Yes, I watched Bram's talk in Berkeley two weeks ago. It indeed runs with much lower power costs. The cost is the power to do a hard-disk seek every long once in a while, rather than to compute sha256 over and over again.

It does incentivize a buildup of physical disk drives. And if it becomes successful, it'll likely spur specialized datacenters and hardware (think specialized tape drives) to try to outcompete in space, so it might end up just as centralized as bitcoin mining.

But the power costs will be much lower.

There also is a proof-of-time component, that does computation, but it is designed to not be parallelizable, which means that throwing additional power at the problem doesn't help you as much. Instead, imagine that each participant must spend a deterministic length of time to compute a solution, and the person closest to the internet backbone wins. But this proof-of-time component is also smaller subsidiary role to the proof-of-space.

So overall, it's a hybrid proof. Bram didn't go into enough detail explaining the actual proof-of-time or space in his talk, so I'm not sure if it all works, but if it does, then you can expect it to use less power.

Proof-of-Stake should use even LESS power, though.