You absolutely could do that if you wanted to basically burn money.
Datacenters and especially ML training hardware is highly capital intensive and depreciates at basically constant rate regardless of utilization.
I see currently no scenario where you wwould be willing to idle this expensive infrastructure just to save pennies on the dollar on a grid connection; carbon credits would have to be nonsensically expensive for this to happen.
Datacenters and especially ML training hardware is highly capital intensive and depreciates at basically constant rate regardless of utilization.
I see currently no scenario where you wwould be willing to idle this expensive infrastructure just to save pennies on the dollar on a grid connection; carbon credits would have to be nonsensically expensive for this to happen.