The correct choice is By offloading provisioning, patching, and auto scaling of the underlying infrastructure to managed generative AI services. Services such as Amazon Bedrock and Amazon SageMaker handle the infrastructure so the team can focus on data preparation, prompt design, and model quality rather than managing servers.

Managed generative AI services provide built in provisioning, automatic scaling, monitoring, and patching for compute and model hosting. This reduces operational overhead and speeds time to value because the team does not need to maintain GPU clusters or the orchestration layers that serve models.

By focusing the service on manual data labeling rather than model training and deployment is incorrect because manual labeling does not remove the burden of provisioning and maintaining training and inference infrastructure and it does not cover hosting workflows.

By using AWS Outposts to run all model workloads on-premises under company management is incorrect because running on premises with Outposts brings hardware operations and patching into the team responsibility and usually increases administration instead of minimizing it.

By requiring the team to operate and patch its own GPU fleet for training and inference is incorrect because operating a dedicated GPU fleet adds significant provisioning, patching, and scaling work which conflicts with the goal to reduce operational effort.

Full AWS Practitioner Certification Question