diff --git a/python/sglang/srt/platforms/npu.py b/python/sglang/srt/platforms/npu.py
index c1382ff76..d01220efb 100644
--- a/python/sglang/srt/platforms/npu.py
+++ b/python/sglang/srt/platforms/npu.py
@@ -56,9 +56,7 @@ class NPUDeviceMixin(DeviceMixin):
         return torch.npu.mem_get_info(device_id)
 
     def is_pin_memory_available(self, device=None) -> bool:
-        if device is not None and str(device) == "cpu":
-            return False
-        return True
+        return False
 
     @classmethod
     def seed_everything(cls, seed: int | None = None) -> None:
diff --git a/test/registered/unit/platforms/test_platform_interface.py b/test/registered/unit/platforms/test_platform_interface.py
index 5614dd7f0..cd9548672 100644
--- a/test/registered/unit/platforms/test_platform_interface.py
+++ b/test/registered/unit/platforms/test_platform_interface.py
@@ -381,10 +381,13 @@ class TestNpuDeviceMixin(CustomTestCase):
             mock_npu.synchronize.assert_called_once()
 
     def test_pin_memory_available_for_npu_targets(self):
+        # Pinned memory stays disabled on NPU: torch_npu's pinned-memory +
+        # non_blocking H2D path is not verified against CANN (see
+        # NPUSRTPlatform.is_pin_memory_available).
         base = NPUSRTPlatform()
-        self.assertTrue(base.is_pin_memory_available())
-        self.assertTrue(base.is_pin_memory_available(device="npu"))
-        self.assertTrue(base.is_pin_memory_available(device=torch.device("npu", 0)))
+        self.assertFalse(base.is_pin_memory_available())
+        self.assertFalse(base.is_pin_memory_available(device="npu"))
+        self.assertFalse(base.is_pin_memory_available(device=torch.device("npu", 0)))
         self.assertFalse(base.is_pin_memory_available(device="cpu"))
         self.assertFalse(base.is_pin_memory_available(device=torch.device("cpu")))