I'm not sure why, but I've never heard singularitarians consider that computation and intelligence is bounded by more hard constraints, like physics, than will let exponential growth continue indefinitely.
The "extrapolation of exponential growth = 'Singularity'" idea is a really crappy argument, and you shouldn't give much weight to it. For me it comes to two postulates: 1. AI is feasible. 2. When AI arrives on the scene, large changes are in store.
But I'll play my bets and say that NP-Hard will stay hard, the speed of light won't be broken, and at a certain point you need to switch to reversible computation because of landauer entropy, which will set space constraints of it's own.