In that case, if you do not require a GPU you can use the agent in services mode (running several tasks in parallel). Otherwise scaling solution are using the AWS autoscaler, or deploying several agents, or using a k8s cluster