diff --git a/tensorflow/compiler/xla/python/tpu_driver/client/tpu_client.py b/tensorflow/compiler/xla/python/tpu_driver/client/tpu_client.py
index f86a7869ec4..43c0d1a40c3 100644
--- a/tensorflow/compiler/xla/python/tpu_driver/client/tpu_client.py
+++ b/tensorflow/compiler/xla/python/tpu_driver/client/tpu_client.py
@@ -53,10 +53,10 @@ class TpuBackend(xla_client.Backend):
     if worker == 'local' or 'local://' in worker:
       # We usually want to cache for local backends to prevent double
       # initialization, except where `force` == True.
-      if force:
-        return TpuBackend(_tpu_client.TpuClient.Get(worker))
       if worker == 'local':
         worker = 'local://'
+      if force:
+        return TpuBackend(_tpu_client.TpuClient.Get(worker))
       if TpuBackend._local_backend is None:
         logging.info('Starting the local TPU driver.')
         TpuBackend._local_backend = TpuBackend(