Perhaps an example will help. Consider the standard $\mathbb{R}$. You may be aware that intervals of the form $(a,b)$ are open. But intervals of the form $[a,b]$ are not, these are closed. In particular singletons $\{a\}$ are not open.
So consider now open intervals $U_n=(-1/n, 1/n)$ for any $n\in\mathbb{N}$. You can then check that $\bigcap_{n=1}^\infty U_n=\{0\}$ which is not open.
Now of course you can invent a new definition such that an arbitrary intersection of open subsets is again open. And in fact we do consider such situations. This is also known as the Alexandrov topology. So $\mathbb{R}$ is not Alexandrov, but for example every topological space that is finite as a set is Alexandrov. But this situation is rather rare in practice and less useful then the classical definition of topology. For example $T_1$ Alexandrov space is automatically discrete. And 99% of spaces we encounter "in the wild" are non-discrete $T_1$.
We can invent definitions however we want. But the rule of thumb is: a definition is as it is, because it is useful.