A more careful analysis using a linear model showed that a constant growth rate model does not fit the data well, underpredicting the levy at both ends and overpredicting in the center:
Examining the residuals confirms this:
Adding a quadratic term to the model produces a better fit:
Because the graph is concave down, the rate of growth is slowing. This shows that the model needed to get to a $100 million dollar levy in 15 years does not fit the data, because that curve would be concave up.
The R code and output are posted here.