Lambda has huge potential for very defined work load. In this case, I do not get it. As you mentioned the idea of having 3 regions with 3000 cores? Are you doing ML on K8s? Another aspect is the caching with CDN and internally, I do not get that either.