<p>Nanofluidics exhibits ultrafast transport, highly sensitive to confinement size and surface interactions. This transport behavior is particularly prominent in graphene nanoslits, but its molecular origin remains elusive. Here, we utilize machine learning-based molecular dynamics simulations with ab initio accuracy to probe water transport between graphene sheets. We find that interlayer distance (<i>H</i>)-dependent transport variations result from the competition between parallel transport and fluctuations perpendicular to sheet, a universal mechanism across various graphene surfaces, regardless of strain. At <i>H</i> ≤ 12.5 Å, large variations are dominated by the parallel transport of the interfacial layer, stemming from the distribution of severely undercoordinated “fast water”. At <i>H</i> &gt; 12.5 Å, variations become subtler as increased intermediate water leads to more perpendicular fluctuations. Moreover, we demonstrate that reducing interfacial friction at specific confinement sizes can further enhance nanofluidic transport several times over. Our findings suggest an effective strategy for regulating nanofluidic transport.</p>