Currently, AI labs and hyperscalers alike face the financial reality of GPU depreciation. If you have chips, you need them operating immediately to start recovering that investment. And to do that, they need power. If they wait even three to five years, the GPUs are typically relegated to lower-level tasks, or just sold on secondary markets and replaced with the latest, and undoubtedly very expensive, generation.
Jim Simonelli, CTO of Schneider Electric’s Data Center Business, comes to this conversation with a view informed by Schneider’s joint venture with AlphaStruxure, which designs, builds, owns, and operates energy infrastructure for many data center companies. First, he doesn’t see most of the major data center companies wanting to actually own and operate power systems. Second, he believes most of these companies see the value in being “grid interactive” even when bringing their own generation or capacity, so that needs to be part of the system design from the beginning. Earlier this year, AWS confirmed to Latitude Media that its goal is to “get back to being grid-tied.”
He puts this preference down to the unique load profile of massive data centers, which can be highly variable and benefit from the power smoothing of a grid connection, particularly with the help of energy storage systems.
How this trend plays out is important. With generation, particularly gas, devoted solely to a single data center customer, the market is in some ways insulated from those costs being shifted onto ratepayers — but those ratepayers are also kept away from the benefits of that generation on the grid.
While the focus has understandably on this risk of costs being unfairly socialized, it’s worth considering the value of these assets being socialized. The grid needs power, it needs improvements, it needs investments in resilience. If paid for properly, this generation can find value in being available to all users of the grid. Otherwise we could be looking at a further balkanized power system, with the most advanced systems dedicated exclusively to the largest companies in the country.
Navigating the tradeoffs and regulatory questions is a big task ahead. But we have the superintelligence of AI to help, right?
A version of this story was published in the AI-Energy Nexus newsletter on October 22. Subscribe to get pieces like this — plus expert analysis, original reporting, and curated resources — in your inbox every Wednesday.
