There's the rub! It depends, of course, upon how you define [itex]e^x[/itex].
Many modern Calculus texts first define [itex]ln(x)= \int_a^x dx/x[/itex] and then define [itex]e^x[/itex] as the inverse function so your question never arises. That's my preferred method.
Robert1986 suggests that you define [itex]e^x[/itex] as as power series. That's perfectly reasonable, and again, your question does not arise.
But it is probably most common to define [itex]a^x[/itex] by first defining [itex]a^n[/itex] by "a multiplied by itself n times" then extending to all real numbers by requiring that [itex]a^{n+m}= (a^n)(a^m)[/itex], [itex](a^n)^m= a^{mn}[/itex], and continuity. Using that definition, we have
[tex]\frac{f(x+h)-f(x)}{h}= \frac{a^{x+h}- a^x}{h}= \frac{a^xa^h- a^x}{h}[/tex]
[tex]= \frac{a^x(a^h- 1)}{h}= a^x\frac{a^h- 1}{h}[/tex]
So that
[tex]\frac{da^x}{dx}= \lim_{h\to 0}\frac{a^{h+1}- a^x}{h}= a^x\lim_{h\to 0}\frac{a^h- 1}{h}[/tex]
is just [itex]C_aa^x[/itex] where [itex]C_a[/itex] is that limit.
It is easy to show that if 0< a< 1, then [itex]C_a[/itex] is negative, that [itex]C_2[/itex] is less than 1, that [itex]C_3[/itex] is greater than 1, and that [itex]C_a[/itex] continuously increases as a increases. That means that there must exist some value of a, between 2 and 3, such that [itex]C_a= 1[/itex] and we define "e" to be that value of a.
From [itex]\lim_{h\to 0}\frac{e^h- 1}{h}= 1[/itex] we can say that, for h close to 0, [itex]\frac{e^h- 1}{h}[/itex] is close to 1 so that [itex]e^h- 1[/itex] is close to h. Then [itex]e^h[/itex] is close to 1+ h and so e is close to [itex](1+ h)^{1/h}[/itex] which is common limit definition of e: [itex]e= \lim_{h\to 0}(1+ h)^{1/h}[/itex] or, taking n= 1/h, [itex]e= \lim_{n\to\infty} (1+ 1/n)^n[/itex].