AI's Endless Hunger for Energy: Retired Jet Engines Step In to Power Data Centers
Originally published in Turkish. Read the original →
With connections to the electricity grid taking years, tech companies and data center operators have started powering AI servers by converting passenger jet engines that have reached the end of their flying life into natural gas-fired power turbines.
Training generative AI models and running the inference processes that answer the instant queries of millions of users has triggered an infrastructure race of unprecedented scale worldwide. This competition, which initially took shape around chip supply and access to advanced GPU clusters, has now hit a far more basic and physical barrier: electrical power. Lining up billions of dollars' worth of AI accelerators on racks is no longer enough on its own, because the grid capacity needed to feed these giant data centers cannot keep up with the dizzying speed of the silicon world. To get past this bottleneck, tech giants and data center developers are turning to an unusual solution borrowed from aviation: retired jet engines that have reached the end of their flying life.
Grid Queues and Years of Waiting
The amount of energy a modern AI data center consumes now rivals the total electricity demand of a mid-sized city. But traditional power grids were not designed to handle such intense and sudden load increases. In regions where data center investment is concentrated, above all the United States, connecting a new facility to the high-voltage grid or building a conventional natural gas combined-cycle plant from scratch can take 5 to 7 years because of bureaucratic permits, transmission line shortages and equipment lead times.
Many of the industry's leading figures, including Microsoft CEO Satya Nadella, have said plainly in recent statements that the real problem is no longer finding GPUs but getting the energy needed to plug in and run the thousands of chips they already have. In an environment where even a delay of a few months in the AI race means losing billions of dollars in market share, tech companies have no patience to wait years for grid approval. That urgency is exactly what turned old passenger jet engines sitting in aircraft boneyards into some of the most sought-after equipment in the energy sector.
From the Sky to Data Centers: How Jet Engines Become Generators
Because safety standards in commercial aviation are extremely strict, the turbofan engines that serve on Boeing and Airbus passenger aircraft are retired from the sky after they reach a certain number of flight hours and pressure cycles. But the fact that an engine can no longer carry passengers at an altitude of 10,000 meters does not mean its mechanical life is completely over. Engineering firms convert these jet engines, whose basic working principle is to compress air, ignite it with fuel and produce a high-speed gas exhaust, into stationary electricity generators called aeroderivative gas turbines.
In this conversion process, the huge fan assembly at the front of the aircraft engine is removed, and the engine's core is reconfigured to run directly on natural gas instead of jet fuel. The enormous thrust that pushes the aircraft forward in the sky is used on the ground to spin a power turbine and the alternator connected to it. While ordering a giant industrial gas turbine from scratch takes years, modified jet engine units can be installed on site in about 24 months or less and supply electricity directly to the data center behind the meter, off the grid. What's more, just as an aircraft takes to the runway and reaches full power within minutes, these systems can come online within just a few minutes and produce energy at full capacity.
The Demand Boom, from Elon Musk's xAI Facilities to Stargate Projects
One of the first and most notable players to put old jet engines at the heart of data centers was Elon Musk. Musk's AI company xAI brought online mobile power units made of converted jet engine turbines in neighboring Mississippi to feed the second phase of the giant Colossus supercomputer cluster it built in Memphis to train its Grok models. Likewise, infrastructure companies such as Crusoe, which builds the multibillion-dollar data center campuses that giants including OpenAI are running in Texas, have placed large orders for these turbines to keep servers running until grid power arrives.
This unexpected demand boom has also created a brand-new market for companies operating at the intersection of aviation and energy:
* ProEnergy: It overhauls the legendary General Electric CF6-80C2 engines, which served for years on wide-body aircraft such as the Boeing 747 and 767, and turns them into gas turbines it calls the PE6000. A single unit can generate about 48 to 50 megawatts (MW) of electricity, which means a few engines side by side can carry the entire load of a giant data center on their own. * FTAI Aviation: It set up a dedicated power unit to convert CFM56 engines, one of the most-produced engines in commercial aviation history and the power behind the Boeing 737 and Airbus A320 families, into 25-megawatt electricity generators for data centers. Boom Supersonic: Even the startup developing a civilian supersonic passenger jet has begun signing new deals to sell the core of its own Symphony* jet engine design to AI data centers as a stationary power turbine.
The Price of Speed: Low Efficiency and the Carbon Emissions Debate
Although retired jet engines buy tech companies invaluable time, this approach comes with a serious environmental and economic bill. Aviation-derived simple-cycle gas turbines run at much lower fuel efficiency than a modern combined-cycle natural gas plant. In other words, they burn more natural gas to produce the same amount of electricity and release much higher amounts of carbon dioxide and nitrogen oxides (NOx) into the atmosphere per megawatt-hour.
Environmental organizations and local governments sharply criticize big tech companies that have made zero-emission pledges for setting up what are practically small fossil fuel power plants next to their data centers so as not to fall behind in the AI race. Experts stress that old jet engines cannot be a permanent solution to AI's energy crisis and are merely an expensive bridge technology used until nuclear energy investments, small modular reactors (SMRs) and advanced renewable energy grids come online. Still, the current picture lays bare that the limits of the AI revolution are no longer set by algorithms, but by the laws of physics and megawatts.
