We investigated whether artificial intelligence (AI)-based tumor microenvironment profiling correlates with treatment efficacy in unresectable hepatocellular carcinoma (HCC) patients treated with immune checkpoint inhibitor (ICI) therapies. Spatial distribution of immune/non-immune cells from pretreatment H&E images of 163 patients was retrospectively analyzed using an AI/deep-learning model. High tumor endothelial cell (TEC) density was associated with significantly longer progression-free survival (PFS) in the atezolizumab plus bevacizumab (atezo-bev) cohort (HR 0.51 [0.27-0.97]; p = 0.037) but not in the anti-PD-1 monotherapy cohort (HR 1.02 [0.59-1.77]; p = 0.935). Conversely, inflamed immune phenotype, characterized by high intratumoral TIL densities, predicted longer PFS after anti-PD-1 monotherapy (HR 0.50 [0.25-0.99]; p = 0.042) but not after atezo-bev (HR 0.92 [0.50-1.69]; p = 0.762). Our exploratory analysis using AI/deep-learning model demonstrated high TEC density predicted superior outcomes with atezo-bev, while TIL presence correlated with improved anti-PD-1 monotherapy efficacy in HCC patients, suggesting potential clinical applicability in treatment selection.