> Solar is very, very cheap and almost totally worthless without storage.

For say an AI training-oriented data center, you could scale down the power usage when supply is limited. You could change power limits on the CPU/GPUs, put the machines in sleep mode or powered off entirely. So the required storage would just be a slightly bigger UPS.

Not sure if the economics works out, but at least technically it's possible as it's more flexible than user-based loads.

▲

loeg 2 days ago | parent | next [-]

You don't want to waste your GPU capex by not running those suckers at 100%. (Other datacenter workloads it makes some sense to demand-regulate, but not AI.)

▲

anonymousDan 3 days ago | parent | prev [-]

AI based training is an almost ideal match with this kind of supply. You could even imagine migrating long running training jobs to different parts of the world based on energy availability to optimise costs.

▲

idiotsecant 2 days ago | parent [-]

So the model is buy some of the worlds most expensive hardware and let it sit idle for half the time? If I want to save the same throughput I need to buy at least twice the hardware!

Load throttling is one of those ideas that seems great as long as someone else is doing it.

	▲	anonymousDan a day ago \| parent [-]
		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.