Costimulatory molecules (CMGs) play essential roles in multiple cancers. However, lncRNAs regulating costimulatory molecules have not been fully explored in gastric cancer (GC). Public data of GC patients were obtained from The Cancer Genome Atlas database. R software v4.1.1, SPSS v13.0, and GraphPad Prism 8 were used to perform all the analyses. The Limma package was used for differential expression analysis. The survival package was used for patient prognosis analysis. The gene set enrichment analysis (GSEA), gene ontology (GO), and the Kyoto encyclopedia of genes and genomes (KEGG) analysis were used for pathway enrichment analysis. qRT-PCR was used to detect the RNA level of target lncRNA. CCK-8 and colony formation assay were used to assess the proliferation ability of GC cells. The transwell assay was used to evaluate the invasion and migration ability of GC cells. We first identified CMG-related lncRNAs (CMLs) through co-expression analysis. Then, an eight-CML-based signature was constructed to predict patient overall survival (OS), which showed satisfactory predictive efficiency (the training cohort: 1-year AUC = 0.764, 3-year AUC = 0.810, 5-year AUC = 0.840; the validation cohort: 1-year AUC = 0.661, 3-year AUC = 0.718, 5-year AUC = 0.822). The patients in the high-risk group tend to have a worse prognosis. GSEA showed that epithelial-mesenchymal transition, KRAS signaling, and angiogenesis were aberrantly activated in high-risk patients. GO and KEGG analyses indicated that the biological difference between high- and low-risk patients was mainly enriched in the extracellular matrix. Immune infiltration analysis showed that macrophages (M1 and M2), dendritic cells, monocytes, Tregs, and T regulatory cells were positively correlated with the risk scores, partly responsible for the worsening OS of high-risk patients. Finally, lncRNA AP000695.2 was selected for further experiments. The result showed that AP000695.2 was upregulated in GC cell lines and could facilitate the proliferation, invasion, and migration of GC cells. In summary, this study established an effective prognosis model based on eight CMLs, which would be helpful for further therapy options for cancer. Also, we found that AP000695.2 could promote GC cell malignant phenotype, making it an underlying therapy target in GC.