[anonymousDan]

Ha, that's a great point. I guess if you have some latency sensitive inference workload the capacity will effectively be dynamic, but that is likely uncorrelated with local energy prices I imagine.