As stated in the comments, there is is considerable ambiguity here. We are given two values of the per unit cost function, let's call it $U(n)$. Logically, we could assume that $U(n)$ were linear but, at best, that could only be true in a range (as it is clearly a decreasing function of $n$, linearity in a large range would allow $U(n)<0$ which is not realistic). But it seems to me that that the only realistic assumption (that we can have enough information to solve for!) is that the total cost function is linear. That is to say, if we define $C(n)$ to be the total amount to be paid (thus $C(n)=U(n)\times n$) then $$C(n)=m\times n + b$$ for suitable constants $m,b$.
Let's see how this plays out. We are given that $$C(100)=18\times 100=1800\;\;\;\;\;\&\;\;\;\;\;C(500)=15.5\times 500=7750$$
A quick calculation to solve for $m,b$ then shows that $$C(n)=14.875n+312.5$$
As noted in the solution by @Clarinetist this lets you solve for $C(2000)=30062.5$ and the associated per unit cost of $15.03125$ .
It is interesting to note that, in the limit as $n\to \infty$ we get the marginal cost of a unit to be $$\lim_{n \to \infty}\;\frac {C(n)}n=14.875$$ Which feels fairly plausible in light of the given data.
$18each"? That's not how I would interpret the question. – Will R Sep 04 '16 at 23:5818$– srip Sep 05 '16 at 00:00