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‐γ 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.