diff --git a/CMakeLists.txt b/CMakeLists.txt
index bf374cd..082aa0b 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -95,6 +95,7 @@ set(TRITON_BUILD_CONTAINER "" CACHE STRING "Triton container to use a base for b
 set(TRITON_BUILD_CONTAINER_VERSION "" CACHE STRING "Triton container version to target")
 set(TRITON_BUILD_ONNXRUNTIME_VERSION "" CACHE STRING "ONNXRuntime version to build")
 set(TRITON_BUILD_ONNXRUNTIME_OPENVINO_VERSION "" CACHE STRING "ONNXRuntime OpenVINO version to build")
+set(TRITON_BUILD_ONNXRUNTIME_PARALLEL "" CACHE STRING "Limit ONNX Runtime internal build parallelism")
 set(TRITON_BUILD_TARGET_PLATFORM "" CACHE STRING "Target platform for ONNXRuntime build")
 set(TRITON_BUILD_CUDA_VERSION "" CACHE STRING "Version of CUDA install")
 set(TRITON_BUILD_CUDA_HOME "" CACHE PATH "Path to CUDA install")
@@ -324,6 +325,9 @@ endif() # TRITON_ENABLE_ONNXRUNTIME_OPENVINO
 #
 if(TRITON_ONNXRUNTIME_DOCKER_BUILD)
   set(_GEN_FLAGS "")
+  if(NOT "${TRITON_BUILD_ONNXRUNTIME_PARALLEL}" STREQUAL "")
+    set(_GEN_FLAGS ${_GEN_FLAGS} "--ort-build-parallel=${TRITON_BUILD_ONNXRUNTIME_PARALLEL}")
+  endif() # TRITON_BUILD_ONNXRUNTIME_PARALLEL
   if(NOT ${TRITON_BUILD_TARGET_PLATFORM} STREQUAL "")
     set(_GEN_FLAGS ${_GEN_FLAGS} "--target-platform=${TRITON_BUILD_TARGET_PLATFORM}")
   endif() # TRITON_BUILD_TARGET_PLATFORM
diff --git a/tools/gen_ort_dockerfile.py b/tools/gen_ort_dockerfile.py
index 17ec6a3..f82664f 100755
--- a/tools/gen_ort_dockerfile.py
+++ b/tools/gen_ort_dockerfile.py
@@ -382,12 +382,16 @@ def dockerfile_for_linux(output_file):
     else:
         cuda_archs = "75-real;80-real;86-real;90-real;100f;110f;120f"
 
+    ort_parallel_arg = "--parallel"
+    if FLAGS.ort_build_parallel is not None:
+        ort_parallel_arg += f" {FLAGS.ort_build_parallel}"
+
     df += """
 WORKDIR /workspace/onnxruntime
-ARG COMMON_BUILD_ARGS="--config ${{ONNXRUNTIME_BUILD_CONFIG}} --parallel --skip_submodule_sync --build_shared_lib \
+ARG COMMON_BUILD_ARGS="--config ${{ONNXRUNTIME_BUILD_CONFIG}} {} --skip_submodule_sync --build_shared_lib \
     --compile_no_warning_as_error --build_dir /workspace/build --cmake_extra_defines CMAKE_CUDA_ARCHITECTURES='{}'  --cmake_extra_defines CMAKE_POLICY_VERSION_MINIMUM=3.5 --build_wheel"
 """.format(
-        cuda_archs
+        ort_parallel_arg, cuda_archs
     )
 
     df += """
@@ -541,6 +545,13 @@ def preprocess_gpu_flags():
         choices=["Debug", "Release", "RelWithDebInfo"],
         help="ORT build configuration.",
     )
+    parser.add_argument(
+        "--ort-build-parallel",
+        type=int,
+        required=False,
+        default=None,
+        help="Limit ONNX Runtime internal build parallelism.",
+    )
     parser.add_argument(
         "--target-platform",
         required=False,