Abstract Purpose: Microsatellite instability (MSI) is currently the only predictive biomarker of efficacy of immune checkpoint inhibitors (ICI) in metastatic colorectal cancers (mCRC). However, 10% to 40% of patients with MSI mCRC will experience a primary resistance to ICI. Experimental Design: In two cohorts of patients with MSI mCRC treated with ICI (exploratory, N = 103; validation, N = 35), 3′ RNA sequencing was performed from primary tumors. Previously described single-cell transcriptomic signatures of tumor microenvironment (TME) were analyzed. Results: In the exploratory cohort, the unsupervised clustering allowed the identification of three clusters of tumors with distinct transcriptional profiles: cluster A (“stromalHIGH-proliferationLOW”), cluster B (“stromalHIGH-proliferationMED”), and cluster C (“stromalLOW-proliferationHIGH”), with an enrichment of patients progressing at first disease assessment under ICI in cluster A (30% vs. 12% in cluster B and 8.1% in cluster C; P = 0.074). Progression-free survival (PFS) was also significantly shorter in patients belonging to cluster A, compared with clusters B or C (P < 0.001) with 2-year PFS rates of 33.5%, 80.5%, and 78.3%, respectively. In multivariate analysis, PFS was still significantly longer in patients belonging to cluster B [HR, 0.19; 95% confidence interval (CI), 0.08–0.45; P < 0.001] and cluster C (HR, 0.25; 95% CI, 0.10–0.59; P = 0.02), compared with patients belonging to cluster A. The association of this clustering with PFS under ICI was confirmed in the validation cohort. PFS related to non–ICI-based regimens was not significantly different according to cluster. Conclusions: This unsupervised transcriptomic classification identified three groups of MSI mCRCs with different compositions of TME cells and proliferative capacities of TME/tumor cells. The “stromalHIGH-proliferationLOW” cluster is associated with a poorer prognosis with ICI treatment.