Alternative way to reach the result, that is already proposed in the other answers:
Consider the non-linear maximization problem $$\max_{x_i}\sum_{i=1}^{n}\frac{x_i^2}{(x_i^2+1)^2}$$ subject to $x_1+x_2+\ldots+x_n=\sqrt{n}$ and $x_i\ge0$. In that case the Lagrange function is defined as $$Λ(x, λ)=\sum_{i=1}^{n}\frac{x_i^2}{(x_i^2+1)^2}+λ\left(\sum_{i=1}^{n}x_i-\sqrt{n}\right)$$ The equation $\nablaΛ=0$ yields $$λ=\frac{2x_i(1-x_i^2)}{(1+x_i^2)^3}$$ for all $1\le x_i \le n$. Obviously the point $x^0$ with $$x^0_i=\begin{cases}1, & 1\le i \le \lfloor \sqrt{n}\rfloor \\ \\0, & \lfloor \sqrt{n}\rfloor \le i \le n \end{cases}, \qquad λ=0$$ solves the above system (both sides are equal to zero) and is thus a stationary point. The value of the objective function in $x^0$ is equal to $$\lfloor \sqrt{n}\rfloor\frac14$$
It is easy to see that for $n \ge 16$ this values is greater than $1$ and thus larger than the RHS of the given inequality. For $n<16$ this point does not yield a value larger than the RHS.
Perhaps there are other stationary points that yield a higher value for the objective function, even for $n<16$.