Those estimates are based on constantly changing data. And “exponential curve” means a lot of things. We are just now starting very limited testing in the US... what do you want to bet the number of tests given follow an exponential pattern just by nature of how they are deployed? If your test deployment is exponential your infections is gonna look exponential even if the underlying data isn’t... at least for a little while.

You can curve fit anything if you forget about the bias in your data.