More and more services are meant to be deployed on unreliable hardware. Lose a few nodes? Meh who cares. For ease of patching, I currently always deploy 2+ workers even for simple sites, take one down, patch, test and rejoin the load balancer is far easier than take it down, patch as fast as you can type and everything works right the first try. (Works as long as no dev makes a database migration that breaks backwards compatibility... I hate deploying those changes.)
You still need 1 vmware cluster on proper SAN for the really important stuff like the main HAProxy and Postgres nodes, but for the nth container hosts, nope don't need that expensive vSphere license.
On the other hand, VMWare is kinda squeezed by the crazy amount of cores and RAM modern machines have, they tried charging for vCPU and RAM, but got yelled at so hard they won't try again. Except for charging even more for support, I don't see how they can prevent revenue from cratering from node count reductions and moves to the cloud.