diff --git a/Dockerfile.mxfp4 b/Dockerfile.mxfp4
index bedcf4a..9df3e65 100644
--- a/Dockerfile.mxfp4
+++ b/Dockerfile.mxfp4
@@ -101,7 +101,7 @@ ARG FLASHINFER_SHA=f349e52496a72a00d8c4ac02c7a1e38523ff7194
 ARG CUTLASS_SHA=c7516ad20f3d022fdbc93e9468643bf3b577e02c
 
 RUN --mount=type=cache,id=uv-cache,target=/root/.cache/uv \
-     uv pip install nvidia-nvshmem-cu13 "apache-tvm-ffi<0.2"
+     uv pip install "nvidia-nvshmem-cu13<3.6" "apache-tvm-ffi<0.2"
 
 # Clone FlashInfer (cached for faster rebuilds)
 RUN --mount=type=cache,id=git-flashinfer,target=/git-cache/flashinfer \
@@ -272,7 +272,7 @@ ENV PATH=$VLLM_BASE_DIR:$PATH
 
 # Final extra deps
 RUN --mount=type=cache,id=uv-cache,target=/root/.cache/uv \
-    uv pip install ray[default] fastsafetensors nvidia-nvshmem-cu13
+    uv pip install ray[default] fastsafetensors "nvidia-nvshmem-cu13<3.6"
 
 # Build metadata (generated by build-and-copy.sh)
 COPY build-metadata.yaml /workspace/build-metadata.yaml