Crusoe kills $1.25bn Boom turbine deal for AI data centres
- Crusoe ended a $1.25 billion agreement to purchase 29 stationary power turbines from Boom Supersonic for its Abilene, Texas data centre campus.
- Boom's Superpower business was designed to cross-subsidise development of the company's Overture supersonic passenger jet.
- Crusoe's Abilene campus, serving Oracle and OpenAI, currently runs on the grid with gas turbine backup and intends to use alternative turbine suppliers.
- Boom indicated it still plans to deliver roughly 250 megawatts of Superpower units to other sites within the coming year.
Crusoe ended a $1.25 billion agreement to buy 29 stationary power turbines from Boom Supersonic, the TechCrunch report on the deal’s dissolution confirms. Blake Scholl, Boom’s chief executive, disclosed the termination on X, saying turbines are no longer part of Crusoe’s near-term primary power mix at its Abilene, Texas campus. The arrangement had been Boom’s launch customer deal for its Superpower business: 42-megawatt units that were to start arriving in 2027.
Crusoe operates a 1.2 gigawatt data centre at Abilene that serves Oracle and OpenAI, currently drawing power from the grid with gas turbine backup. A separate 900 megawatt facility under construction for Microsoft will use on-site gas turbines from other suppliers. A company spokesperson said energy plans have not fundamentally changed, adding that Crusoe still intends to use turbines for power, just not Boom’s.
The immediate question is what happens to Boom’s cross-subsidy model. Scholl had previously told TechCrunch that profits from stationary power were meant to fund development of the company’s Overture supersonic passenger jet, and Boom raised roughly $300 million largely to commercialise the Superpower line. Losing the anchor customer for that unit does not close the programme. Boom indicated it still plans to deliver around 250 megawatts of Superpower units to other sites within the coming year, targeting one gigawatt by 2028. But the gap between stated ambition and contracted reality matters when the plan depends on one customer underwriting the business case.
For the AI infrastructure sector, the episode highlights a recurring governance problem. Data centres are scaling power demand at extraordinary speed, and procurement decisions carry long-term consequences for grid stability, local energy planning and emissions. When a $1.25 billion arrangement unravels before hardware is delivered, it raises a practical question about how thoroughly these dependencies are stress-tested before announcements are made. The Abilene campus serves some of the most prominent AI models in commercial deployment; its power architecture is not an abstract planning exercise.
There is a quieter implication for anyone building or studying AI systems honestly. The energy cost of training and running large models is a factual constraint that shapes what is feasible and what is responsible. Deals like this one make that constraint visible, even when they fall apart. Whether the Superpower turbine eventually reaches Crusoe or not, the underlying demand for reliable, large-scale power at data centres is not going away.
The harder question, the one that survives this particular contract’s collapse, is whether the industry’s power planning can keep pace with its compute ambitions, or whether the announcements will continue to outrun the infrastructure.