Notwithstanding remarkable treatment success with anti-PD-1 monoclonal antibody, oncogenic mechanism of PD-L1 regulation in gastric cancer (GC) remains poorly understood. We hypothesized that ARID1A might be related to tumor PD-L1 expression in GC. We found that tumor PD-L1 positivity was associated with loss of ARID1A and showed trend toward better survival of patients with various molecular subtypes of GC (experimental set, n = 273). Considering heterogeneous ARID1A expression, we validated this using whole tissue sections (n = 159) and found that loss of ARID1A was correlated with microsatellite instability-high (MSI-H), Epstein-Barr virus (EBV), and PD-L1 positivity. Furthermore, for patients with MSI-H tumors, the degree of PD-L1 expression was significantly higher in ARID1A-deficient tumors. After ARID1A knockdown in GC cell lines, total and membranous PD-L1 protein, and PD-L1 mRNA levels were increased based on Western blot, flow cytometry, and qRT-PCR, respectively. With IFN-gamma treatment, PD-L1 expression was significantly increased both in ARID1A-deficient cancer cells and controls, but the increase was not more pronounced in the former. Loss of ARID1A increased PD-L1 via activating AKT signaling, while LY294002 (PI3K inhibitor) decreased PD-L1 levels. Furthermore, we found that 3 MSI-H tumors showing highest expression of PD-L1 had simultaneous KRAS mutation and loss of ARID1A, suggesting a possible synergistic role boosting PD-L1. Our results strongly indicate that loss of ARID1A is tightly associated with high PD-L1 expression in GC. These results would increase our understanding of the oncogenic mechanism of PD-L1 regulation in GC, and also help to find the optimal candidates for immunotherapy.