The spins of growing super-massive black holes (BHs) dictate the radiative efficiencies of the accretion process, and hold evidence regarding their past growth. Theoretical models broadly predict two distinct evolutionary scenarios for the spins of the most massive BHs, suggesting that such systems may either "spin-down" to very low spins, or alternatively "spin-up" to the maximal allowed value. Our current ability to observationally test the relevance of these two scenarios is, however, highly limited. I will present new constraints on the radiative efficiencies (and therefore, BH spins) of a large sample of luminous unobscured active galactic nuclei at z∼1.5-3.5, powered by some of the most massive black holes known. The analysis relies on a large set of literature data with estimates of BH masses, bolometric luminosities and physical accretion rates, based on the virial approach and on simple scaling laws emerging from accretion disk models. Most of the extremely massive BHs in the sample (i.e., M_BH ≥ 3✕10^9 Msun) show very high BH spins (a*), with typical values well above a*~0.7. This strongly supports a "spin-up" scenario, which is driven by either prolonged accretion or a series of anisotropically oriented accretion episodes. Considering the fact that these extreme BHs require long-duration or continuous accretion to account for their high masses, it is argued that the most probable scenario for the super-massive black holes under study is that of an almost continuous sequence of randomly yet not isotropically oriented accretion episodes. I will also mention other recent studies which suggest that the most massive BHs in the Universe are spinning at high rates.