What powers the AI agent you asked a question to this morning? If you guessed “electricity,” you’re technically right and practically missing the entire story. The more interesting answer is that somebody, somewhere, had to decide where that electricity comes from, and those decisions are getting strange enough to make headlines.
Here’s one that just landed. Crusoe, a Denver-based startup that builds AI data centers, walked away from a $1.25 billion agreement to buy 29 stationary power turbines from Boom Supersonic. Yes, that Boom Supersonic, the company known for trying to bring back faster-than-sound passenger flight. The turbines in question were natural-gas-fired units branded Superpower, rated at 42 megawatts each. Crusoe, which recently raised $3.9 billion, decided it would rather keep its options open with a flexible mix of energy sources. Boom, in the process, lost the first customer it had publicly named for its power business.
Why a jet company was selling power plants at all
This is the part that sounds like a mix-up but isn’t. A jet engine and a stationary gas turbine are cousins. Both spin a shaft by burning fuel and pushing hot gas through blades. One pushes an aircraft forward; the other spins a generator. Companies that know how to build the first thing can often build the second thing, which is why “aerospace firm pivots into power generation” is a real business move rather than a typo.
And the timing made sense on paper. AI data centers are unusually hungry customers. Training and running large models means racks of chips drawing power continuously, and the grid connections those facilities need can take years to secure. So operators started looking for ways to make their own electricity on site. Gas turbines are an obvious fit: they’re compact, they turn on relatively fast, and they don’t wait on a utility queue.
What this tells us about how AI actually gets built
If you spend time reading about AI agents, you get used to a certain vocabulary. Models, prompts, tools, memory, reasoning. It feels like software all the way down. Stories like this one are a reminder that every agent you interact with sits on top of a physical supply chain that includes concrete, cooling systems, transformers, and, apparently, turbine deals worth more than a billion dollars.
That matters for non-technical readers in a few practical ways:
- Capacity is a bottleneck, not an afterthought. When an AI company can’t ship a feature or raises prices on usage, the reason may have less to do with the model and more to do with how much computing power it can actually plug in.
- Energy choices are strategy choices. Crusoe’s shift toward a flexible mix suggests that locking into one supplier, one fuel, and one hardware line is a risk when demand forecasts keep moving.
- The AI boom pulls in unexpected companies. A supersonic aircraft startup pitching power plants to a data center builder is not a story anyone would have written five years ago.
Reading the cancellation without overreading it
I want to be careful here, because canceled deals invite dramatic interpretation. We know what was reported: the agreement covered 29 turbines, the value was $1.25 billion, Crusoe ended it, and the stated direction is a more flexible energy approach. We don’t have detailed reasoning from either company beyond that, and I’m not going to invent it.
What I will say is that flexibility is a reasonable instinct when you’re buying infrastructure for a market this young. Committing a billion-plus dollars to a specific turbine line from a company whose main business is aircraft means betting on delivery timelines, servicing, and fuel economics all holding steady. Keeping a mix of sources means you can adjust as prices, regulations, and your own customer demand change. For a company that just raised billions, preserving optionality may simply be worth more than the certainty of a single large order.
For Boom, losing a named launch customer is a genuine setback for the power side of the business. A first customer does more than generate revenue; it signals to everyone else that the product is real and someone credible is willing to stake money on it.
The takeaway for the rest of us
Next time you watch an AI agent book something, summarize something, or write code for you, it’s worth picturing the less glamorous layer underneath: a building somewhere, a cooling loop, and a very long conversation about where the next few hundred megawatts are coming from. Those conversations don’t always end in a signed deal. This one ended in a canceled one.
The AI industry likes to talk about intelligence. Increasingly, its hardest problems are about electricity. That’s not a contradiction, it’s just what scaling looks like when the software gets big enough to have a physical footprint.
🕒 Published: