Picture a procurement doc on someone’s second monitor in Denver. Line item: 29 natural gas-fired Superpower turbines. Vendor: Boom Supersonic, the company better known for trying to make passenger jets go fast again. Total: $1.25 billion. Somebody had to sign that. Somebody had to build a power model around it, sketch a site plan around it, maybe tell an investor about it.
Now picture the same doc with a line through it. Crusoe, the AI data center builder that recently raised $3.9 billion, has ended the agreement. No turbines were ever delivered. Boom lost the first customer it had named for its power business, a company headquartered in the same city.
I review AI tools for a living, which means I spend most of my week separating what a product does today from what a slide deck says it will do next quarter. This deal is that same problem with a much bigger invoice attached.
Nothing was delivered, and that is the whole story
The detail I keep circling back to is not the dollar figure. It’s the sequence. An agreement existed, it was large enough to anchor an entire product line for the seller, and then it ended before a single unit arrived on a pad. The commitment was real in every way that generates a press release and unreal in the only way that generates electricity.
If you have ever evaluated an AI toolkit based on a roadmap, you already know this shape. The feature is “coming.” The integration is “in private beta.” The pricing tier that makes the math work is “rolling out.” None of it is lying, exactly. It’s just that announced capacity and available capacity are two different inventories, and only one of them shows up when you need it.
What I’d actually take from this
I am not going to pretend I know why Crusoe walked. The verified facts do not include a reason, and I am not in the business of inventing one. What I can talk about is the pattern, because it repeats at every scale of this industry, from billion-dollar power contracts down to the $29-a-month agent platform you are considering for your team.
- A named launch customer is a marketing asset, not a guarantee. Boom had one. Then it didn’t. If your evaluation of a vendor rests on “big company X signed with them,” understand that the signature and the shipment are separate events.
- Untested product lines carry untested risk. The turbines were a new line of stationary power plants from a company whose core reputation is in aviation. That is not disqualifying. Plenty of good products come from sideways moves. But it means there is no track record to check, and no track record means you are buying the pitch.
- Exit terms matter more than entry terms. The most useful thing in that canceled agreement was whatever clause let Crusoe leave cleanly before delivery. When you sign an annual contract for an AI tool, the escape hatch is the part worth reading twice.
The AI infrastructure crunch makes everyone a little credulous
Compute demand is pulling the entire industry toward commitments made on faith. When you need power and you need it on a timeline, a supplier with an unproven line and an aggressive quote starts to look reasonable. The same dynamic operates in software. When your team needs an agent framework that does six things, and one vendor claims all six while the boring incumbent does four of them well, the six-thing pitch wins the meeting.
It loses the quarter, usually. That’s my experience reviewing these things. The four-thing tool ships four things. The six-thing tool ships two and a changelog.
How I’d run the same evaluation
My standing rule when I test a toolkit: I only score what I can make work in a session. If a feature requires a waitlist, a sales call, or a “coming soon” badge, it scores zero. Not a low number. Zero. Because a zero is what you get if the roadmap slips, and roadmaps slip more often than they hold.
Applied to hardware, the equivalent is not scoring capacity until it’s generating. Applied to your stack, it means building your workflow around the features that exist right now, with a plan that survives the promised ones never arriving.
Two companies, one lesson each
Crusoe gets to keep its $3.9 billion and find another power source. That’s the advantage of walking away early, and it’s the same advantage you get from a month-to-month plan and a portable data format.
Boom has a harder problem. Losing your launch customer before delivery means the product line’s only external validation just evaporated, and every future buyer will ask about it. Reputational damage from a cancellation is not symmetrical, and it rarely lands on the party with the most cash.
The takeaway for anyone picking tools this month is unglamorous and free. Buy what works. Treat what’s promised as a bonus. And keep the door unlocked, because deals this size come apart, which tells you something about how much weight your $200-a-month subscription should carry.
đź•’ Published: