Error Deploying Model to Endpoint with initial_instance_count=2

When I attempt to deploy a model to an endpoint with initial_instance_count=1, it works fine. Attempting to deploy the same model to an endpoint with initial_instance_count=2, however, always results in the following error:

UnexpectedStatusException: Error hosting endpoint mm-xxxxxxxx-content-recommender-2024-09-27-18-32-34: Failed. Reason: error: Failed to download model data from URL "s3://sagemaker-us-west-2-xxxxxxxxxxxx/tensorflow-inference-2024-09-27-18-32-34-972/model.tar.gz". Please ensure that an S3 VPC endpoint exists in route table or NAT gateway for the VPC mode and the URL is reachable from within the subnets provided.. Try changing the instance type or reference the troubleshooting page https://docs.aws.amazon.com/sagemaker/latest/dg/async-inference-troubleshooting.html

The deployment code is as follows:

from sagemaker.tensorflow import TensorFlowModel
from sagemaker.tensorflow import TensorFlowPredictor
from time import gmtime, strftime

vpc_config = {
  "Subnets": ["subnet-abc123", "subnet-abc456"], # Need subnets in at least 2 distinct availability zones.
  "SecurityGroupIds": ["sg-abc123"]
}

mm = TensorFlowModel(
  entry_point="inference.py",
  source_dir="multi/code",
  model_data=f"s3://{bucket}/{prefix}/{multi_archive}",
  role=role,
  framework_version="2.16",
  container_log_level=logging.WARNING,
  vpc_config=vpc_config,
  env={ "SAGEMAKER_TFS_DEFAULT_MODEL_NAME": "ranking" })

timestamp_suffix = strftime('%Y-%m-%d-%H-%M-%S', gmtime())
endpoint_name = f"mm-xxxxxxxx-{timestamp_suffix}"

mm_predictor = mm.deploy(
  initial_instance_count=2,
  instance_type="ml.c5.9xlarge",
  endpoint_name=endpoint_name
  )

I tried setting model_data_download_timeout=1200, but it didn't help.