diff --git a/python/sglang/srt/model_executor/model_runner.py b/python/sglang/srt/model_executor/model_runner.py
index 607766703..d3ff835bb 100644
--- a/python/sglang/srt/model_executor/model_runner.py
+++ b/python/sglang/srt/model_executor/model_runner.py
@@ -194,6 +194,11 @@ _is_npu = is_npu()
 _is_cpu_amx_available = cpu_has_amx_support()
 _is_cpu_arm64 = is_host_cpu_arm64()
 
+if _is_npu:
+    from sglang.srt.hardware_backend.npu.utils import init_npu_backend
+
+    init_npu_backend()
+
 MLA_ATTENTION_BACKENDS = [
     "aiter",
     "flashinfer",
diff --git a/python/sglang/srt/server_args.py b/python/sglang/srt/server_args.py
index 044463090..065ce26ba 100644
--- a/python/sglang/srt/server_args.py
+++ b/python/sglang/srt/server_args.py
@@ -891,12 +891,7 @@ class ServerArgs:
 
     def _handle_npu_backends(self):
         if self.device == "npu":
-            from sglang.srt.hardware_backend.npu.utils import (
-                init_npu_backend,
-                set_default_server_args,
-            )
-
-            init_npu_backend()
+            from sglang.srt.hardware_backend.npu.utils import set_default_server_args
 
             set_default_server_args(self)