Recent pan-cancer analysis revealed the global pattern and potential aetiologies of copy number variation signatures in human cancers, particularly those derived from non-hematopoietic tissues. In sharp contrast, the generally low CNV burden in leukemia leaves the CNV landscape and variations largely unexplored, impeding understanding of CNV in leukemia development. Through a comprehensive compilation of public datasets, we constructed LeukAtlas ( https://ngdc.cncb.ac.cn/leukemia ), a user-friendly database encompassing 12,597 CNVs from 1446 AML samples across diverse subtypes and age groups, providing tools for multidimensional CNV analysis. Our analyses suggested the CNV levels significantly varied among AML patients. We discovered two previously unknown CNV patterns in adult AML patients, dominated by segmental LOH and/or minor gain, which have been shown to be associated with chromosomal instability in solid tumors. Additionally, we defined two potential new AML subgroups based on CNVs status, providing new stratification markers within the existing karyotype framework. Representing the most extensive CNV collection in AML, LeukAtlas is a valuable resource for exploring the role of CNVs in the pathogenesis and prognosis stratification of leukemia. Interrogation of this database uncovers novel subclasses with unique CNV profiles and reveals heterogeneous CNV patterns in AML, demonstrating the potential role of chromosomal instability in AML progression.© 2025. The Author(s), under exclusive licence to Springer Nature Limited.