The deal is dead.
Crusoe, the Denver AI data center builder that recently closed a $3.9 billion raise, has walked away from its $1.25 billion agreement to buy 29 stationary natural-gas turbines from Boom Supersonic, another Denver company. Boom’s Superpower units were rated at 42 megawatts apiece. Do the arithmetic and you get somewhere around 1.2 gigawatts of planned generation that just evaporated from a spreadsheet. Crusoe says it’s moving to a flexible mix of energy sources instead. Boom, meanwhile, has lost the first and only customer it had publicly named for its power business.
I review AI tools for a living, which means I spend most of my time two or three abstraction layers above this story. But the reason your agent framework times out at 4pm on a Tuesday traces back to decisions exactly like this one. So it’s worth sitting with.
Announced capacity is not capacity
This is the part that keeps getting lost. Every week brings another press release about gigawatts under development, megawatts contracted, sites secured. Those numbers get repeated as though they’re already humming. They’re not. They’re intentions, and intentions get canceled.
A $1.25 billion purchase order is not a small intention. It’s the kind of number that anchors a company’s story about itself. And it still went away quietly enough that most people covering AI infrastructure didn’t blink. If a deal that large can dissolve without much noise, apply appropriate skepticism to every capacity figure you read in a vendor deck this quarter.
For anyone evaluating AI platforms, that’s the practical takeaway. When a provider tells you about the compute coming online next year, ask what’s actually built versus what’s ordered versus what’s been sketched on a napkin with a term sheet attached.
Crusoe made the boring, correct call
I don’t read this as a failure on Crusoe’s side. Committing 1.2 gigawatts to a single unproven product line from a single supplier is a concentration bet. If the delivery schedule slips, if the units underperform, if the supplier hits turbulence, you own that problem across your entire buildout. A mix of energy sources is less exciting and considerably harder to screw up.
There’s also the plain fact that Boom Supersonic’s core business is building a supersonic passenger jet. The stationary power line is an adjacent bet on shared engine technology. Adjacent bets can work. They also sit lower on the priority list than the thing the company was founded to do, which is a reasonable thing for a customer to worry about when they’re writing a ten-figure check.
Crusoe got a $3.9 billion raise done. Capital that fresh tends to come with pressure to show the plan is solid rather than clever. Swapping a concentrated supplier bet for a diversified one is what a company does when it wants to survive its own growth curve.
Boom’s harder problem
Losing your only named customer for a new product line is worse than losing revenue. It removes the proof point. Enterprise buyers for industrial power equipment want to see somebody else go first. Crusoe was going first. Now nobody is, at least not publicly.
That doesn’t mean the turbines are bad. It means Boom has to rebuild the commercial case from a colder start, while also building an airplane. Two very hard things at once, with the easier-to-explain one now short a reference customer.
What this means if you’re just trying to ship
The AI tooling space runs on an assumption that compute keeps getting cheaper and more available on a predictable curve. That assumption depends on electricity showing up, on schedule, at the sites where the racks are going. Stories like this one are a reminder that the supply chain underneath your API calls includes gas turbines, interconnect queues, and multi-year procurement decisions that can reverse.
A few things I’d actually do with this information:
- Treat provider capacity roadmaps as forecasts, not commitments, and plan for the version where the timeline slips a year
- Avoid architectures that assume one vendor’s region will always have headroom, because power constraints hit regionally before they hit globally
- Watch which infrastructure companies favor flexibility over splashy exclusive deals, since those are the ones less likely to strand you
Crusoe chose flexibility over a headline. That’s a mildly disappointing news cycle and a considerably better operating decision. The interesting question is how many other announced gigawatts across this industry are sitting in the same category, waiting for someone to quietly revise them downward.
🕒 Published: