You can compute a loop by essentially building a CPU out of gates. But you're right that this would be impractically slow even with this latest improvement in performance.
It's not clear to me that this is still "impractically slow" for all purposes. An 8080 is a few thousand gates. What can you do on an 8080 in 100 cycles? You can now do that in about an hour on secret data by naively simulating the chip. That actually sounds like it might make some things practical that weren't before...
Nor I, TBH, but I've not been thinking about it very hard or very long. It just surprised me that we're no longer talking about months or years, where things are more obviously impractical for just about everything.