diff --git a/vllm/model_executor/layers/fused_moe/experts/flashinfer_cutedsl_moe.py b/vllm/model_executor/layers/fused_moe/experts/flashinfer_cutedsl_moe.py
index 5805a4dd5..a1db26619 100644
--- a/vllm/model_executor/layers/fused_moe/experts/flashinfer_cutedsl_moe.py
+++ b/vllm/model_executor/layers/fused_moe/experts/flashinfer_cutedsl_moe.py
@@ -23,6 +23,7 @@ from vllm.model_executor.layers.quantization.utils.quant_utils import (
 from vllm.platforms import current_platform
 from vllm.utils.flashinfer import (
     flashinfer_cutedsl_grouped_gemm_nt_masked,
+    has_flashinfer_cutedsl_grouped_gemm_nt_masked,
     scaled_fp4_grouped_quantize,
     silu_and_mul_scaled_nvfp4_experts_quantize,
 )
@@ -60,7 +61,11 @@ class FlashInferCuteDSLExperts(mk.FusedMoEExpertsModular):
     @staticmethod
     def _supports_current_device() -> bool:
         p = current_platform
-        return p.is_cuda() and p.is_device_capability_family(100)
+        return (
+            p.is_cuda()
+            and p.is_device_capability_family(100)
+            and has_flashinfer_cutedsl_grouped_gemm_nt_masked()
+        )
 
     @staticmethod
     def _supports_no_act_and_mul() -> bool:
diff --git a/vllm/model_executor/layers/fused_moe/experts/trtllm_fp8_moe.py b/vllm/model_executor/layers/fused_moe/experts/trtllm_fp8_moe.py
index f57a05dc6..671435a88 100644
--- a/vllm/model_executor/layers/fused_moe/experts/trtllm_fp8_moe.py
+++ b/vllm/model_executor/layers/fused_moe/experts/trtllm_fp8_moe.py
@@ -27,6 +27,7 @@ from vllm.model_executor.layers.quantization.utils.quant_utils import (
     kMxfp8Static,
 )
 from vllm.platforms import current_platform
+from vllm.utils.flashinfer import has_flashinfer_trtllm_fused_moe
 
 logger = init_logger(__name__)
 
@@ -61,8 +62,11 @@ class TrtLlmFp8ExpertsBase:
     def _supports_current_device() -> bool:
         """Supports only Blackwell-family GPUs."""
         p = current_platform
-        # Add check flashinfer trtllm is available
-        return p.is_cuda() and p.is_device_capability_family(100)
+        return (
+            p.is_cuda()
+            and p.is_device_capability_family(100)
+            and has_flashinfer_trtllm_fused_moe()
+        )
 
     @staticmethod
     def _supports_no_act_and_mul() -> bool:
diff --git a/vllm/model_executor/layers/fused_moe/experts/trtllm_nvfp4_moe.py b/vllm/model_executor/layers/fused_moe/experts/trtllm_nvfp4_moe.py
index 1df50197c..7960bdf44 100644
--- a/vllm/model_executor/layers/fused_moe/experts/trtllm_nvfp4_moe.py
+++ b/vllm/model_executor/layers/fused_moe/experts/trtllm_nvfp4_moe.py
@@ -24,6 +24,7 @@ from vllm.model_executor.layers.quantization.utils.quant_utils import (
     kNvfp4Static,
 )
 from vllm.platforms import current_platform
+from vllm.utils.flashinfer import has_flashinfer_trtllm_fused_moe
 
 
 class TrtLlmNvFp4ExpertsBase:
@@ -80,7 +81,11 @@ class TrtLlmNvFp4ExpertsBase:
     def _supports_current_device() -> bool:
         """Supports only Blackwell-family GPUs."""
         p = current_platform
-        return p.is_cuda() and p.is_device_capability_family(100)
+        return (
+            p.is_cuda()
+            and p.is_device_capability_family(100)
+            and has_flashinfer_trtllm_fused_moe()
+        )
 
     @staticmethod
     def _supports_no_act_and_mul() -> bool:
diff --git a/vllm/utils/flashinfer.py b/vllm/utils/flashinfer.py
index fed44d04f..0db05851b 100644
--- a/vllm/utils/flashinfer.py
+++ b/vllm/utils/flashinfer.py
@@ -244,7 +244,7 @@ def has_flashinfer_cutedsl_grouped_gemm_nt_masked() -> bool:
     required_functions = [
         ("flashinfer.cute_dsl.blockscaled_gemm", "grouped_gemm_nt_masked"),
         ("flashinfer", "scaled_fp4_grouped_quantize"),
-        ("flashinfer", "silu_and_scaled_nvfp4_experts_quantize"),
+        ("flashinfer", "silu_and_mul_scaled_nvfp4_experts_quantize"),
     ]
 
     for module_name, attr_name in required_functions: