diff --git a/.github/scripts/build-cuda.sh b/.github/scripts/build-cuda.sh
index 32874e70b..e69548f02 100644
--- a/.github/scripts/build-cuda.sh
+++ b/.github/scripts/build-cuda.sh
@@ -12,19 +12,19 @@ elif [ "${build_arch}" = "aarch64" ]; then
     build_capability="75;80;90"
 
     # CUDA 12.8-12.9: Add sm100/sm120
-    [[ "${cuda_version}" == 12.8.* || "${cuda_version}" == 12.9.* ]] && build_capability="75;80;90;100;120"
+    [[ "${cuda_version}" == 12.8.* || "${cuda_version}" == 12.9.* ]] && build_capability="75;80;87;90;100;120"
 
     # CUDA 13.0+: Add sm100/sm110/sm120
-    [[ "${cuda_version}" == 13.*.* ]] && build_capability="75;80;90;100;110;120"
+    [[ "${cuda_version}" == 13.*.* ]] && build_capability="75;80;87;90;100;103;110;120;121"
 else
     # By default, target Pascal through Hopper.
     build_capability="60;70;75;80;86;89;90"
 
     # CUDA 12.8+: Add sm100 and sm120; remove < sm70 to align with PyTorch 2.8+cu128 minimum
-    [[ "${cuda_version}" == 12.8.* || "${cuda_version}" == 12.9.* ]] && build_capability="70;75;80;86;89;90;100;120"
+    [[ "${cuda_version}" == 12.8.* || "${cuda_version}" == 12.9.* ]] && build_capability="70;75;80;87;86;89;90;100;120;121"
 
     # CUDA 13.0+: Remove < sm75 to align with PyTorch 2.9+cu130 minimum
-    [[ "${cuda_version}" == 13.*.* ]] && build_capability="75;80;86;89;90;100;120"
+    [[ "${cuda_version}" == 13.*.* ]] && build_capability="75;80;87;90;100;103;110;120;121"
 fi
 
 [[ "${build_os}" = windows-* ]] && python3 -m pip install ninja