AI data-center developers are hunting for used gas turbines and retired jet engines, because ordering brand-new power equipment could leave their servers waiting for electricity until the next decade. Some companies are buying refurbished machines, while others are converting aircraft-engine technology into generators that can run alongside their computing facilities. The scramble reflects a widening gap between AI companies' construction timelines and the power industry's delivery schedules.

New gas turbines now take an average of six years to deliver in the United States, according to energy consultancy Wood Mackenzie. Delivery discussions are already focused on 2030 and 2031, a timeline that clashes with an AI industry trying to bring enormous computing campuses online much sooner.

Turbine demand outstrips supply

Global turbine orders have topped 100 gigawatts, compared with manufacturers' estimated annual capacity of just 60 to 70 gigawatts, Reuters reported. Siemens Energy, GE Vernova, and Mitsubishi Heavy Industries have all said orders are outpacing what they can produce.

Gas turbines generate large amounts of electricity by burning natural gas to spin equipment connected to a generator. AI data centers need that power around the clock to run densely packed chips and cooling systems, and connecting new facilities to public electricity grids can itself take years.

The pressure is only expected to grow. The International Energy Agency expects global data-center electricity consumption to nearly double from 485 terawatt-hours in 2025 to 950 in 2030, while consumption at AI-focused facilities is projected to triple over the same period. Read together, the numbers frame the industry's core problem: the IEA's projection horizon is 2030, the same period into which new-turbine deliveries are currently being scheduled.

From airplane engines to AI power plants

One way around the backlog is an aeroderivative turbine, a power generator built using technology adapted from aircraft jet engines that can sidestep the multiyear wait for newly built machines. Other companies are going further by seeking older machines and retired aircraft engines that can be rebuilt for stationary power production.

For example, AI infrastructure firm Crusoe ordered 13 units from Missouri-based ProEnergy in April, each rated at 50 megawatts, for a combined 650 megawatts. The jet-engine-derived equipment is intended for future U.S. data centers, though the contract does not mean that Crusoe bought engines removed from retired aircraft.

Second-hand machines can also carry a substantial discount. Used turbines in good condition sell for about $1.5 million to $1.7 million per megawatt, compared with roughly $2 million to $2.5 million for new units, according to equipment brokers and energy analysts interviewed by Reuters. One broker described a sale of two smaller turbines totaling 29 megawatts for $12.5 million; equivalent new equipment would have cost about $35 million and taken five years to arrive, he said.

"That secondary market is very, very hot right now," EthosEnergy CEO Ana Amicarella told Reuters, adding that asking prices are climbing as more buyers compete for the same machinery.

The problem with second-hand power

Building a power plant beside a data center also brings fuel needs, air emissions, and permits into the equation. The Environmental Protection Agency says stationary turbines and engines used by data centers are subject to federal air-pollution standards. Setups like these are often called behind-the-meter generation, because the equipment feeds a single facility directly rather than the public grid.

Last week, OpenAI promised to cover electricity and grid costs at future campuses it owns and operates after voters backed tighter restrictions on data-center construction — a sign of how directly local politics now touch AI companies' power planning.

The gap between ambition and reality remains wide. Power-market research firm Cleanview identified about 90 gigawatts of proposed on-site generation at U.S. data centers in its mid-2026 analysis; only about 2 gigawatts was operating at the time. The roughly 88-gigawatt difference offers a concrete measure of how early in the buildout most of these projects still sit.

Developers are looking beyond turbines, too. Finnish manufacturer Wärtsilä told Reuters it had booked around 3 gigawatts of orders from U.S. data centers since spring 2025 for generators powered by gas engines rather than turbines.