The demands of AI data centers have created a truly unprecedented era in the speed and scale at which power generation needs to be built. But there has been uncertainty about whether we truly need as much infrastructure as tech giants are demanding.
What’s becoming clear, though, is that the growth of AI has neither quickly reached a level where training models no longer require massive compute infrastructure, nor has it fizzled due either to a bubble bursting or else lackluster revenue from AI companies. The industry, as Varun Sivaram of Emerald AI said to me this week, is in a “goldilocks” period, where demand for compute increases steadily as AI models evolve.
For now, forecasts of growth into the tens of gigawatts can be believed — and the speed at which hyperscalers need this power is faster than any developer, utility, or regulator was prepared for.
The recent rise of AI data centers is quickly rewriting the script for literally all energy sources, from the restart of nuclear plants to the deferred retirements of coal plants. It provides the economic context for sources that had been only slowly entering the mix, such as enhanced geothermal, fuel cells, and nuclear SMRs. But none have generated the headlines that gas has in 2025.
Some of the major recent announcements include:
- Pacifico Energy project, Texas: Pacifico Energy has unveiled plans for a massive 5 GW off-grid fossil gas generation facility in Texas specifically designed to power a hyperscale data center campus, marking one of the largest single-site energy deployments for digital infrastructure. This is unique in that even at this scale, it’s designed to be entirely off-grid and supported by 1.8 GW of battery storage.
- Meta Louisiana Campus: Meta’s announced $10 billion data center — which Trump recently said would cost $50 billion — will be supplied by three fossil gas power plants, collectively producing 2.25 GW; Louisiana regulators recently granted their approval. The development is scheduled for commissioning in 2028 or 2029, with the possibility of expansion up to 5 GW.
- Vantage Data Centers Frontier Campus, Texas: Vantage said on August 19 that it would build a $25 billion, 1.4 GW data center campus called “Frontier” on 1,200 acres in Shackelford County, Texas. Access to gas supply from the Permian Basin was key to the location selection.
- OpenAI’s Stargate data center complex, Texas: The Abilene, Texas, project will begin with 10 simple-cycle gas turbines from GE Vernova and Solar Turbines, providing 360 megawatts of power, with plans to scale up to gigawatt-scale.
- Homer City Energy Campus,Pennsylvania: At the site of a former coal plant, developers led by Homer City Redevelopment and Kiewit are building a 4.5 GW gas-fired generation facility to supply a planned hyperscale data center campus. This project will utilize hydrogen-enabled turbines and is slated to start operation in 2027.
- Crusoe and Tallgrass data center, Wyoming: The 1.8 GW project will begin with gas power, with plans to incorporate carbon capture and storage as well as renewables as it scales to 10 GW.
- Powerconnex Inc. New Albany Energy Center, Ohio: This new data center campus in Ohio will include an onsite fossil gas power plant expected to generate up to 120 MW of electricity, supporting high-density compute loads for AI enterprises. Construction starts in the fourth quarter of this year.
- Thunderhead and HMC Partnership, multiple locations: Thunderhead Energy Solutions and Harbert Management Corporation are developing up to 1.5 GW of behind-the-meter fossil gas generation capacity. These projects are designed for expedited deployment and scalable growth at U.S. data centers, enabling up to 100 MW of capacity within two years.
Data center operators need firm power at scale, and gas developers have seized the opportunity to position themselves as the clear leaders, despite challenging economics, limitations in the supply chain, and the sheer scale of the capital required to bring projects online when compared to renewables and storage. The administration has provided every possible support to the gas industry, while dramatically altering the economics of renewables as well as their permitting and regulation.
What’s behind the sense that gas is back
But is gas winning? It depends.
The gas power industry went through a massive wave of deployments in the early 2000s as the fracking revolution brought low-cost supply to the U.S. But that was followed by a classic pattern of oversupply, depressed prices, and halted construction. Renewables and battery storage rose to prominence, and they now represent the vast majority of generation added to the grid each year in red and blue states alike; gas remains a distant third.
The high costs of gas — driven by tariffs, tight supplies of equipment, labor shortages, and construction materials — are keeping it less competitive than renewables, particularly in markets like ERCOT. In addition, the gas supply in the U.S. is largely spoken for, and the opening up of further LNG exports is creating global demand for gas and upward pressure on fuel prices.
Michael Thomas of Cleanview recently illustrated similar findings. His analysis found that while gas continues to be deployed at a pace of around 8.8 GW per year, that is no greater than the pace in the early 2000s boom (below, this point is illustrated in this week’s featured chart).
In the meantime, renewables and storage continue to make up the majority of new generation. So what explains the sense that gas is back, and winning the competition to power AI data centers?
First, there is clearly a powerful narrative at work. In the past, when new data centers would have ready access to existing grid power, there was no need to announce the source of generation. More often, the hyperscalers would contract renewable power separately as an offset to any emissions associated with the local grid that powered the data center.
But after data centers started growing into the hundreds of megawatts, building generation in tandem with the data center became a requirement. At this scale, gas generation became a more common choice because it is the “known quantity” when it comes to firm power. Gas is being presented as a bridge to a cleaner mix of power in the future, as recently evidenced by Oracle’s plan to use gas generation for a major Texas data center, as well as xAI’s Colossus data center in Memphis. It’s also being deployed as a behind-the-meter solution in cases where “bring your own generation” is the quickest path to power.
The result? More often than not, when major data centers are announced, so are major gas power plants.
Can gas continue to grow?
This will undoubtedly continue into the 2030s. It will be buoyed by both the loosening of supply constraints and increasing demand, particularly among vertically integrated utilities whose model rewards building and operating large capital projects like gas.
In a note to clients this week, the investment bank Jefferies made this observation: “Utilities have more advantages in natural gas development with land, permits, etc. and expertise in operating….Further, the loss of tax credits under OBBBA and normalization under IRA has shifted the playing field back toward utility ownership and natural gas.”
The final question, therefore, is whether gas can continue to grow at this pace — not exactly winning, but increasing its momentum thanks to data centers — while not increasing emissions at the same rate. To answer it, there are two developments I’m watching.
First, carbon capture and storage has in fact been seeing a resurgence, associated not only with power generation, but also with ethanol and fossil gas processing. The EIA sees sustained growth ahead for CCS, capturing 2% of overall energy emissions by 2039.
A recent white paper from Carbon Direct found that outfitting gas-fired generators with CCS “could supply up to 63% of U.S. data center electricity by 2030 while keeping net emissions low,” as CEO Jon Goldberg summarized in his recent opinion piece for Latitude Media. “CCS deployment is not a silver bullet, but its adoption reflects hyperscalers’ willingness to embrace pragmatic, scalable solutions for clean firm power.”
Second, data center developers are increasingly eyeing combined heat and power models of onsite generation. These have massive efficiency gains over traditional gas power generation by using the waste heat to drive absorption chillers for cooling. This option hasn’t yet been deployed at a meaningful scale, as today’s data center developers tend to be conservative about new technologies — but its mere existence indicates a way forward where emissions and efficiency can both be addressed at the site itself.
Combined, these both could be a bit of wishful thinking in an era in the U.S. when accountability for emissions is clearly taking a backseat to all-out growth. But there’s real evidence that hyperscalers care about the emissions associated with their infrastructure deployments, as they continue to invest in zero-carbon solutions like enhanced geothermal or nuclear SMRs. Gas will be part of the picture, but ideally the complex mix of market forces at work today will keep the impacts minimized while the energy transition continues.
A version of this story was published in the AI-Energy Nexus newsletter on August 27. Subscribe to get pieces like this — plus expert analysis, original reporting, and curated resources — in your inbox every Wednesday.
