diff --git a/vllm_ascend/ops/fused_moe/fused_moe.py b/vllm_ascend/ops/fused_moe/fused_moe.py
index f4396992..9adfe085 100644
--- a/vllm_ascend/ops/fused_moe/fused_moe.py
+++ b/vllm_ascend/ops/fused_moe/fused_moe.py
@@ -533,7 +533,7 @@ class AscendSharedFusedMoE(SharedFusedMoE, AscendFusedMoE):
             # NOTE: This is exactly the opposite of `maybe_all_reduce_tensor_model_parallel`
             forward_context = get_forward_context()
             moe_comm_type = forward_context.moe_comm_type
-            if moe_comm_type in {MoECommType.ALLTOALL, MoECommType.MC2} \
+            if moe_comm_type in {MoECommType.ALLTOALL, MoECommType.MC2, MoECommType.FUSED_ALLTOALL} \
                     and not shared_expert_dp_enabled():
                 shared_out = tensor_model_parallel_all_reduce(shared_out)
         else: