Easy to do, easy to implement but hard to bypass.
Also it tells me something about the network that is not vying for a slice of the attention economy and isn't going to do everything it can to keep me on the site.
This comment got me thinking that it might be worth using their second-to-last location to try and derive some vector. Obviously that's super informative as you already know the edge of the map they left, but maybe it's useful?
It's a good question, but in a closed system (like you have in space) the heat from the turbine loop has to go somewhere in order to make it useful. Let's say you have a coolant loop for the gpus (maybe glycol). You take the hot glycol, run it through your heat exchanger and heat up your cool, pressurized ammonia. The ammonia gets hot (and now the glycol is cool, send it back). You then take the ammonia and send it through the turbine and it evaporates as it expands and loses pressure to spin the turbine. But now what? You have warm, vaporized, low pressure ammonia, and now you need to cool it down to start over. Once it's cool you can pressurize it again so you can heat it up to use again, but you have to cool it, and that's the crux of the issue.
The problem is essentially that everything you do releases waste heat, so you either reject it, or everything continues to heat up until something breaks. Developing useful work from that heat only helps if it helps reject it, but it's more efficient to reject it immediately.
A better, more direct way to think about this might be to look at the Seebeck effect. If you have a giant radiator, you could put a Peltier module between it and you GPU cooling loop and generate a little electricity, but that would necessarily also create some waste heat, so you're better off cooling the GPU directly.
So, in doing a bit of research from a link in one of the other comments, this is lcos, levelized cost of storage. I understand that to be roughly equivalent to the marginal cost of using it, including the capex divided over the unit volume. That same article uses $125/kwh as the capex, which is in line with your (and my) expectations of the cost to install.
$65/mwh works out to $0.065/kwh, so that makes sense. Effectively you can read this as "it costs $65/mwh to store and then consume electricity using these batteries"
Oh man, I've been playing with GCP's vertex AI endpoints, and this is so representative of my experience. It's actually bananas how difficult it is, even compared to other GCP endpoints
If it matters to you, Reolink is a Chinese owned company. Not passing judgement one way or another, but if avoiding Unifi over the remote incident matters, I could see this factoring in as well.
Easy to do, easy to implement but hard to bypass. Also it tells me something about the network that is not vying for a slice of the attention economy and isn't going to do everything it can to keep me on the site.