In the next 10–20 years, several observatories will aim to detect the signatures of oxygenic photosynthesis on exoplanets, though targets must be carefully selected. Most known potentially habitable exo-planets orbit cool M-dwarf stars, which have limited emission in the photosynthetically active region of the spectrum (PAR, 400<λ<700\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$400< \\lambda < 700$$\\end{document} nm) used by Earth’s oxygenic photoautotrophs. Still, recent experiments have shown that model cyanobacteria, algae, and non-vascular plants grow comfortably under simulated M-dwarf light, though vascular plants struggle. Here, we hypothesize that this is partly due to the different ways they harvest light, reflecting some general rule that determines how photosynthetic antenna structures may evolve under different stars. We construct a simple thermodynamic model of an oxygenic antenna-reaction centre supercomplex and determine the optimum structure, size and absorption spectrum under light from several star types. For the hotter G (e.g. the Sun) and K-stars, a small modular antenna is optimal and qualitatively resembles the PSII-LHCII supercomplex of higher plants. For the cooler M-dwarfs, a very large antenna with a steep ’energy funnel’ is required, resembling the cyanobacterial phycobilisome. For the coolest M-dwarfs an upper limit is reached, where increasing antenna size further is subject to steep diminishing returns in photosynthetic output. We conclude that G- and K-stars could support a range of niches for oxygenic photo-autotrophs, including high-light adapted canopy vegetation that may generate detectable bio-signatures. M-dwarfs may only be able to support low light-adapted organisms that have to invest considerable resources in maintaining a large antenna. This may negatively impact global coverage and therefore detectability.