Quantum computers can potentially achieve an exponential speedup versus classical computers on certain computational tasks, as was recently demonstrated in systems of superconductingqubits. However, these qubits have large footprints due to the need of ultra low-loss capacitors. The large electric field volume of \textit{quantum compatible} capacitors stems from their distributed nature. This hinders scaling by increasing parasitic coupling in circuit designs, degrading individual qubit addressability, and limiting the minimum achievable circuit area. Here, we report the use of van der Waals (vdW) materials to reduce the qubit area by a factor of >1000. These qubit structures combine parallel-plate capacitors comprising crystalline layers of superconducting niobium diselenide (NbSe2) and insulating hexagonal-boron nitride (hBN) with conventional aluminum-based Josephson junctions. We measure a vdW transmon T1 relaxation time of 1.06 μs, demonstrating that a highly-compact capacitor can reach a loss-tangent of <2.83×10−5. Our results demonstrate a promising path towards breaking the paradigm of requiring large geometric capacitors for long quantum coherence in superconducting qubits, and illustrate the broad utility of layered heterostructures in low-loss, high-coherence quantum devices.[/expand]