diff --git a/vllm/model_executor/layers/fused_moe/rocm_aiter_fused_moe.py b/vllm/model_executor/layers/fused_moe/rocm_aiter_fused_moe.py
index 495b9daaff4..d9d888296b7 100644
--- a/vllm/model_executor/layers/fused_moe/rocm_aiter_fused_moe.py
+++ b/vllm/model_executor/layers/fused_moe/rocm_aiter_fused_moe.py
@@ -252,7 +252,8 @@ def rocm_aiter_fused_experts(
 
     else:
         quant_method = QuantMethod.NO.value
-        # mxfp4: both w4a4 (quark) and w4a16 (oracle CK) use BLOCK_1X32
+        # mxfp4 i.e. w4a4, w4a16 uses BLOCK_1X32
+        # mxfp6 and mxfp8 are unsupported in AITER currently and use emulation instead
         if quant_config.use_mxfp4_w4a4 or quant_config.use_mxfp4_w4a16:
             quant_method = QuantMethod.BLOCK_1X32.value
         # w8a8 block-scaled
diff --git a/vllm/model_executor/layers/quantization/quark/quark_moe.py b/vllm/model_executor/layers/quantization/quark/quark_moe.py
index d92acb85c26..8737d9e7372 100644
--- a/vllm/model_executor/layers/quantization/quark/quark_moe.py
+++ b/vllm/model_executor/layers/quantization/quark/quark_moe.py
@@ -1025,12 +1025,8 @@ class QuarkOCP_MX_MoEMethod(QuarkMoEMethod):
             get_current_vllm_config().model_config.hf_config, "model_type", None
         )
 
-        # TODO(aiter): extend once rocm_aiter_fused_experts gains dispatch
-        # for the other OCP MX schemes. Today its CK MoE kernel only has an
-        # entry for `w_mxfp4` (w4a16); mixed schemes like `w_mxfp4_a_mxfp6_*`
-        # fall through to QuantMethod.NO and raise "Unsupported kernel config
-        # for moe heuristic dispatch".
-        _AITER_NATIVE_OCP_MX_SCHEMES = ("w_mxfp4",)
+        # TODO: Remove once all OCP MX schemes use the kernel abstraction
+        _AITER_NATIVE_OCP_MX_SCHEMES = ("w_mxfp4", "w_mxfp4_a_mxfp4")
         self.emulate = (
             not current_platform.supports_mx()
             or self.ocp_mx_scheme not in _AITER_NATIVE_OCP_MX_SCHEMES