diff --git a/dask_cloudprovider/cloudprovider.yaml b/dask_cloudprovider/cloudprovider.yaml
index b8448af6..fe6c8740 100755
--- a/dask_cloudprovider/cloudprovider.yaml
+++ b/dask_cloudprovider/cloudprovider.yaml
@@ -105,12 +105,12 @@ cloudprovider:
     scheduler_machine_type: "n1-standard-1" # size of the machine type to use for the scheduler
     worker_machine_type: "n1-standard-1" # size of the machine type to use for all workers
     filesystem_size: 50 # amount in GBs of hard drive space to allocate
-    ngpus: "" # number of GPUs to use. If provided, will be used for both scheduler and worker
-    gpu_type: "" # type of gpus to use. (e.g. 'nvidia-tesla-t4'). You can view the possible values through ``gcloud compute accelerator-types list``. If provided, will be used for both scheduler and worker
-    scheduler_ngpus: "" # number of GPUs to use on scheduler
-    scheduler_gpu_type: "" # type of gpus to use. (e.g. 'nvidia-tesla-t4'). You can view the possible values through ``gcloud compute accelerator-types list``.
-    worker_ngpus: "" # number of GPUs to use on worker
-    worker_gpu_type: "" # type of gpus to use. (e.g. 'nvidia-tesla-t4'). You can view the possible values through ``gcloud compute accelerator-types list``.
+    ngpus: null # number of GPUs to use. If provided, will be used for both scheduler and worker
+    gpu_type: null # type of gpus to use. (e.g. 'nvidia-tesla-t4'). You can view the possible values through ``gcloud compute accelerator-types list``. If provided, will be used for both scheduler and worker
+    scheduler_ngpus: null # number of GPUs to use on scheduler
+    scheduler_gpu_type: null # type of gpus to use. (e.g. 'nvidia-tesla-t4'). You can view the possible values through ``gcloud compute accelerator-types list``.
+    worker_ngpus: null # number of GPUs to use on worker
+    worker_gpu_type: null # type of gpus to use. (e.g. 'nvidia-tesla-t4'). You can view the possible values through ``gcloud compute accelerator-types list``.
     disk_type: "pd-standard" # type of disk to use: pd-standard, pd-ssd
     docker_image: "daskdev/dask:latest" # docker image to use
     auto_shutdown: true # Shutdown instances automatically if the scheduler or worker services time out.
diff --git a/dask_cloudprovider/gcp/instances.py b/dask_cloudprovider/gcp/instances.py
index f0c308a2..c39a1eae 100644
--- a/dask_cloudprovider/gcp/instances.py
+++ b/dask_cloudprovider/gcp/instances.py
@@ -667,12 +667,12 @@ def __init__(
             self.scheduler_ngpus = (
                 scheduler_ngpus
                 if scheduler_ngpus is not None
-                else self.config.get("scheduler_ngpus", 0)
+                else self.config.get("scheduler_ngpus") or 0
             )
             self.worker_ngpus = (
                 worker_ngpus
                 if worker_ngpus is not None
-                else self.config.get("worker_ngpus", 0)
+                else self.config.get("worker_ngpus") or 0
             )
         else:
             if scheduler_ngpus is not None or worker_ngpus is not None:
diff --git a/dask_cloudprovider/gcp/tests/test_gcp.py b/dask_cloudprovider/gcp/tests/test_gcp.py
index bf96e9af..039923f4 100644
--- a/dask_cloudprovider/gcp/tests/test_gcp.py
+++ b/dask_cloudprovider/gcp/tests/test_gcp.py
@@ -53,6 +53,17 @@ async def test_init():
     assert cluster.status == Status.created
 
 
+@pytest.mark.asyncio
+async def test_init_gpu_config_defaults():
+    """Regression test for https://github.com/dask/dask-cloudprovider/pull/479."""
+    skip_without_credentials()
+
+    cluster = GCPCluster(asynchronous=True)
+    assert cluster.ngpus is None
+    assert cluster.scheduler_ngpus == 0
+    assert cluster.worker_ngpus == 0
+
+
 @pytest.mark.asyncio
 async def test_get_cloud_init():
     skip_without_credentials()