We propose to measure the stellar winds of a nearly complete sample of O stars in the SMC with evolutionary masses above 30 Msol. Recent works indicate that stars of such high masses remain largely chemically mixed, hot, and compact during their evolution. This strongly questions the accepted models for gravitational wave progenitors as well as for various types of core-collapse SNe. These models explicitly assume a scaling of stellar mass loss with metallicity, which is not yet verified empirically in the upper mass range. Stellar winds are best studied in the UV, but up to now UV spectra exist for only a handful among the most massive O stars in the SMC. Their analyses revealed serious problems with commonly adopted mass loss recipes, but the existing sample of UV spectra is highly insufficient. We will obtain excellent quality UV COS spectra of 18 O-type stars with evolutionary masses above 30 Msol, dramatically increasing (by a factor of 3) and largely completing the existing sample. The COS spectra will be analyzed using highly advanced non-LTE stellar atmosphere models. We will determine stellar wind mass loss rates, velocity fields, clumping properties, abundances, and rotation rates. On this basis, empirical recipes for the dependence of the stellar mass-loss rate on metallicity will be derived and compared to theoretical predictions. The analyses of spectral lines and spectral energy distribution will provide hints on binary status. As the result, each star will be well characterized. The comparison with single and binary star evolutionary models will disrupt the current understanding of massive star lives and deaths at low metallicities.