A $1.25 billion supply deal that dies before a single unit gets delivered was never a supply deal — it was a press release with a dollar sign attached.
That’s the blunt read on Crusoe walking away from its agreement to buy 29 natural gas-fired Superpower turbines from Boom Supersonic. Both companies are based in Denver. Both are in the business of promising large things. One of them just decided the other’s large thing wasn’t worth waiting for, and Boom lost the only customer it had publicly named for its power business.
I review AI tools and agents for a living, which means I spend most of my time separating demos from products. The pattern here is familiar enough that I recognized it instantly. A capability gets announced. A number gets attached. Coverage follows. Then, months later, the thing quietly doesn’t happen and nobody circles back.
Why the number was always the weakest part
$1.25 billion sounds like commitment. It isn’t. It’s a ceiling on a purchase order that hadn’t been executed. No deliveries occurred. That means the figure represented intent, not spend, and intent is the cheapest thing in AI infrastructure right now.
Crusoe recently raised $3.9 billion. When you’re sitting on that kind of capital, a billion-dollar hardware order is a line item you can restructure without a board fight. The asymmetry matters: for Crusoe, this was one power-sourcing option among several. For Boom, it was the launch customer for an entire product line. Same contract, wildly different stakes.
That asymmetry is the part readers should sit with. When you see a partnership announcement between a well-funded buyer and a company trying to prove a new product category exists, the announcement is worth far more to one side than the other. The smaller side needs the validation. The bigger side is buying optionality.
What this says about AI data center power
The reason a supersonic jet company was selling stationary gas turbines at all is that AI data centers need power faster than the grid can supply it. That demand has created a market where anyone with turbine engineering experience can plausibly pitch a power product. Boom did exactly that. Crusoe, which builds AI data centers, was a logical first buyer.
Logical isn’t the same as durable. The specifics of why Crusoe backed out aren’t in the public record I’ve seen, and I’m not going to invent a reason. What I can say is that a buyer with $3.9 billion in fresh capital reversing a power-sourcing decision before delivery tells you the decision wasn’t load-bearing to begin with. You don’t cancel the thing your business depends on. You cancel the thing you were testing.
The pattern to watch for
If you follow AI infrastructure news, this cancellation is a useful calibration exercise. Here’s what separates a real commitment from a announcement:
- Deliveries, not orders. Units in the field beat units on paper every time. This deal had zero of the former.
- Who needs it more. If one party’s stock narrative depends on the partnership and the other’s doesn’t, the deal is fragile.
- Whether the buyer has alternatives. Crusoe had options. That’s exactly why it could leave.
- Whether the seller has other named customers. Boom had one. Now it has none disclosed.
None of this makes Boom’s turbine product bad. I haven’t seen one run, and neither has Crusoe. What it does is reset the product back to pre-validation status. A launch customer is a form of proof, and that proof has been withdrawn.
Why I care about a turbine story on an AI tools site
Because the same credibility problem runs through the entire AI stack. Agent frameworks announce enterprise partnerships that never ship. Model providers announce capacity deals that get restructured. Tooling startups announce integrations with companies that barely know they exist. The vocabulary changes; the mechanics don’t.
When I test an AI agent, the question I ask is whether it does the thing when nobody’s watching. A $1.25 billion turbine order that evaporated before delivery is the infrastructure-scale version of a demo that only works on the presenter’s laptop. The number was real. The commitment was conditional. Those aren’t the same, and the press cycle treats them as if they are.
Crusoe gets to keep building data centers with whatever power solution it actually chose. Boom gets to find a new first customer and explain the gap. And anyone who covered the original deal as settled fact gets a free lesson in reading purchase agreements as forecasts.
Announcements are marketing. Deliveries are evidence. Price them accordingly.
🕒 Published: