A growing number of AI data centers are using retired passenger-jet engines, converted to burn natural gas, as a source of power. According to the New York Times, as covered by TechCrunch, xAI has installed eight converted gas turbines in Mississippi. These generators are reportedly less efficient and emit more than gas turbines designed purely for power generation, yet demand for them keeps growing. The biggest reason is short lead times. With Mitsubishi Power quoting around 2031 for new orders, aeroderivative turbines that can secure power several years sooner have become a leading option for speeding up AI data center construction.
ProEnergy wins a 650 MW order, and xAI deploys eight converted turbines
On April 16, 2026, US power equipment maker ProEnergy announced that it had received an order from AI infrastructure company Crusoe for 13 PE6000 aeroderivative gas turbine generators. Each unit produces 50 MW, for a total of 650 MW, and delivery is scheduled by summer 2027. The installation sites are described only as "upcoming hyperscale data center projects," and no specific facilities have been named. TechCrunch discussed the contract in connection with a data center development in Abilene, Texas, but ProEnergy's announcement does not explicitly say the units are for Abilene. A Crusoe senior vice president described the procurement as a strategic move to quickly secure the power that large data centers need.
The PE6000 is a gas turbine built by refurbishing the core of a GE CF6-80C2 engine from a retired passenger aircraft and combining it with power-generation components. According to ProEnergy's product materials, its gross output under ISO standard conditions is 50.2 MW, and it can start up in about five minutes. It runs on natural gas and also supports liquid fuels and hydrogen blends.
Crusoe is not the only company deploying such equipment. According to the New York Times, xAI installed eight ProEnergy converted turbines in Mississippi to power its "Colossus 2" AI supercomputer. The report cites state permit records and satellite imagery as evidence of the installation. A Rystad Energy analyst also explained that converting aircraft engines makes it possible to bring a data center online in 24 months.
Meanwhile, in September 2026, Crusoe canceled a plan to purchase 29 42 MW turbines from Boom Supersonic for a total of $1.25 billion. Deliveries had originally been due to begin in 2027, and Crusoe said it "will continue to use turbines, but not Boom's." The supplier changed, but the policy of using gas turbines to secure data center power did not.
Summer 2027 versus around 2031: a gap of three to four years from major manufacturers
The main reason aeroderivative turbines are drawing attention is the difference in lead times compared with conventional power-generation gas turbines. The 13 units ProEnergy booked in April 2026 are scheduled for delivery by summer 2027, while Mitsubishi Power indicated in October of the same year that new orders would be delivered around 2031. That is a gap of roughly three to four years.
Akihiro Ondo, CEO of Mitsubishi Power Asia Pacific, told Bloomberg on October 27 that new orders would be delivered around 2031. Delivery slots for 2028 and 2029 are already almost entirely filled, he said. However, this outlook applies to new gas turbine orders in general and is not a like-for-like comparison with ProEnergy's 50 MW-class product in terms of model, output, or contract terms. Even so, it is clear that securing a new power-generation gas turbine takes a long time in today's market.
The difference in lead times stems from differences in manufacturing processes and supply chains. New gas turbines are constrained by the manufacturer's production capacity, so buyers must wait until a production slot opens up after ordering. The PE6000, by contrast, reuses the core of a retired aircraft engine and sources the necessary parts mainly from North American supply chains. ProEnergy says this approach reduces the risk of parts-procurement and manufacturing delays. The company's materials also cite a track record of delivering eight units totaling 400 MW to an AI company, with 12 months from order to delivery. However, the customer was not named, and the figure was published by the manufacturer itself, which should be kept in mind.
Delays in power equipment directly affect data center business plans. In October 2026, Microsoft reportedly withdrew as the anchor tenant of Nscale's data center in Loughton, UK, after the local utility said it could not supply sufficient power until the early-to-mid 2030s. Even if computing equipment is ready, a data center cannot operate without secured power.
Among Japanese companies, Mitsubishi Power, part of the Mitsubishi Heavy Industries group, sits at the center of this supply constraint. Ondo expects global new gas turbine orders in 2026 to reach 70–80 GW. Mitsubishi Heavy set a goal in 2025 of doubling production capacity within two years, yet delivery for new orders is still said to be around 2031. It will take time before investment in expanded production translates into shorter lead times.
Less efficient, yet still chosen: why aeroderivative turbines win
The weaknesses of aeroderivative turbines are efficiency and environmental impact. ProEnergy publishes a gross heat rate of 8,800 Btu/kWh (lower heating value, LHV) for the PE6000, which converts to an efficiency of about 38.8%. By comparison, GE Vernova's large 7HA.03 gas turbine is rated at 43.3% net efficiency (LHV) in simple cycle. In combined cycle, where the turbine's exhaust heat also drives a steam turbine, the 7HA series reaches 64% efficiency.
That is a gap of about 4.5 percentage points between simple-cycle units and about 25 points against combined cycle. Note, however, that the PE6000 figure is based on gross output and the 7HA.03 figure on net output, and both are manufacturer claims, so this is not a strictly equivalent comparison. ProEnergy cites nitrogen oxide (NOx) emissions of 2.5 ppm with a selective catalytic reduction (SCR) system, but this too is a manufacturer-provided figure. The New York Times also points out that aeroderivative turbines are less efficient and emit more than purpose-built power-generation machines, and that more units are needed to reach the same generating capacity.
Because of these drawbacks, the advantage of aeroderivative turbines may not last. According to TechCrunch, some experts believe concerns about efficiency and environmental impact will grow in importance over the next one to two years. And even if manufacturing and delivery are fast, lengthy environmental permitting could erase the benefit of early operation.
In fact, xAI's gas turbine installations have sparked regulatory conflict. On March 10, 2026, the Mississippi Department of Environmental Quality (MDEQ) issued a Clean Air Act permit for 41 gas turbines totaling 1.2 GW in Southaven. xAI describes this as a permanent construction permit, but environmental groups claim turbines were installed before the permit was granted, and several groups filed objections in April. The outcome of those challenges could not be confirmed.
Because each aeroderivative turbine has relatively small output, large amounts of capacity require many units. More units can mean more emission sources and more permitting issues to check. However, public reporting does not make clear how many of the 41 turbines covered by the Southaven permit are aeroderivative. Environmental permitting is also a constraint common to gas turbines in general, not just aeroderivatives.
Still, as long as a three-to-four-year lead-time gap with new gas turbines remains, aeroderivative units look advantageous for projects that can obtain the necessary permits. Bringing a data center online several years earlier has economic value that justifies accepting lower efficiency.
Will "bridge power" last longer? Lead times are the key to watch
In an October 23, 2025 article, XenoSpectrum pointed out that aeroderivative turbines were being used as "bridge power" for AI data centers and that this role could last longer than expected. At the time, ProEnergy said it could deliver new orders in 2027 and that customers expected to use the units for five to seven years. The Crusoe contract, booked in April 2026 with delivery by summer 2027, is broadly consistent with the lead time the manufacturer had indicated.
If new gas turbines do not arrive until around 2031, aeroderivative turbines delivered in summer 2027 would supply power for three to four years after that. The previously stated five-to-seven-year usage period likely includes continued use as backup power even after new equipment arrives or grid connection is completed. What began as a stopgap for temporary power shortages could become a permanent part of data centers' power mix.
Two main conditions could change this outlook. First, major manufacturers such as Mitsubishi Power and GE Vernova could expand production capacity and begin offering delivery earlier than 2029 for new orders placed in 2027. If new gas turbines can be procured quickly, the reason to choose less efficient aeroderivative units weakens.
Second, lead times for aeroderivative turbines themselves could lengthen. ProEnergy's 13 units for Crusoe have roughly 14 to 17 months from contract to delivery deadline. If future new contracts stretch beyond 24 months, it would suggest that supply constraints are starting to affect aeroderivative turbines too. The 24-month benchmark is a rule of thumb set here to judge a significant lengthening, based on this contract and the manufacturer's published 12-month past delivery record.
In addition, the outcome of the objections to xAI's permit will influence how far adoption expands. Even if equipment can be procured quickly, regulatory delays would erode the lead-time advantage.
What supports the competitiveness of aeroderivative turbines is not generation efficiency but the time value of bringing AI data centers online several years sooner. Demand is likely to persist as long as supply shortages at major manufacturers continue, but if lead times for new gas turbines shorten, equipment selection will likely turn back to generation efficiency, emissions, and long-term operating costs.