The only thing that makes you favour the exponential is Occam's razor – it's simpler. Well, in this case, it's just too simple to fit even what we know.
I don't think that's really Occam's razor. It's not just about picking what's simplest, but on what makes the least amount of wild assumptions. I'd say a logarithmic curve makes less assumptions, and an exponential curve makes too many.
Regardless, even if the predictions of when it happens are on a slightly different curve, that doesn't rule that level of technology out. There are many more problems with the concept of a singularity, mostly with the correlation that processing power equals intelligence.