US gas power development jumped 50% in six months to feed AI data centers. The turbine shortage means your next data center job might include a power plant

US gas-fired power capacity in development jumped 50% in the first half of 2026 — from 252 gigawatts to 378 GW — almost entirely on the back of AI data center demand, according to a new Global Energy Monitor tracker update. Roughly two-thirds of those projects don't have a turbine manufacturer signed yet, which is why a growing share of data center developers are building their own power plant on-site instead of waiting.

How big is the buildout, and how much of it is data centers?

Global Energy Monitor, which maintains the most widely cited independent database of power plants worldwide, found that gas capacity earmarked to connect directly to US data centers nearly doubled in six months — from 97 GW to 189 GW. That's more than half the total national gas pipeline now tied to AI infrastructure specifically. Texas leads by a wide margin: 122 GW of gas capacity in development statewide, up 51% since January, with 77 GW of that — nearly two-thirds — planned to power data centers directly. The surge put the US ahead of China for the first time in years, with roughly double China's gas capacity in development and close to a quarter of the world's total.

Why is a turbine, not a permit, the thing that stalls the project?

Normally the long pole in a gas plant's schedule is environmental review and interconnection. Right now it's equipment. GE Vernova, Siemens Energy, and Mitsubishi Power control most of the large-frame gas turbine market, and their combined order books are booked out toward 2030 — with some large-frame units reported sold out through 2028. Global Energy Monitor's data shows about two-thirds of US gas projects in development have no turbine manufacturer under contract at all, and for the subset tied specifically to data centers, more than half are still unassigned. A developer can clear permitting, finish site work, and pour foundations, and the project still sits idle waiting on hardware that hasn't been built yet — the same equipment-over-construction schedule risk Construction AI Brief flagged in Nvidia's own chip backlog last week, now showing up one layer down the supply chain in the power plants those chips run on.

What's the workaround, and what risk comes with it?

Developers racing to power data centers before their turbine order ships are increasingly turning to reciprocating gas engines — smaller units, closer to what you'd find on a standby generator, deployed in banks of dozens instead of a handful of large turbines. xAI used this approach at its first Colossus site in Memphis, reportedly reaching power-on in about 122 days using modular units. But speed cuts both ways on compliance. xAI's larger, turbine-based buildout in Southaven, Mississippi drew a lawsuit from the Southern Environmental Law Center alleging the company operated dozens of gas turbines without the air permits state law requires — a reminder that skipping the queue on equipment doesn't skip the queue on environmental review, even when the owner treats the schedule as the priority.

What should an estimator actually check before pricing a data center job?

• Ask whether the project includes behind-the-meter generation at all. A data center scope that used to end at the utility meter now sometimes includes the power plant feeding it — turbine or reciprocating engine skids, fuel gas supply piping, exhaust and emissions controls, noise mitigation, and the civil work under all of it.
• Find out who holds the air permit, and when it's due. On a fast-tracked job, permitting can lag construction instead of leading it. That's the owner's legal exposure, but it's the GC's schedule risk if the plant can't be commissioned without it.
• Price the equipment lead time separately from the construction schedule. If the owner is waiting on a turbine with a multi-year backlog, that date — not the GC's construction sequence — sets when the facility actually energizes. Say so in the contract rather than letting an equipment delay read as a construction delay.
• Treat reciprocating-engine jobs as a different scope than turbine jobs, not a smaller version of the same one. More units, more modular foundations, more fuel gas connections, and a compressed schedule that leaves less room to catch a missed permit before startup.

The equipment shortage forcing developers to build their own gas plants on-site is turning a scope that used to belong entirely to a utility into one that shows up in a GC's or electrical sub's bid package — fuel piping, emissions gear, and permitting risk included.

The next line item in your data center bid might not be a building system at all — it might be a power plant.

Construction AI Brief publishes three times a week. Subscribe at constructionaibrief.com.