diff --git a/pytorch_lightning/callbacks/model_checkpoint.py b/pytorch_lightning/callbacks/model_checkpoint.py
index cd307d18bc03a..8ea585db3c4ce 100644
--- a/pytorch_lightning/callbacks/model_checkpoint.py
+++ b/pytorch_lightning/callbacks/model_checkpoint.py
@@ -249,7 +249,7 @@ def state_key(self) -> str:
         )
 
     def on_pretrain_routine_start(self, trainer: "pl.Trainer", pl_module: "pl.LightningModule") -> None:
-        """When pretrain routine starts we build the ckpt dir on the fly."""
+        """When pretrain routine starts we resolve the ckpt dir on the fly."""
         if self._save_on_train_epoch_end is None:
             # if the user runs validation multiple times per training epoch or multiple training epochs without
             # validation, then we run after validation instead of on train epoch end
@@ -600,9 +600,6 @@ def __resolve_ckpt_dir(self, trainer: "pl.Trainer") -> None:
 
         self.dirpath = ckpt_path
 
-        if not trainer.fast_dev_run and trainer.training_type_plugin.should_rank_save_checkpoint:
-            self._fs.makedirs(self.dirpath, exist_ok=True)
-
     def __warn_if_dir_not_empty(self, dirpath: _PATH) -> None:
         if self.save_top_k != 0 and self._fs.isdir(dirpath) and len(self._fs.ls(dirpath)) > 0:
             rank_zero_warn(f"Checkpoint directory {dirpath} exists and is not empty.")
diff --git a/pytorch_lightning/plugins/io/xla_plugin.py b/pytorch_lightning/plugins/io/xla_plugin.py
index c40b6a1ada037..9c2f0088d12e5 100644
--- a/pytorch_lightning/plugins/io/xla_plugin.py
+++ b/pytorch_lightning/plugins/io/xla_plugin.py
@@ -11,11 +11,13 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+import os
 from typing import Any, Dict, Optional
 
 from pytorch_lightning.plugins.io.torch_plugin import TorchCheckpointIO
 from pytorch_lightning.utilities import _OMEGACONF_AVAILABLE, _TPU_AVAILABLE
 from pytorch_lightning.utilities.apply_func import apply_to_collection
+from pytorch_lightning.utilities.cloud_io import get_filesystem
 from pytorch_lightning.utilities.types import _PATH
 
 if _TPU_AVAILABLE:
@@ -36,6 +38,8 @@ def save_checkpoint(self, checkpoint: Dict[str, Any], path: _PATH, storage_optio
             path: write-target path
             storage_options: Optional parameters when saving the model/training states.
         """
+        fs = get_filesystem(path)
+        fs.makedirs(os.path.dirname(path), exist_ok=True)
         # Todo: TypeError: 'mappingproxy' object does not support item assignment
         # Ref: https://github.com/pytorch/xla/issues/2773
         if _OMEGACONF_AVAILABLE:
diff --git a/pytorch_lightning/plugins/training_type/single_tpu.py b/pytorch_lightning/plugins/training_type/single_tpu.py
index 1f15f1cf8d885..011604468e1f5 100644
--- a/pytorch_lightning/plugins/training_type/single_tpu.py
+++ b/pytorch_lightning/plugins/training_type/single_tpu.py
@@ -75,9 +75,6 @@ def pre_dispatch(self, trainer: "pl.Trainer") -> None:
         self.tpu_local_core_rank = xm.get_local_ordinal()
         self.tpu_global_core_rank = xm.get_ordinal()
 
-    def save(self, state_dict: Dict, path: _PATH) -> None:
-        xm.save(state_dict, path)
-
     def save_checkpoint(self, checkpoint: Dict[str, Any], filepath: _PATH) -> None:
         """Save model/training states as a checkpoint file through state-dump and file-write.
 
diff --git a/pytorch_lightning/trainer/trainer.py b/pytorch_lightning/trainer/trainer.py
index bacb4cbbd5475..6805009d745b8 100644
--- a/pytorch_lightning/trainer/trainer.py
+++ b/pytorch_lightning/trainer/trainer.py
@@ -1700,10 +1700,6 @@ def world_size(self) -> int:
         # some training types define a world size
         return getattr(self.training_type_plugin, "world_size", 1)
 
-    @property
-    def should_rank_save_checkpoint(self) -> bool:
-        return self.training_type_plugin.should_rank_save_checkpoint
-
     @property
     def _distrib_type(self) -> _StrategyType:
         return self._accelerator_connector._distrib_type