diff --git a/vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py b/vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py
index 17f5be76ce40..c306eeb5aa7a 100644
--- a/vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py
+++ b/vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py
@@ -708,8 +708,6 @@ def register_kv_caches(self, kv_caches: dict[str, torch.Tensor]):
         caches_data = []
         # With hybrid allocator, layers can share a kv cache tensor
         seen_base_addresses = []
-        xfer_buffers = (self.host_xfer_buffers
-                        if self.use_host_buffer else kv_caches)
 
         # Note(tms): I modified this from the original region setup code.
         # K and V are now in different regions. Advantage is that we can