diff --git a/vllm/model_executor/layers/quantization/utils/w8a8_utils.py b/vllm/model_executor/layers/quantization/utils/w8a8_utils.py
index d77722499d0e9..ac55ef037079a 100644
--- a/vllm/model_executor/layers/quantization/utils/w8a8_utils.py
+++ b/vllm/model_executor/layers/quantization/utils/w8a8_utils.py
@@ -12,7 +12,7 @@
 
 def sparse_cutlass_supported() -> bool:
     # sparse cutlass is not supported on Rocm
-    if current_platform.is_rocm():
+    if not current_platform.is_cuda():
         return False
 
     capability_tuple = current_platform.get_device_capability()
@@ -23,7 +23,7 @@ def sparse_cutlass_supported() -> bool:
 
 def cutlass_fp8_supported() -> bool:
     # cutlass is not supported on Rocm
-    if current_platform.is_rocm():
+    if not current_platform.is_cuda():
         return False
 
     capability_tuple = current_platform.get_device_capability()