In north-western Tibet (34.0textlessspan class="inline-formula"textgreater$^textrm∘$textless/spantextgreater N, 82.2textlessspan class="inline-formula"textgreater$^textrm∘$textless/spantextgreater E) near lake Aru Co, the entire ablation areas of two glaciers (Aru-1 and Aru-2) suddenly collapsed on 17 July and 21 September 2016. The masses transformed into ice avalanches with volumes of 68 and textlessspan class="inline-formula"textgreater83×10$^textrm6$textless/spantextgreater mtextlessspan class="inline-formula"textgreater$^textrm3$textless/spantextgreater and ran out up to 7 km in horizontal distance, killing nine people. The only similar event currently documented is the textlessspan class="inline-formula"textgreater130×10$^textrm6$textless/spantextgreater mtextlessspan class="inline-formula"textgreater$^textrm3$textless/spantextgreater Kolka Glacier rock and ice avalanche of 2002 (Caucasus Mountains). Using climatic reanalysis, remote sensing, and three-dimensional thermo-mechanical modelling, we reconstructed the Aru glaciers’ thermal regimes, thicknesses, velocities, basal shear stresses, and ice damage prior to the collapse in detail. Thereby, we highlight the potential of using emergence velocities to constrain basal friction in mountain glacier models. We show that the frictional change leading to the Aru collapses occurred in the temperate areas of the polythermal glaciers and is not related to a rapid thawing of cold-based ice. The two glaciers experienced a similar stress transfer from predominant basal drag towards predominant lateral shearing in the detachment areas and during the 5–6 years before the collapses. A high-friction patch is found under the Aru-2 glacier tongue, but not under the Aru-1 glacier. This difference led to disparate behaviour of both glaciers, making the development of the instability more visible for the Aru-1 glacier through enhanced crevassing and terminus advance over a longer period. In comparison, these signs were observable only over a few days to weeks (crevasses) or were absent (advance) for the Aru-2 glacier. Field investigations reveal that those two glaciers were underlain by soft, highly erodible, and fine-grained sedimentary lithologies. We propose that the specific bedrock lithology played a key role in the two Tibet and the Caucasus Mountains giant glacier collapses documented to date by producing low bed roughness and large amounts of till, rich in clay and silt with a low friction angle. The twin 2016 Aru collapses would thus have been driven by a failing basal substrate linked to increasing pore water pressure in the subglacial drainage system in response to increases in surface melting and rain during the 5–6 years preceding the collapse dates.