Bitcoin does not derive its value from mining hashpower. Bitcoins value incentivizes and indeed, pays for hashpower. As codewench says, if all the ASIC miners started churning out invalid blocks, Bitcoin could proceed with an old Celeronbut of course that would not happen, because miners want to make money.
This has always struck me as possibly the biggest realistic attack vector for the network, the lack of emergency difficulty adjustment. The Celeron would work great, provided you get through 2016 or so blocks, but depending on how much hashpower you've lost, that could take a prohibitively long time? Is the solution here pretty much "we can hardfork if the time ever comes"