Gastric and gastroesophageal junction (G/GEJ) cancers carry a poor prognosis, and despite recent advancements, most patients die of their disease. Although immune checkpoint blockade became part of the standard-of-care for patients with metastatic G/GEJ cancers, its efficacy and impact on the tumor microenvironment (TME) in early disease remain largely unknown. We hypothesized higher efficacy of neoadjuvant immunotherapy plus chemotherapy in patients with nonmetastatic G/GEJ cancer. In the phase 2 PANDA trial, patients with previously untreated resectable G/GEJ tumors (n = 21) received neoadjuvant treatment with one cycle of atezolizumab monotherapy followed by four cycles of atezolizumab plus docetaxel, oxaliplatin and capecitabine. Treatment was well tolerated. There were grade 3 immune-related adverse events in two of 20 patients (10%) but no grade 4 or 5 immune-related adverse events, and all patients underwent resection without treatment-related delays, meeting the primary endpoint of safety and feasibility. Tissue was obtained at multiple time points, allowing analysis of the effects of single-agent anti-programmed cell death ligand 1 (PD-L1) and the subsequent combination with chemotherapy on the TME. Twenty of 21 patients underwent surgery and were evaluable for secondary pathologic response and survival endpoints, and 19 were evaluable for exploratory translational analyses. A major pathologic response (≤10% residual viable tumor) was observed in 14 of 20 (70%, 95% confidence interval 46–88%) patients, including 9 (45%, 95% confidence interval 23–68%) pathologic complete responses. At a median follow-up of 47 months, 13 of 14 responders were alive and disease-free, and five of six nonresponders had died as a result of recurrence. Notably, baseline anti-programmed cell death protein 1 (PD-1)+CD8+ T cell infiltration was significantly higher in responders versus nonresponders, and comparison of TME alterations following anti-PD-L1 monotherapy versus the subsequent combination with chemotherapy showed an increased immune activation on single-agent PD-1/L1 axis blockade. On the basis of these data, monotherapy anti-PD-L1 before its combination with chemotherapy warrants further exploration and validation in a larger cohort of patients with nonmetastatic G/GEJ cancer. ClinicalTrials.gov registration: NCT03448835.