The AI race has hit a problem with no quick fix: electricity. According to TechCrunch, citing a New York Times report, AI data centers are starting to generate their own power with old aircraft jet engines converted into generators. According to an analyst estimate, such units can get a data center running in about 24 months, compared with about seven years to connect 100 megawatts to the grid in Northern Virginia, according to the International Energy Agency.
One of the leading companies in the field is ProEnergy, which converts CF6 engines, used on Boeing 747s, into generators called PE6000. According to Datacenter Dynamics, each unit supplies up to 48 megawatts, and according to Gizmodo the company has already sold 21 such generators to data centers. Crusoe, OpenAI's partner on the Stargate project, also ordered 29 turbines that will supply a full gigawatt to its Texas site.
The solution is fast but not clean. The turbines burn natural gas and draw criticism from environmental groups and residents, like the NAACP lawsuit against Elon Musk's xAI over pollution from turbines powering its Memphis data center. There are also supply limits: the number of old engines is limited, and converting them is expensive.
The issue shows how AI is no longer just a story of software and chips, but of physical infrastructure: power, cooling and water. In Israel, where the power grid is relatively small, the question of where and how large AI data centers will be built is becoming one of the central questions in infrastructure planning for the coming years.
