There was a time when cloud was significantly cheaper then owning.
I'd expect that there are people who moved to the cloud then, and over time started using services offered by their cloud provider (e.g., load balancers, secret management, databases, storage, backup) instead of running those services themselves on virtual machines, and now even if it would be cheaper to run everything on owned servers they find it would be too much effort to add all those services back to their own servers.
The cloud wasn’t about cheap, it was about fast. If you’re VC funded, time is everything, and developer velocity above all else to hyperscale and exit. That time has passed (ZIRP), and the public cloud margin just doesn’t make sense when you can own and operate (their margin is your opportunity) on prem with similar cloud primitives around storage and compute.
Elasticity is a component, but has always been from a batch job bin packing scheduling perspective, not much new there. Before k8s and Nomad, there was Globus.org.
(Infra/DevOps in a previous life at a unicorn, large worker cluster for a physics experiment prior, etc; what is old is a new again, you’re just riding hype cycle waves from junior to retirement [mainframe->COTS on prem->cloud->on prem cloud, and so on])
I'd expect that there are people who moved to the cloud then, and over time started using services offered by their cloud provider (e.g., load balancers, secret management, databases, storage, backup) instead of running those services themselves on virtual machines, and now even if it would be cheaper to run everything on owned servers they find it would be too much effort to add all those services back to their own servers.