Kiruna-type iron oxide-apatite (IOA) deposits are important sources for Fe, necessary for steel production, and other elements such as REE, crucial for new technologies. IOA deposits occur worldwide (Sweden, Chile, USA, China, Iran etc.) and range in age from Late Archean (2.5 Ga) to the present. However, their formation is still under debate. Hypotheses vary from a (magmatic-) hydrothermal origin to direct crystallization from an immiscible Fe-rich melt. In order to investigate which hypotheses works best, we measured trace element concentrations and Fe-isotope ratios in-situ in magnetites (Fe3O4) from the Cretaceous Los Colorados IOA deposit (~350 Mt Fe) in the Chilean Iron Belt. Analyses showed that magnetite cores have an igneous texture and chemistry, while the surrounding magnetite rims indicate lower temperature (magmatic-) hydrothermal formation conditions. Since a coactive cooperation between both processes could not be explained by one of the existing models, we developed a completely novel formation model for Kiruna-type IOA deposits. In our proposed scenario the decompression of an oxidized, andesitic and volatile-rich magma, typical for arc-volcanism, results in degassing of volatiles such as H2O and Cl. The exsolved fluid bubbles are expected to nucleate preferentially on surfaces of oxide crystals such as magnetite where surface tension is lower. The bulk density of these bubble-magnetite pairs is expected to be lower than the surrounding magma and will thus float upwards as a bubble-magnetite suspension that is additionally enriched in dissolved Fe due to complexation with Cl. This suspension will cause the formation of massive magnetite deposits in regional-scale transcurrent faults with magmatic-hydrothermal as well as with igneous characteristics. High temperature decompression experiments confirmed that the flotation model is physically possible and clearly showed upward accumulation of magnetite upon decompression and fluid exsolution in contrast to gravitational settling of these dense minerals expected without exsolved fluids. This flotation scenario is in agreement with the geochemical and isotopic signatures observed at Los Colorados and other Kiruna-type IOA deposits. Mineral flotation on exsolved fluid bubbles may also change classical views on crystal fractionation and thus the formation of monomineralic layers in mafic layered intrusions (e.g., Skaergaard, Bushveld complex), where dense magnetite layers overlie less dense anorthosite layers.