Since the Mesozoic, Africa has been under extension with shorter periods of compression associated with obduction of ophiolites on its northern margin. Less frequent than “normal” subduction, obduction is a first order process that remains enigmatic. The closure of the Neo-Tethys Ocean, by the Upper Cretaceous, is characterized by a major obduction event, from the Mediterranean region to the Himalayas, best represented around the Arabian Plate, from Cyprus to Oman. These ophiolites were all emplaced in a short time window in the Late Cretaceous, from ∼100 to 75 Ma, on the northern margin of Africa, in a context of compression over large parts of Africa and Europe, across the convergence zone. The scale of this process requires an explanation at the scale of several thousands of kilometres along strike, thus probably involving a large part of the convecting mantle. We suggest that alternating extension and compression in Africa could be explained by switching convection regimes. The extensional situation would correspond to steady-state whole-mantle convection, Africa being carried northward by a large-scale conveyor belt, while compression and obduction would occur when the African slab penetrates the upper–lower mantle transition zone and the African plate accelerates due to increasing plume activity, until full penetration of the Tethys slab in the lower mantle across the 660 km transition zone during a 25 Myr long period. The long-term geological archives on which such scenarios are founded can provide independent time constraints for testing numerical models of mantle convection and slab–plume interactions.