diff --git a/vllm/model_executor/layers/quantization/quark/quark_moe.py b/vllm/model_executor/layers/quantization/quark/quark_moe.py
index ebe37d8dc5..bce888415c 100644
--- a/vllm/model_executor/layers/quantization/quark/quark_moe.py
+++ b/vllm/model_executor/layers/quantization/quark/quark_moe.py
@@ -1322,6 +1322,11 @@ class QuarkOCP_MX_MoEMethod(QuarkMoEMethod):
                 gemm1_clamp_limit=getattr(layer, "swiglu_limit", None),
             )
 
+    @property
+    def supports_eplb(self) -> bool:
+        # AITER shuffle keeps expert dim outermost, so EPLB row moves are layout-safe.
+        return True
+
     @property
     def is_monolithic(self) -> bool:
         if self.moe_kernel is not None: