The timing and mechanism of formation of the Tibet Plateau remain elusive, and even the present-day structure of the Tibetan lithosphere is hardly resolved, due to conflicting interpretations of the geophysical data. We show here that significant advances in our understanding of this orogeny could be achieved through a better assessment of the composition and rheological properties of the deepest parts of the Tibetan lithosphere, leading in particular to a reinterpretation of the global tomographic cross sections. We report mantle phlogopite xenocrysts and carbonate-bearing ultramafic cumulates preserved in Eocene potassic rocks from the Eastern Qiangtang terrane, which provide evidence that the lithospheric mantle in Central Tibet was enriched in H 2 O and CO 2 prior to the India-Asia collision. Rheological calculations and numerical modeling suggest that (1) such metasomatized mantle would have been significantly weaker than a normal mantle but buoyant enough to prevent its sinking into the deep mantle; (2) the slow seismic anomalies beneath Central Tibet may image a weakened lithosphere of normal thickness rather than a lithosphere thinned and heated by the convective removal of its lower part; and (3) melting of such soft and fusible metasomatized mantle would have been possible during intracontinental subduction, supporting a subduction origin for the studied Eocene potassic magmatism. These results demonstrate that the inheritance a soft and buoyant precollisional Tibetan lithosphere may have conditioned the growth and the present-day structure of the Tibet Plateau. Plain Language Summary The Tibetan Plateau is the largest relief on the Earth's surface, but also one of the least understood geodynamic phenomena, due to the lack of constrain on its growth evolution and mechanism. Here we have discovered, enclosed in Eocene (35-Ma-old) lavas from Central Tibet, hydrous and carbonated minerals derived from the underlying upper mantle, which reveal an unsuspected abundance of water and CO 2 in this upper mantle. We show that, consequently, the upper mantle beneath Central Tibet has been anomalously weak since before the collision of India with Asia. It was previously thought to have disappeared, foundered in a process called "delamination," but we demonstrate here that this was an artifact: Most of the tibetan mantle is still there, but it is very weak and therefore does not appear on the seismic images. This implies that models expecting the Tibet Plateau to be a result of a delam-ination process are probably wrong; instead, we propose a model in which the weak upper mantle beneath our study area was easily indented by the strong adjacent continental block; this model manages to explain both Eocene magmatism and uplift in our study area.