Let us argue by induction. The case $n = 2$ has been done above.
Consider $n_1,...,n_m$ as above, now with $m \geq 3$. Without loss of generality, we may assume that $2 \leq n_1 \leq n_2 \leq ... \leq n_m$, since if any $n_i$ is $1$ then the problem is trivial. The hypothesis implies $(n_1, (n_2...n_m)) = 1$. Let $d = (n_2,...,n_m)$. Consider $n > n_1...n_m= M$.
First, note that there is some $d > k \geq 0$ such that $n - kn_1$ is a multiple of $d$. This is because the numbers $n,n-n_1,n-2n_1,...,n-(d-1)n_1$ are $d$ different numbers, which leave $d$ different remainders when divided by $d$, because the difference of any two of them is of the form $jn_1$ with $j < d$, hence is not a multiple of $d$. One of these must leave remainder $0$, since there are only $d$ many possible remainders. The claim follows. Note that $k < d$, so $kn_1 < dn_1 \leq M$, therefore $n-kn_1 > 0$ is a positive multiple of $d$. Let $n - kn_1 = ld$.
Suppose we can show that $l > \frac{n_2n_3...n_m}{d^{m-1}}$. Then, since $\left(\frac{n_2}d,...,\frac{n_m}d\right) = 1$, we get by the induction hypothesis that $l = \frac{\sum_{i=2}^m k_in_i}{d}$ for positive $k_i$, therefore $n = kn_1 + ld = kn_1 + \sum k_in_i$ is a non-negative combination of the $n_i$, hence we are done.
To show that the condition holds, we go in steps.
Note that $dl = n - kn_1$. Since $k <d$ we get $dl > n - dn_1$. Since $n > M$ we get $dl > M - dn_1$. Finally, since $d$ divides $n_2$ we have $d \leq n_2$ therefore $dl > M - n_2n_1$.
Note that $n_1 \geq \frac{n_3...n_m}{n_3...n_m - 1}$ , since the left hand side is smaller than $2$ (it is equal to $2$ when $n_3...n_m = 2$ , but use the condition on $n_i$, and the gcd condition to derive a contradiction. Hence $n_3...n_m > 2$)
Therefore, multiplying by $n_2$ on both sides , and noting that $M - n_2n_1 = n_2n_1(n_3...n_m - 1)$, we get $M - n_2n_1 > n_2n_3...n_m$. Of course, this implies $M - n_2n_1 > \frac{n_2n_3...n_m}{d^{m-2}}$.
Now the result follows by combining the bullet points.