Recent advances in genomic technologies have revolutionized acute myeloid leukemia (AML) understanding by identifying potential novel actionable genomic alterations. Consequently, current risk stratification at diagnosis not only relies on cytogenetics, but also on the inclusion of several of these abnormalities. Despite this progress, AML remains a heterogeneous and complex malignancy with variable response to current therapy. Although copy-number alterations (CNAs) are accepted prognostic markers in cancers, large-scale genomic studies aiming at identifying specific prognostic CNA-based markers in AML are still lacking. Using 367 AML, we identified four recurrent CNA on chromosomes 11 and 21 that predicted outcome even after adjusting for standard prognostic risk factors and potentially delineated two new subclasses of AML with poor prognosis. ERG amplification, the most frequent CNA, was related to cytarabine resistance, a cornerstone drug of AML therapy. These findings were further validated in The Cancer Genome Atlas data. Our results demonstrate that specific CNA are of independent prognostic relevance, and provide new molecular information into the genomic basis of AML and cytarabine response. Finally, these CNA identified two potential novel risk groups of AML, which when confirmed prospectively, may improve the clinical risk stratification and potentially the AML outcome.