Background: Type 2 diabetes (T2D) is an established late effect of cancer treatment among long-term survivors of childhood cancer. Genetic factors underpinning T2D in diverse populations of survivors have not been well studied. Methods: We conducted a multi-ancestry genome-wide association study (GWAS) for clinically ascertained T2D among survivors of European (EUR; N = 3,102 with 261 cases) and African (AFR; N = 574 with 43 cases) ancestry from SJLIFE. Replication analyses were performed between ancestries in SJLIFE and in EUR survivors in CCSS (N = 5,965; 270 self-reported cases). Two published T2D polygenic risk scores (PRSs) were assessed in survivors: a 338-variant multi-ancestry PRS (N~1.4 million; 49% non-EUR descent) and a ~6.9 million-variant EUR-only PRS (N~160K; DIAGRAM Consortium and UK Biobank Study). Treatment-related T2D risk effect modification was evaluated for abdominal irradiation and alkylating agents. Results: In SJLIFE AFR survivors, one novel locus with suggestive significance was identified (5p15.2: OR = 10.19, P = 5.1x10^-7), replicating in both SJLIFE EUR (P = 0.011) and CCSS EUR survivors (P = 0.021). A EUR-specific genome-wide significant association at 8q11.21 (SNTG1 intronic variant; OR = 1.99; P = 4.4x10^-8) was replicated in CCSS (P = 8.1x10^-3). Two other loci with suggestive associations (P < 5x10^-6) in SJLIFE EUR survivors replicated in SJLIFE AFR survivors (P < 0.05), achieving genome-wide significance in multi-ancestry meta-analysis (2p25.3: OR = 2.05, P = 4.5x10^-8; 19p12: OR = 2.43 P = 5.7x10^-9). Each of the three novel trans-ancestral loci overlapped putative Polycomb-repressed regions, i.e., chromatin-based gene regulation elements, in pancreatic cells and displayed treatment-related effect heterogeneity across ancestry groups. Notably, AFR survivors with risk alleles experienced disproportionately greater T2D risk if treated with alkylating agents (2p25.3: OR = 3.95; 19p12: OR = 5.74; 5p15.2: OR = 17.81). Increases in the T2D odds per multi-ancestry PRS standard deviation were consistent in survivors across ancestries (SJLIFE EUR: OR = 1.84, P = 1.1x10^-16; SJLIFE AFR: OR = 1.80, P = 2.8x10^-3, CCSS EUR: OR = 1.60, P = 8.4x10^-13). However, T2D risk association with the EUR-only PRS was absent in AFR survivors (OR = 0.97, P = 0.95). Conclusions: Multi-ancestry genetic analyses revealed three novel T2D risk alleles associated with disproportionately greater alkylating agent-related risk among African-ancestry survivors. Furthermore, an external multi-ancestry T2D PRS was associated with increased risk in diverse ancestry survivors, whereas a EUR-only PRS was not useful for African-ancestry survivors. This study supports precision diabetes surveillance and survivorship care for all childhood cancer survivors, including those in minority ancestry groups.