Crusoe Walked Away From 1.2 Gigawatts of Boom Turbines
Crusoe canceled its $1.25 billion order for 29 Boom Supersonic Superpower turbines, and the split shows how fast the math on behind-the-meter power for AI data centers is moving.
The most unusual power deal in the AI buildout is over. On September 25, Boom Supersonic founder and CEO Blake Scholl posted on X that "Crusoe and Boom are no longer launching our Superpower turbine together." With that, Crusoe's $1.25 billion order for 29 of Boom's 42-megawatt natural-gas turbines, roughly 1.21 gigawatts of on-site generation, is gone.
Scholl's explanation was short. Turbines, he wrote, "are no longer part of Crusoe's near term primary power mix at Abilene/etc., so a launch partnership just didn't make sense," per TechCrunch. Crusoe's own framing was narrower. Spokesperson Andrew Schmitt told TechCrunch the company's energy plans "haven't changed" and that it will still use turbines, "just not Boom's."
Both statements can be true. Together they say a lot about how AI developers are now pricing the choice between the grid and a power plant in the parking lot.
What was actually on order
The deal was announced on December 9, 2025, alongside a $300 million Boom round led by Darsana Capital Partners, with Altimeter Capital, Ark Invest, Bessemer Venture Partners, Robinhood Ventures and Y Combinator participating, TechCrunch reported at the time. Crusoe was the launch customer for Superpower, a stationary turbine that shares about 80% of its parts with Symphony, the engine Boom is developing for its Overture supersonic airliner. First deliveries were scheduled for 2027.
The pitch was built for West Texas. Per Unite.AI, Boom says Superpower holds its full 42 megawatts at 110 degrees Fahrenheit and above, runs without water, burns natural gas with diesel backup and ships at roughly container scale. Scholl has argued that legacy turbines "lose 20 to 30 percent of their generation capacity at 110 degrees." For a data center that peaks on the hottest afternoon of the year, that derating is not a footnote. It is the difference between the capacity you bought and the capacity you have in August.
The price worked out to a bit over $1 million per megawatt of nameplate.
Crusoe's power map changed underneath the deal
Crusoe was founded in 2018 as a bitcoin miner that burned stranded natural gas at oil fields. It has since become one of the largest independent builders of AI campuses, and its power choices now look much more like a utility customer's than a wildcatter's.
Its flagship Abilene, Texas campus, the 1.2-gigawatt site that serves Oracle and OpenAI, runs primarily on grid electricity with gas-turbine backup, according to TechCrunch. A separate 900-megawatt Abilene facility being built for Microsoft does use on-site gas turbines. Crusoe says it picks power "right for each site," in Schmitt's words, and a single-vendor launch commitment to an unproven turbine sits awkwardly with that.
The portfolio has also been reshuffled. In June, Bloomberg reported that Crusoe paused its 1.8-gigawatt campus in Cheyenne, Wyoming, which it had been developing with Blackstone-backed Tallgrass and which was designed around natural gas. DatacenterDynamics quoted Crusoe saying the pause came "at the request of our customer." Black Hills Corp. has since said it will develop the site directly with an undisclosed large-load customer. A gas-heavy greenfield campus leaving the pipeline removes one obvious home for a gigawatt of new turbines.
Then came the money. On September 17, Crusoe closed a $3.9 billion Series F at a $30.9 billion valuation, led by Atreides Management, Mubadala Capital and Valor Equity Partners. Per Unite.AI, the company counts more than $140 billion in contracted value, more than 6 gigawatts of gross contracted capacity and about 1 gigawatt operational. A company with that balance sheet can buy proven turbines from established manufacturers and wait for grid interconnects. It does not need to underwrite a new product line to secure power.
The behind-the-meter bet is getting harder to size
On-site generation became fashionable because grid connections were too slow. That problem has not gone away. But building your own power has its own bottlenecks, and they are showing up in public.
The day before Scholl's post, Oracle sent a force majeure notice on Project Jupiter, the 2.45-gigawatt Stargate campus in New Mexico, TechCrunch and CNBC reported. That campus is meant to run on Bloom Energy fuel cells fed by natural gas, and the 17-mile Energy Transfer pipeline that would supply it has slipped from August 2026 to February 2027 after the New Mexico State Land Office repeatedly denied right-of-way permits. An air-quality permit for the fuel cells is still pending, with a state decision due by November 23.
The Jupiter episode makes the underlying point. Putting generation behind the meter swaps grid risk for fuel, permitting and equipment risk. The equipment market is tight too. GE Vernova's gas turbine backlog reached 116 gigawatts at the end of the second quarter, up from 100 gigawatts in the first, Utility Dive reported, and the company is already taking reservations for 2031 deliveries. CEO Scott Strazik said data center operators are now about 20% of its gas turbine customers.
A new entrant promising 2027 units into a market booked years out is a compelling offer on paper. The catch is that a buyer with a gigawatt of AI tenants on contract cannot easily absorb first-production risk on the machines keeping those tenants online.
What Boom keeps
Boom says it is moving on. Scholl said the company will deliver about 250 megawatts of Superpower units to other sites next year and is targeting 1 gigawatt in 2028. Unite.AI reports a longer-term goal of more than 4 gigawatts a year by 2030. Boom has not named the customers behind the 2027 volume.
Losing the launch customer still matters. A reference buyer at Crusoe's scale was proof that a hyperscale-class developer would trust the product. Without it, Boom has to sell 42-megawatt blocks one site at a time, most likely to smaller developers, industrial users and operators who cannot get into GE Vernova's queue.
Why it matters
The AI power story in 2025 was simple: the grid is too slow, so build your own. The 2026 version is more selective. Developers with capital and leverage, like Crusoe after its Series F, are mixing grid supply, proven turbines and backup rather than betting whole campuses on one on-site technology. The newest power suppliers are being pushed toward buyers who have no better option.
For anyone tracking AI infrastructure, the useful signal is not that turbines are out. Crusoe says they are not. It is that power is now bought like any other critical component: from multiple vendors, site by site, with a bias toward what has already run at scale. A 1.21-gigawatt order disappearing in a single X post shows how quickly that preference can move.
