diff --git a/cpp/kernels/fmha_v2/fmha_test.py b/cpp/kernels/fmha_v2/fmha_test.py
index 3523ee1d1002..ad2957b42d70 100644
--- a/cpp/kernels/fmha_v2/fmha_test.py
+++ b/cpp/kernels/fmha_v2/fmha_test.py
@@ -1,7 +1,12 @@
 import subprocess
 
 import pytest
-from cuda import cuda, nvrtc
+
+try:
+    from cuda.bindings import driver as cuda
+    from cuda.bindings import nvrtc
+except ImportError:
+    from cuda import cuda, nvrtc
 
 
 def ASSERT_DRV(err):
diff --git a/requirements.txt b/requirements.txt
index 55d2777964bf..78c3b37c43e6 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -3,7 +3,7 @@
 accelerate>=0.25.0
 build
 colored
-cuda-python # Do not override the custom version of cuda-python installed in the NGC PyTorch image.
+cuda-python>=12,<13
 diffusers>=0.27.0
 lark
 mpi4py
diff --git a/tensorrt_llm/_ipc_utils.py b/tensorrt_llm/_ipc_utils.py
index c19fa516babd..d0e30a61d6c5 100644
--- a/tensorrt_llm/_ipc_utils.py
+++ b/tensorrt_llm/_ipc_utils.py
@@ -17,17 +17,20 @@
 import sys
 from typing import List, Tuple
 
-from cuda import cuda, cudart
-from cuda.cudart import cudaError_t
+try:
+    from cuda.bindings import driver as cuda
+    from cuda.bindings import runtime as cudart
+except ImportError:
+    from cuda import cuda, cudart
 
 from ._utils import mpi_comm
 from .logger import logger
 from .mapping import Mapping
 
 
-def _raise_if_error(error: cudaError_t | cuda.CUresult):
-    if isinstance(error, cudaError_t):
-        if error != cudaError_t.cudaSuccess:
+def _raise_if_error(error: cudart.cudaError_t | cuda.CUresult):
+    if isinstance(error, cudart.cudaError_t):
+        if error != cudart.cudaError_t.cudaSuccess:
             raise RuntimeError(f"CUDA Runtime API error: {repr(error)}")
     if isinstance(error, cuda.CUresult):
         if error != cuda.CUresult.CUDA_SUCCESS:
diff --git a/tensorrt_llm/_mnnvl_utils.py b/tensorrt_llm/_mnnvl_utils.py
index 4e0b7554e16e..c5ead1ece7cb 100644
--- a/tensorrt_llm/_mnnvl_utils.py
+++ b/tensorrt_llm/_mnnvl_utils.py
@@ -18,7 +18,11 @@
 
 import pynvml
 import torch
-from cuda import cuda
+
+try:
+    from cuda.bindings import driver as cuda
+except ImportError:
+    from cuda import cuda
 
 from ._dlpack_utils import pack_strided_memory
 from ._utils import mpi_comm
diff --git a/tensorrt_llm/auto_parallel/cluster_info.py b/tensorrt_llm/auto_parallel/cluster_info.py
index 514504dbe81d..8bdf5eba29e3 100644
--- a/tensorrt_llm/auto_parallel/cluster_info.py
+++ b/tensorrt_llm/auto_parallel/cluster_info.py
@@ -5,7 +5,11 @@
 
 import pynvml
 import torch
-from cuda import cudart
+
+try:
+    from cuda.bindings import runtime as cudart
+except ImportError:
+    from cuda import cudart
 
 from tensorrt_llm._utils import DictConversion
 from tensorrt_llm.logger import logger
diff --git a/tensorrt_llm/runtime/generation.py b/tensorrt_llm/runtime/generation.py
index ec223f31c1a3..bf6e228f7699 100755
--- a/tensorrt_llm/runtime/generation.py
+++ b/tensorrt_llm/runtime/generation.py
@@ -29,7 +29,10 @@
 import torch
 import tensorrt as trt
 # isort: on
-from cuda import cudart
+try:
+    from cuda.bindings import runtime as cudart
+except ImportError:
+    from cuda import cudart
 
 from tensorrt_llm.runtime.memory_pools.memory_pools_allocator import \
     MemoryPoolsAllocator
diff --git a/tensorrt_llm/runtime/multimodal_model_runner.py b/tensorrt_llm/runtime/multimodal_model_runner.py
index 9d6be2bddad8..4d71846c3fe0 100644
--- a/tensorrt_llm/runtime/multimodal_model_runner.py
+++ b/tensorrt_llm/runtime/multimodal_model_runner.py
@@ -13,7 +13,12 @@
 from typing import Optional, Tuple
 
 import torch.nn.functional as F
-from cuda import cudart
+
+try:
+    from cuda.bindings import runtime as cudart
+except ImportError:
+    from cuda import cudart
+
 from huggingface_hub import hf_hub_download
 from PIL import Image, UnidentifiedImageError
 from safetensors import safe_open
diff --git a/tests/integration/defs/sysinfo/get_sysinfo.py b/tests/integration/defs/sysinfo/get_sysinfo.py
index a7de1871e517..4efd791d7420 100644
--- a/tests/integration/defs/sysinfo/get_sysinfo.py
+++ b/tests/integration/defs/sysinfo/get_sysinfo.py
@@ -24,7 +24,11 @@
 
 import psutil
 import pynvml
-from cuda import cuda
+
+try:
+    from cuda.bindings import driver as cuda
+except ImportError:
+    from cuda import cuda
 
 # Logger
 logger = logging.getLogger(__name__)
diff --git a/tests/microbenchmarks/all_reduce.py b/tests/microbenchmarks/all_reduce.py
index 13cb4271bbbc..4aad4170ec59 100644
--- a/tests/microbenchmarks/all_reduce.py
+++ b/tests/microbenchmarks/all_reduce.py
@@ -18,7 +18,10 @@
 # isort: off
 import torch
 # isort: on
-from cuda import cuda, cudart
+try:
+    from cuda.bindings import runtime as cudart
+except ImportError:
+    from cuda import cudart
 
 import tensorrt_llm as tllm
 from tensorrt_llm import Mapping, Tensor
diff --git a/tests/microbenchmarks/build_time_benchmark.py b/tests/microbenchmarks/build_time_benchmark.py
index 3313cae1cda6..fad8ae2a9af5 100644
--- a/tests/microbenchmarks/build_time_benchmark.py
+++ b/tests/microbenchmarks/build_time_benchmark.py
@@ -7,7 +7,11 @@
 import traceback
 
 import tensorrt as trt
-from cuda import cudart
+
+try:
+    from cuda.bindings import runtime as cudart
+except ImportError:
+    from cuda import cudart
 
 import tensorrt_llm
 from tensorrt_llm import (AutoConfig, AutoModelForCausalLM, BuildConfig,
diff --git a/tests/unittest/_torch/multi_gpu/test_lowprecision_allreduce.py b/tests/unittest/_torch/multi_gpu/test_lowprecision_allreduce.py
index 3aa6871fbe4f..9d271a1d6e60 100644
--- a/tests/unittest/_torch/multi_gpu/test_lowprecision_allreduce.py
+++ b/tests/unittest/_torch/multi_gpu/test_lowprecision_allreduce.py
@@ -34,7 +34,10 @@
 def run_single_rank(dtype, strategy, message_size):
     import numpy as np
     import torch
-    from cuda import cuda
+    try:
+        from cuda.bindings import driver as cuda
+    except ImportError:
+        from cuda import cuda
 
     import tensorrt_llm
     from tensorrt_llm._torch.distributed import AllReduce, AllReduceStrategy
diff --git a/tests/unittest/trt/functional/test_allreduce_norm.py b/tests/unittest/trt/functional/test_allreduce_norm.py
index a6a8ee477e11..dbd4448c45d1 100644
--- a/tests/unittest/trt/functional/test_allreduce_norm.py
+++ b/tests/unittest/trt/functional/test_allreduce_norm.py
@@ -21,7 +21,10 @@
 import torch
 # isort: on
 
-from cuda import cudart
+try:
+    from cuda.bindings import runtime as cudart
+except ImportError:
+    from cuda import cudart
 from parameterized import parameterized
 from utils.util import create_session, run_session, unittest_name_func
 
diff --git a/tests/unittest/trt/functional/test_allreduce_prepost_residual_norm.py b/tests/unittest/trt/functional/test_allreduce_prepost_residual_norm.py
index 295e92c0aec9..bf0592193d59 100644
--- a/tests/unittest/trt/functional/test_allreduce_prepost_residual_norm.py
+++ b/tests/unittest/trt/functional/test_allreduce_prepost_residual_norm.py
@@ -21,7 +21,10 @@
 import torch
 # isort: on
 
-from cuda import cudart
+try:
+    from cuda.bindings import runtime as cudart
+except ImportError:
+    from cuda import cudart
 from parameterized import parameterized
 from utils.util import create_session, run_session, unittest_name_func
 
diff --git a/tests/unittest/trt/functional/test_nccl.py b/tests/unittest/trt/functional/test_nccl.py
index 760e6538123f..61b9eacc825d 100644
--- a/tests/unittest/trt/functional/test_nccl.py
+++ b/tests/unittest/trt/functional/test_nccl.py
@@ -21,7 +21,10 @@
 import torch
 # isort: on
 
-from cuda import cudart
+try:
+    from cuda.bindings import runtime as cudart
+except ImportError:
+    from cuda import cudart
 from parameterized import parameterized
 from utils.util import create_session, run_session, unittest_name_func
 
diff --git a/tests/unittest/trt/functional/test_pp_reduce_scatter.py b/tests/unittest/trt/functional/test_pp_reduce_scatter.py
index fba41cbf76fc..660a499c9226 100644
--- a/tests/unittest/trt/functional/test_pp_reduce_scatter.py
+++ b/tests/unittest/trt/functional/test_pp_reduce_scatter.py
@@ -21,7 +21,10 @@
 import torch
 # isort: on
 
-from cuda import cudart
+try:
+    from cuda.bindings import runtime as cudart
+except ImportError:
+    from cuda import cudart
 from parameterized import parameterized
 from utils.util import create_session, run_session, unittest_name_func
 
diff --git a/tests/unittest/utils/util.py b/tests/unittest/utils/util.py
index 31dcb8efd9e7..379102b767d8 100644
--- a/tests/unittest/utils/util.py
+++ b/tests/unittest/utils/util.py
@@ -7,7 +7,13 @@
 import pytest
 import tensorrt as trt
 import torch
-from cuda import cuda, nvrtc
+
+try:
+    from cuda.bindings import driver as cuda
+    from cuda.bindings import nvrtc
+except ImportError:
+    from cuda import cuda, nvrtc
+
 from parameterized import parameterized
 
 import tensorrt_llm