The rapid expansion of AI workloads is transforming the US power landscape, with utilities increasingly turning to natural gas to meet soaring and volatile demand, raising questions about long-term decarbonisation.
Artificial intelligence is forcing a rethink of how the US powers its digital economy, and the answer so far is increasingly natural gas. What began as a planning challenge for utilities has turned into a scramble to secure firm electricity for hyperscale data centres, with developers and power companies now treating generation, transmission and storage as part of the same industrial package.
The scale of the demand shock is striking. Goldman Sachs Research has projected global data centre electricity demand to reach 84GW by 2027, up 50% from 2023, with AI workloads accounting for more than a quarter of that total. Bain, meanwhile, expects US data centre power consumption to more than double between 2024 and 2030. That kind of growth is colliding with a grid built for slower-moving demand and longer planning cycles.
The problem is not simply volume but volatility. Traditional computing loads are relatively stable, whereas AI training and inference can produce abrupt spikes as large GPU clusters switch on and off. Utilities are used to forecasting seasonal peaks and predictable industrial demand; they are far less equipped for megawatt-scale jumps within seconds. ERCOT, the Texas grid operator, has already moved to adjust its forecasting methods to account for the gap between projects announced and projects that actually materialise.
For now, gas is filling the gap because it is dispatchable, widely available and quicker to bring online than most alternatives. Interconnection queues for new wind and solar projects can stretch for years, while new combined-cycle gas plants can be delivered in a fraction of that time. That speed matters when data centre operators want power now, not in the next planning cycle.
That is why utilities are no longer simply selling electricity to large customers; they are co-building infrastructure around them. In Louisiana, Entergy is moving ahead with a multibillion-dollar gas buildout tied to Meta’s new campus, and reports this week said the social media company will fund additional gas plants, transmission lines, storage and nuclear-related upgrades as part of the broader site strategy. The scale of that arrangement underlines how tightly AI expansion is now bound to energy infrastructure.
Across the market, the shift is spreading well beyond one state. Axios reported that nearly three-quarters of planned on-site power equipment for new data centres now runs on natural gas, a sharp change from the near-universal reliance on grid electricity seen little more than a year ago. Elsewhere, developers are experimenting with everything from mobile turbines to fuel cells in an effort to secure immediate supply.
The engineering pressures are showing up on the transmission side too. Investment in many parts of the US slowed after 2015, leaving regional networks short of headroom just as large-load requests surged. In Texas, CenterPoint Energy said it saw a 700% jump in large-load interconnection requests over a 12-month period, while Virginia still has tens of gigawatts of data centre proposals in the queue. The Department of Energy has separately warned that congestion is already severe in multiple regions.
Costs are rising with the strain. New combined-cycle gas plants are significantly more expensive to build than they were only a few years ago, and capacity markets are also tightening. In PJM, the grid covering much of the Mid-Atlantic and Midwest, the clearing price for the 2026-27 delivery year jumped far above the level seen two years earlier, a sign of the premium now attached to guaranteed supply.
For industrial decarbonisation, the longer-term issue is that today’s workaround is tomorrow’s emissions lock-in. Gas plants typically operate for around 30 years, which means infrastructure approved now may still be running well beyond most net-zero deadlines. That would be a problem even before accounting for methane leakage across wells, pipelines and compressor stations.
The climate stakes are becoming harder to ignore. Researchers cited in recent reporting estimate that the current wave of gas buildout tied to AI could add tens of millions of tonnes of carbon dioxide to the atmosphere by the end of the decade. Another study found that US data centres already consume more than 4% of national electricity, with more than half of that power still coming from fossil fuels.
There are, however, some longer-term routes away from this dependence. The Inflation Reduction Act’s storage incentives are designed to make batteries more competitive, while restarted or extended nuclear plants are emerging as one of the few firm zero-carbon options capable of serving round-the-clock AI demand. Google’s agreement with NextEra to support a nuclear restart in Iowa reflects that logic. So does the broader push for flexible storage, demand response and cleaner on-site generation.
But none of those solutions move at the pace the AI market is demanding. The result is a widening gap between the speed of digital expansion and the slower machinery of power planning. For now, gas is the bridge. The question for utilities, regulators and industrial customers is how long that bridge can hold before it becomes a permanent road.
- https://www.zerohedge.com/energy/us-grid-wasnt-built – Please view link – unable to able to access data
- https://www.tomshardware.com/tech-industry/artificial-intelligence/meta-will-fund-seven-new-gas-plants-to-power-its-7gw-louisiana-data-center – Meta has partnered with Entergy Corp. to fund the construction of seven new natural gas power plants to support its AI data center operations at the Hyperion campus in Richland Parish, Louisiana. These plants will supply an additional 5.2 gigawatts of power, on top of the previously approved 2.3 gigawatts from three existing plants, bringing the total to over 7 gigawatts across ten facilities. The project also includes Meta funding 240 miles of transmission lines, energy storage systems, and nuclear power upgrades. While the projects await regulatory approval, they represent a significant investment in fossil-fueled infrastructure to meet AI energy demands.
- https://www.envirolink.org/2026/02/10/ai-data-centers-turn-to-natural-gas-as-power-demand-soars-threatening-climate-goals/ – The artificial intelligence boom is driving a massive surge in natural gas-powered data centers across the United States, potentially undermining climate commitments as tech companies scramble to meet unprecedented energy demands. With traditional power grids unable to keep pace with AI’s voracious electricity appetite, developers are increasingly turning to on-site fossil fuel generation, creating what experts warn could be a climate disaster in the making. Companies are getting creative with their power solutions—sometimes in surprising ways. Boom Supersonic, known for developing supersonic aircraft engines, recently signed a $1.25 billion deal to provide 29 jet-engine gas turbines for OpenAI data centers. Meanwhile, Meta’s Texas facility will rely on over 800 mobile mini-turbines, and massive projects like the 1.2-gigawatt Stargate campus in Abilene are sprouting across the country. Globally, more than 1,000 gigawatts of new gas-fired power projects are now in development—a 31% jump in just one year. The environmental implications are staggering. Cornell University researchers estimate this buildout could add 44 million metric tons of carbon dioxide to the atmosphere by 2030, equivalent to emissions from 10 million cars. Many of these improvised power solutions are actually worse for the climate than traditional power plants, as they use less efficient engines and turbines that produce more emissions per unit of electricity generated. Texas leads this fossil fuel rush with nearly 58 gigawatts of planned natural gas capacity—more than the next four states combined. The trend raises serious questions about whether the AI revolution’s promise of technological progress is worth its potential climate costs.
- https://www.axios.com/2026/02/04/ai-natural-gas-power-data-centers/ – A recent report by Cleanview reveals that the surge in artificial intelligence (AI) is driving a significant increase in natural gas use at data centers. Nearly 75% of the planned on-site power equipment for these centers operates on natural gas, highlighting a departure from prior strategies that relied exclusively on electric grids. This marks a notable shift, as just over a year ago, nearly all developers intended to use grid power for their operations. Cleanview analyzed plans for 46 data centers, which collectively aim to generate 56 gigawatts of power independently. The rush to secure reliable and immediate electricity—driven by the intense demand of AI—appears to be outweighing the long-term cost and environmental benefits of renewable sources like wind, solar, and batteries. Although natural gas has faced political opposition in the past, it is now regaining prominence, particularly in regions at the heart of the AI boom. The trend also boosts companies like Bloom Energy, which produces natural gas-powered fuel cells, and exemplifies the changing energy landscape influenced by AI’s growth.
- https://www.theguardian.com/environment/2026/jan/29/gas-power-ai-climate – The US is leading a huge global surge in new gas-fired power generation that will cause a major leap in planet-heating emissions, with this record boom driven by the expansion of energy-hungry datacenters to service artificial intelligence, according to a new forecast. Projects in development are expected to grow global capacity by nearly 50% amid growing concern over impact on the planet. US gas projects in development will, if they are completed, produce 12.1bn tonnes in carbon dioxide emissions over their lifetimes.
- https://www.bain.com/insights/can-the-us-energy-grid-keep-up-with-ai-data-centers-snap-chart/ – US data centers’ electricity consumption is projected to more than double from 2024 to 2030, according to the latest baseline forecast from Bain’s Data Center Model. Although sufficient supply is expected to meet demand, the baseline scenario estimates total US consumption in 2030 will be about 150 terawatt-hours higher than the amount forecasted by the US Energy Information Administration. Meeting the demand will require a diverse set of coordinated actions, including promoting flexible power solutions (e.g., battery energy storage systems) integrated with the grid to improve its reliability; implementing demand response programs that incentivize adaptable electricity usage; and facilitating on-site power generation—such as natural gas, rooftop solar, and restarted nuclear plants—which can be developed faster than utility-scale solutions.
- https://arxiv.org/abs/2411.09786 – The rapid proliferation of data centers in the US—driven partly by the adoption of artificial intelligence—has set off alarm bells about the industry’s environmental impact. Researchers compiled detailed information on 2,132 US data centers operating between September 2023 and August 2024 and determined their electricity consumption, electricity sources, and attributable CO₂e emissions. Findings reveal that data centers accounted for more than 4% of total US electricity consumption—with 56% derived from fossil fuels—generating more than 105 million tons of CO₂e (2.18% of US emissions in 2023). Data centers’ carbon intensity—the amount of CO₂e emitted per unit of electricity consumed—exceeded the US average by 48%. The data pipeline and visualization tools can be used to assess current and future environmental impacts of data centers.
Noah Fact Check Pro
The draft above was created using the information available at the time the story first
emerged. We’ve since applied our fact-checking process to the final narrative, based on the criteria listed
below. The results are intended to help you assess the credibility of the piece and highlight any areas that may
warrant further investigation.
Freshness check
Score:
7
Notes:
The article was published on April 24, 2026, and has been republished across various platforms, including low-quality sites and clickbait networks. The earliest known publication date of substantially similar content is April 24, 2026. The narrative is based on a press release from Interesting Engineering, which typically warrants a high freshness score. However, the widespread republication across low-quality sites raises concerns about the originality and freshness of the content. Additionally, the article includes updated data but recycles older material, which may affect its freshness. Given these factors, the freshness score is reduced to 7.
Quotes check
Score:
6
Notes:
The article includes direct quotes from Goldman Sachs Research and other sources. However, searches for the earliest known usage of these quotes reveal that they have appeared in earlier material, indicating potential reuse. Variations in quote wording between sources have been noted, which raises concerns about the accuracy and consistency of the quotes. No online matches were found for some quotes, making independent verification challenging. Given these issues, the quotes score is reduced to 6.
Source reliability
Score:
5
Notes:
The narrative originates from a major news organisation, ZeroHedge, which is known for its critical and often controversial perspectives. While it is a reputable source within its niche, its reach and influence are limited compared to mainstream media outlets. The article appears to be summarising or rewriting content from a paywalled source, which raises concerns about the independence and originality of the content. Given these factors, the source reliability score is reduced to 5.
Plausibility check
Score:
7
Notes:
The article makes claims about the strain on the US power grid due to AI data centers and the reliance on natural gas to meet this demand. These claims are plausible and align with industry trends. However, the lack of supporting detail from other reputable outlets and the absence of specific factual anchors raise concerns about the accuracy and depth of the reporting. Given these issues, the plausibility score is reduced to 7.
Overall assessment
Verdict (FAIL, OPEN, PASS): FAIL
Confidence (LOW, MEDIUM, HIGH): MEDIUM
Summary:
The article raises valid concerns about the strain on the US power grid due to AI data centers and the reliance on natural gas to meet this demand. However, issues such as potential reuse of content, reliance on paywalled sources, and lack of independent verification sources significantly undermine its credibility. Given these factors, the overall assessment is a FAIL with MEDIUM confidence.
