Systemic GM-CSF promotes myelopoiesis and inflammation and GM-CSF blockade is being evaluated as treatment for COVID-19-associated hyperinflammation. Alveolar GM-CSF is however required for monocytes to differentiate into alveolar macrophages (AM) that control alveolar homeostasis and dampen inflammation. By mapping cross-species AM development stages to clinical lung samples, we discovered that COVID-19 is marked by defective GM-CSF-dependent AM instruction and accumulation of proinflammatory macrophages. In a multi-center, open-label, randomized, controlled trial in 81 non-ventilated COVID-19 patients with respiratory failure, we found that inhalation of rhu-GM-CSF did not improve mean oxygenation parameters compared with standard treatment. However, more patients on GM-CSF had a clinical response, and GM-CSF inhalation induced higher numbers of virus-specific CD8 effector lymphocytes and class-switched B cells, without exacerbating systemic hyperinflammation. This translational proof-of-concept study provides rationale for further testing of inhaled GM-CSF as non-invasive treatment to improve alveolar gas exchange and simultaneously boost anti-viral immunity in COVID-19.