After XAi’s garnered some controversy for installing ‘quick and dirty’ gas turbines at its Memphir data center, OpenAI is making a daring move to guarantee steady power for its enormous AI infrastructure. According to reports, OpenAI, through its infrastructure partner Crusoe, has purchased 29 GE Vernova LM2500XPRESS gas turbines at its first Stargate data centre site in Abilene, Texas, to meet its high energy requirements. When fully deployed, the 34MW output from each turbine could result in a total output of 986MW. Almost 500,000 Nvidia GB200 NVL72 AI chips could run on that amount of energy.
Crusoe affirmed the purchase, emphasising that these turbines provide a versatile and effective power source for data centres that handle heavy workloads related to artificial intelligence. The jet engine technology that powers the GE LM2500XPRESS units is renowned for its quick reaction times. Their ability to swiftly ramp up or down power output allows for smooth power transitions while preserving grid stability. In order to address environmental concerns, the turbines are also outfitted with Selective Catalytic Reduction (SCR) systems, which lower nitrogen oxide emissions.
In the AI arms race, data centre energy consumption has emerged as a significant obstacle. The unexpected spike in demand is making it difficult for traditional power grids to keep up, which is stressing regional infrastructure and driving up prices. At his Memphis Supercluster, Elon Musk used portable power generators to solve a similar problem, but this caused local opposition due to pollution worries. For long-term data centre energy requirements, he also purchased a power plant abroad to be shipped to the United States.
Only 10 turbine permits are currently in place at OpenAI’s Stargate site, suggesting that the remaining 19 turbine permits are either pending regulatory approvals or may be for future locations. Although it hasn’t been verified, some people think OpenAI might follow Musk’s lead and underreport temporary generator deployments.
Tech companies are looking into long-term solutions like Small Modular Reactors (SMRs) as the race to power AI heats up, but nuclear options are still years away. To keep the lights—and GPUs—on for the time being, gas turbines and mobile power units provide a temporary fix.