Let $(G_t)_{t\in T}$ be a centred separable Gaussian process with canonical semimetric $d_G$. For $\varepsilon>0$, let $M(T,d_G,\varepsilon)$ be the largest cardinality of an $\varepsilon$-separated subset of $T$. There exists a universal constant $c>0$ such that for every $\varepsilon>0$, \begin{align*} \mathbb E\left[\sup_{t\in T}G_t\right] \ge c\varepsilon\sqrt{\log M(T,d_G,\varepsilon)}. \end{align*} Equivalently, the same conclusion may be written with covering numbers after changing the universal constant and replacing $\varepsilon$ by a comparable scale.