diff --git a/docker/Dockerfile b/docker/Dockerfile
index c4576606ebb2..976af6621939 100644
--- a/docker/Dockerfile
+++ b/docker/Dockerfile
@@ -9,7 +9,7 @@ ARG SGL_VERSION
 ARG SGL_DEEP_GEMM_VERSION=0.1.5.post2
 ARG USE_LATEST_SGLANG=0
 ARG GDRCOPY_VERSION=2.5.1
-ARG NCCL_VERSION=2.30.7
+ARG SGL_NCCL_VERSION=2.30.7
 ARG PIP_DEFAULT_INDEX
 ARG UBUNTU_MIRROR
 ARG GITHUB_ARTIFACTORY=github.com
@@ -173,7 +173,11 @@ ARG CUDA_VERSION
 ARG BUILD_TYPE
 ARG SGL_KERNEL_VERSION
 ARG GITHUB_ARTIFACTORY
-ARG NCCL_VERSION
+# Renamed from NCCL_VERSION: the nvidia/cuda base image sets ENV NCCL_VERSION
+# (Debian apt format, e.g. 2.28.3-1), which shadows a same-named ARG inside RUN,
+# so the CUDA-13 nvidia-nccl-cu13 pip install below resolved an invalid PyPI
+# version. A distinct ARG name avoids the collision.
+ARG SGL_NCCL_VERSION
 
 WORKDIR /sgl-workspace
 
@@ -251,7 +255,7 @@ RUN --mount=type=cache,target=/root/.cache/pip \
     && python3 -m pip install --extra-index-url https://download.pytorch.org/whl/cu${CUINDEX} ".[${BUILD_TYPE}]" \
     && if [ "${CUDA_VERSION%%.*}" = "13" ]; then \
            python3 -m pip install --force-reinstall --no-deps \
-               "nvidia-nccl-cu13==${NCCL_VERSION}"; \
+               "nvidia-nccl-cu13==${SGL_NCCL_VERSION}"; \
        fi \
     && if [ "${CUDA_VERSION%%.*}" = "12" ]; then \
            pip list --format=freeze | awk -F'==' '/-cu13(==|$)/ {print $1}' \