Crusoe Drops Boom Supersonic's Turbines From Its AI Power Plan

Crusoe Energy Systems has walked away from a $1.25 billion agreement to buy stationary power turbines from Boom Supersonic, according to TechCrunch, unwinding one of the more unusual deals to emerge from the scramble to power AI data centers.

Boom Supersonic CEO Blake Scholl confirmed the reversal, telling TechCrunch that the turbines are "no longer in Crusoe's near-term plans." The comment is notable mostly for what it does not say: neither company has offered a technical or financial explanation for scrapping a contract of that size, and Scholl's phrasing leaves room for the arrangement to resurface later without either side admitting the original plan failed.

The pairing made sense on paper. Boom is best known for developing the Overture supersonic jet, but the company has also been adapting jet-engine derived turbine technology into stationary generators, betting that the same aerodynamic and combustion engineering that lets an aircraft engine produce enormous thrust in a compact package could be repurposed to produce electricity fast, on-site, and without waiting on a utility interconnection queue. Crusoe, an AI cloud and data center operator that built its early business capturing stranded natural gas at oil wells to run bitcoin miners, has leaned harder than most of its peers into behind-the-meter generation as a way to get GPU clusters online before the grid can catch up.

That urgency is the real story here. Grid interconnection queues in Texas, Virginia, and other data center hot spots now routinely stretch past three years, according to filings tracked by grid operators including ERCOT and PJM, while AI operators are trying to bring gigawatt-scale campuses online in a fraction of that time. Crusoe's answer, most visibly at the Stargate campus it is building in Abilene, Texas with Oracle and OpenAI, has been to pair leased or purchased gas turbines with its data halls so compute can start running before permanent grid power arrives. The Boom deal, announced with fanfare earlier in 2025 as a bet on faster-to-deploy, jet-derived turbines, was meant to be part of that bridge strategy.

Abandoning it points to a supply and integration problem rather than a change of heart about the need for fast power. Traditional turbine manufacturers, chiefly GE Vernova, Siemens Energy, and Mitsubishi Power, have order backlogs stretching into the early 2030s for the large gas turbines data centers want, which is exactly the gap that opened room for a challenger like Boom to pitch aeroderivative units as a faster alternative. But aeroderivative turbines, while quick to install, are typically smaller and less efficient at continuous, heavy-duty base load than the industrial turbines built for utility-scale power plants, and integrating a first-of-its-kind commercial generator into a live data center power plan carries execution risk that a company earning most of its revenue from an unflown aircraft program may not have been positioned to absorb on Crusoe's timeline.

The collapse of the deal also lands at an awkward moment for Boom. The company has spent 2025 trying to convince investors it can be more than a single-product aerospace startup by diversifying into power generation, arguing that its engineering talent and supply chain could translate into a second business line while Overture works through certification and early flight testing. Losing a $1.25 billion anchor customer before delivering meaningful turbine volume undercuts that pitch, even if Scholl's careful wording avoids saying the technology failed.

For Crusoe, the reversal is a reminder that power procurement for AI infrastructure is still being solved deal by deal, with operators testing gas turbines, nuclear power purchase agreements, and now novel aeroderivative suppliers against the same brutal constraint: compute capacity sitting idle is a bigger cost than almost any generation option available. Crusoe has not said what replaces the Boom turbines in its power stack, and TechCrunch's report leaves that gap unresolved, which is itself a signal that the search for fast, reliable, on-site power for AI data centers has not settled on a winner.