Despite improvements in prevention and treatment, severe coronavirus disease 2019 (COVID-19) is associated with high mortality. Phosphoinositide 3-kinase (PI3K) pathways contribute to cytokine and cell-mediated lung inflammation. We conducted a randomized, placebo-controlled, double-blind pilot trial to determine the feasibility, safety, and preliminary activity of duvelisib, a PI3Kδγ inhibitor, for the treatment of COVID-19 critical illness. We enrolled adults aged ≥18 years with a primary diagnosis of COVID-19 with hypoxic respiratory failure, shock, and/or new cardiac disease, without improvement after at least 48 hours of corticosteroid. Participants received duvelisib (25 mg) or placebo for up to 10 days. Participants had daily semi-quantitative viral load measurements performed. Dose modifications were protocol driven due to adverse events (AEs) or logarithmic change in viral load. The primary endpoint was 28-day overall survival (OS). Secondary endpoints included hospital and intensive care unit length of stay, 60-day OS, and duration of critical care interventions. Safety endpoints included viral kinetics and AEs. Exploratory endpoints included serial cytokine measurements and cytometric analysis. Fifteen patients were treated in the duvelisib cohort, and 13 in the placebo cohort. OS at 28 days was 67% (95% confidence interval [CI], 38%-88%) compared to 62% (95% CI, 32%-86%) for placebo (P = .544). Sixty-day OS was 60% versus 46%, respectively (hazard ratio, 0.66 [95% CI, .22-1.96]; P = .454). Other secondary outcomes were comparable. Duvelisib was associated with lower inflammatory cytokines. In this pilot study, duvelisib did not significantly improve 28-day OS compared to placebo for severe COVID-19. Duvelisib appeared safe in this critically ill population and was associated with reduction in cytokines implicated in COVID-19 and acute respiratory distress syndrome, supporting further investigation. NCT04372602.