Think of moving volatility in the other direction.
As volatility approaches zero, any call strike strictly smaller than the ATM strike, $K<K_{ATM}$, will have zero probability of ending in the money, and the corresponding option value will be zero. An infinitesimally small change in stock price will not move $K$ past $K_{ATM}$, so the option value remains zero nearby. Thus, the sensitivity is zero.
Similarly for $K>K_{ATM}$, all options end up in the money, so the $\Gamma$ is also zero (though for the ITM options $\Delta=1$ rather than 0, ignoring interest and dividend rates).
Only strikes very close to $ATM$ have any likelihood of changing between $\Delta=0$ and $\Delta=1$.
Now, note that that $\Gamma = \frac{\partial}{\partial S}\Delta$ for any volatility. Furthermore, for a call $\Delta(S) \rightarrow 1$ as $ S \rightarrow \infty$.
Thus
$$
\int_0^\infty \Gamma(S) dS = 1
$$
That is to say, the area under the gamma curve is always 1.
In high-volatility cases, the $\Gamma$ is "spread out" over a wide range of $S$, so it never gets very big yet adds up to 1. When volatility is low, the $\Gamma$ is all concentrated near $K_{ATM}$ so it has to get very big.
We conclude that as volatility increases, $\Gamma$ decreases near $K_{ATM}$ and increases for other strikes.
(This is a more formal version of SolitonK's answer, which I have upvoted)