
Key Takeaways
- Data center operators are converting used commercial jet engines into trailer-mounted, natural-gas generators capable of producing up to 48 MW each, sidestepping grid interconnection queues that now run roughly seven years.
- Used turbines cost up to 48% less than new units, with secondary-market pricing near $1.3 million per MW, while new power turbines take about six years to procure and are effectively sold out into 2030.
- Texas carries an estimated 474 GW of interconnection requests; an ERCOT audit could delay 49.8 GW of data center load and put up to $15 billion in projects at risk, with completion targeted for December 2026.
- xAI installed eight converted jet-engine power units at its Colossus 2 site near Memphis, and GE Vernova‘s gas turbine backlog climbed from 80 GW at the end of 2025 to 116 GW by mid-2026.
- Enverus analysts dispute the workaround’s scale, arguing the secondary market is too small to offset AI power delays, while environmental and noise opposition mounts.
The Lede
Data center operators are bolting repurposed commercial jet engines onto trailers and running them on natural gas to generate on-site power, up to 48 MW per unit, as grid interconnection queues stretch to roughly seven years and new power turbines remain sold out into 2030. The maneuver, once a niche engineering curiosity, has become a mainstream tactic for AI developers who cannot wait for the grid to catch up.
Key Facts
Grid interconnection queues now run approximately seven years, and procuring a new power turbine takes about six years, pushing availability into 2030 and beyond. Aero-derivative turbines are jet engines adapted to burn natural gas and drive generators; the CFM56, one of the most-produced jet engines in aviation history, is among the models finding a second life in this role.
Used turbines cost up to 48% less than new units, with secondary-market pricing reaching roughly $1.3 million per MW. Texas alone carries an estimated 474 GW of interconnection requests. State regulators paused new data center projects pending an audit that could delay 49.8 GW of load and cost projects up to $15 billion, about 20% of the US pipeline, with ERCOT targeting completion by December 2026.
GE Vernova‘s gas turbine backlog rose from 80 GW at end-2025 to 116 GW by mid-2026, stretching into 2029–2030, with turbine pricing up roughly 300%. ProEnergy repurposes Boeing 747-class engines into trailer-mounted power units, and FTAI Aviation struck a $1.5 billion deal for a new power unit. xAI installed eight converted jet-engine power units for Colossus 2 near Memphis, and a Mississippi permit board approved 41 Southaven gas turbines.
The Economics of Aviation’s Boneyard
The cross-industry arbitrage is straightforward: aviation’s retired engine stock is becoming AI’s power reserve. Engines that once pushed airliners across oceans are being pulled from boneyards, refurbished, and mounted on trailers to spin generators behind data center fences. The economics work because the two industries value the same hardware differently, an engine too worn for flight still has decades of stationary service potential.
The cost mechanics reinforce the appeal. Used turbines cost up to 48% less than new units, and secondary-market pricing reaches roughly $1.3 million per MW. Just as important, these engines sit inside a mature maintenance, repair, and overhaul ecosystem that can service them at scale, a support network that new turbine manufacturing cannot quickly replicate. Procurement timelines tell the story: six years for a new turbine versus months for a repurposed unit.
The manufacturing-capacity winners are already visible. GE Vernova’s backlog growth reflects surging demand, and the company signed a 29-unit aeroderivative deal with Crusoe. FTAI Aviation has been re-rated by investors and struck a tie-up with Palantir. ProEnergy runs a 747-class conversion line. One caveat for readers: the $1.3 million per MW and 48% figures trace to Reuters’ November 9 reporting and should be confirmed against the original before publication.
The Grid Queue, the ERCOT Audit, and the Time-to-Power Gap
The structural shift is that on-site generation lets operators skip the grid queue entirely, changing how AI infrastructure is sited. Instead of chasing substation capacity, developers now scout sites for gas access, permitting tolerance, and trailer space. The grid becomes a backup rather than a prerequisite.
Texas illustrates the bottleneck. The state faces an estimated 474 GW of interconnection requests, and regulators paused new data center projects pending an audit that could delay 49.8 GW of load and put up to $15 billion in projects at risk, roughly 20% of the US pipeline, with ERCOT targeting completion by December 2026.
Enverus analysts counter that the secondary market is too small to move the aggregate needle. In their framing, the binding constraint is a “time-to-power gap,” not raw capacity, a distinction that matters because repurposed engines can shorten the gap for individual campuses without solving it systemically.
The strategy is fracturing across players. Neoclouds are the most aggressive adopters: xAI’s eight units at Colossus 2 and Mississippi’s 41-turbine approval show the appetite. Startups are struggling: Boom Supersonic raised $300 million, saw a $1.25 billion Crusoe deal collapse, and faces questions over its airliner program. Vertical integration is the emerging tell, SpaceX reportedly casting its own gas turbine blades, a move claimed to cut power delays by 18 months, while China is directing domestic turbine makers toward overseas AI data center orders.
Pollution, Permits, and the Noise Flashpoint
Regulatory risk is now first-order. A ruling found that xAI was generating extra electricity illegally, the DOJ intervened in a gas turbine lawsuit, and permits allow 25% higher pollutant limits with carbon monoxide concerns. The EPA has signaled plans for weaker data center turbine limits.
The environmental justice dimension is sharp. Reuters reporting found pollution hits Black communities hardest, and The New York Times framed AI dominance ambitions as carrying major air pollution costs. Noise is a distinct local-opposition driver, the high-pitched turbine whine is separate from emissions concerns and has proven just as potent at planning hearings.
Industry coverage frames the shift as data centers “skipping the grid queue.” The open debate is whether repurposed engines are a durable fix or a marginal stopgap.
The Bottom Line
Power, not compute, now sets the AI scaling clock, and on-site generation bought in 2025–2026 confers a lead that capital alone cannot replicate given 18–36 months of lead time. Watch ERCOT’s December 2026 audit, whether the used-turbine market proves material or marginal, the EPA’s final rule on data center turbine emissions, Boom Supersonic’s fate as a bellwether for startup entry, and Chinese turbine makers as a geopolitical wildcard.


