diff --git a/grpc_servicer/smg_grpc_servicer/sglang/request_manager.py b/grpc_servicer/smg_grpc_servicer/sglang/request_manager.py
index 8814f7f702..673d76bcc0 100644
--- a/grpc_servicer/smg_grpc_servicer/sglang/request_manager.py
+++ b/grpc_servicer/smg_grpc_servicer/sglang/request_manager.py
@@ -28,6 +28,7 @@
     HealthCheckOutput,
     TokenizedEmbeddingReqInput,
     TokenizedGenerateReqInput,
+    WatchLoadUpdateReq,
 )
 from sglang.srt.observability.req_time_stats import (
     APIServerReqTimeStats,
@@ -679,6 +680,15 @@ async def cleanup(request_id):
 
                 cleanup_tasks.append(asyncio.create_task(cleanup(rid)))
 
+        # Forward load info to DataParallelController for token-aware balancing.
+        # Mirrors TokenizerManager._handle_batch_output logic: when dp_size > 1,
+        # each scheduler piggybacks its load (num_reqs, num_tokens) on batch output.
+        # Without this, DPBudget stays at zero and total_tokens/total_requests
+        # policies degenerate to always picking rank 0.
+        if self.server_args.dp_size > 1 and batch_out.load is not None:
+            load_update = WatchLoadUpdateReq(loads=[batch_out.load])
+            self.send_to_scheduler.send_pyobj(load_update)
+
         # Execute all queue.put() operations in parallel
         if put_tasks:
             await asyncio.gather(*put_tasks, return_exceptions=True)