Natural gas is AI's new bottleneck, according to Anthropic's S-1

The article, published by Yahoo Finance and written by Jake Conley, argues that access to energy has become one of the biggest bottlenecks in deploying artificial intelligence, with natural gas taking center stage because it is cheap and abundant in the United States. The trigger is Anthropic's IPO prospectus (S-1), which according to analyst James West, managing director at Melius Research, shows that natural gas is “an AI vertical.” In his note to clients he wrote that “AI compute is a natural gas demand story.”

The backdrop is data center power demand. According to Goldman Sachs research cited in the piece, electricity consumption by these centers in the US would double between 2025 and 2027, after some twenty years of virtually stagnant grid expansion. To get around the bottleneck, developers are increasingly turning to “behind-the-meter” generation: building their own generation infrastructure or buying capacity from an existing provider. Key components, such as GE Vernova's gas turbines, already have waiting lists stretching several years.

Anthropic's S-1 acknowledges this strain in its risk factors section. The company states that as data center operators lean toward gas-fired generation and regulation holds back some of those projects, it expects “increased competition among users for the same natural gas supply and pipeline capacity.” That competition, it adds, could hurt its ability to contract firm fuel supply and transportation, and to secure pipeline connections and laterals to its facilities, as well as drive up electricity costs due to the imbalance between supply and demand.

Anthropic also notes that equipment such as transformers, turbines, generators, substations and transmission equipment has been subject to extended lead times, price increases, tariffs, national security requirements and allocation restrictions due to rising demand. And it warns of a specific risk: it might fail to secure firm gas supply and transportation during a facility's development, causing a planned site to be delayed or abandoned.

That risk has already materialized in another case cited: Oracle, which was denied a pipeline permit needed to power its Project Jupiter data center in New Mexico. In September, Oracle sent the site's developer, a unit of Blue Owl Capital, a force majeure notice allowing it to delay payments if it cannot obtain adequate power supply, which could let it defer rent payments for several years. Blue Owl told Bloomberg that the notice does not change the financial commitments of the multiyear project and that all parties are “fully aligned.”

From all this, West sees a favorable outlook for gas producers such as EQT, oil majors such as Chevron (which has already begun striking power supply deals with data center developers), midstream operators such as Williams and equipment makers such as GE Vernova and Baker Hughes.

As evidence for the thesis, GE Vernova reported that orders in its power segment rose 178% year over year, driven mainly by gas generation equipment such as turbines. Its backlog and capacity reservation agreements for gas equipment grew from 116 GW to 122 GW, with a forecast of 130 GW by year-end.

In short, the article argues that the limit on AI growth is no longer just access to chips, but the availability of gas, pipelines and generation equipment. It is worth remembering that the investment thesis comes from a Melius Research analyst and that the piece is a financial news report, not an independent technical analysis.