A team of data scientists needs to perform complex machine learning computations periodically, but their workload does not justify maintaining a dedicated cluster 24/7. Which solution would be the most cost-effective while meeting the team's computational requirements?