Behind the Circe Cummins 2 gigawatt order in the West Texas Permian Basin is a structural shift in how data center campuses source power, and a set of unresolved questions about carbon, cost, and what happens when the grid catches up.
The grid is becoming the bottleneck for AI. Utility interconnection queues stretch years, transmission upgrades take longer, and the gap between a permitted megawatt and an energized one is now the binding constraint on data-center buildouts. That constraint is pushing the industry's largest campuses toward an answer the sector has long avoided: building their own power plants on-site.
This week's order of roughly 2 gigawatts of natural-gas generation from Cummins by Circe Energy, scheduled for delivery between 2026 and 2030, is one of the clearest signals yet that behind-the-meter gas, meaning customer-owned generation that bypasses the local utility, is moving from fringe idea to procurement reality for AI infrastructure. The equipment is destined for Circe's West Texas AI Infrastructure Campus, a 1,950-acre development platform in the Permian Basin designed for gigawatt-scale AI and high-performance computing deployment, with phased energization targeted to begin in 2027 (Circe Energy PR Newswire release).
The Permian siting is the economic core of the bet. Circe is anchoring the campus near abundant, competitively priced natural gas and proximate to North American energy production, framing that as long-term protection against grid power cost volatility. Dagan Baroco, the company's chief commercial officer, called AI infrastructure "fundamentally a power challenge" in the release, and positioned on-site generation as a hedge against utility interconnection delays, transmission congestion, and uncertain grid upgrade costs (Circe Energy PR Newswire release). The model is also framed as a response to a pattern Baroco describes as hyperscalers "securing land" or "securing a utility queue position" without the ability to actually energize capacity on customer timelines.
That framing sells the appeal, and the tradeoffs deserve equal billing. Behind-the-meter gas is fast to deploy, and depending on siting and structure it can sidestep parts of the utility interconnection process and some grid oversight. That is part of the appeal, and part of the controversy. It is also carbon-intensive, in a sector already under pressure to disclose and reduce emissions. The Permian gas-price thesis is asserted by Circe, not demonstrated, and gas-price volatility remains the single biggest unhedged variable in the long-term power economics claim.
A second caveat is in the announcement itself. The 2 GW figure is an order, not an energized capacity count. A press release is a platform announcement, not a commissioned-order receipt or a delivered-asset count, and prior AI-infrastructure orders have slipped. The procurement-to-energized-megawatts lag is itself a story, and a credible one given how often gigawatt-scale announcements have softened between announcement and energization.
The durable question is what comes after the bridge. On-site gas solves a deployment-speed problem that clean firm power, advanced nuclear, geothermal, and grid-scale storage have not yet solved at gigawatt scale and AI timelines. If those alternatives mature, the Permian bet looks transitional. If they do not, the industry's carbon bill rises with the megawatt count, and the regulatory fight over which generation sits behind the meter and which sits in front of it gets louder. The next time a company announces a multi-gigawatt energy deal, the questions worth asking are the same: who is the off-taker, what is the offtake structure, what replaces this as clean firm power matures, and does the announcement represent energized capacity or just an order?