the horizontal scaling depends on the spicedb instances being aware of each other, because the consistent hashing you're talking about is what keeps cache hit rates high and performance good. we're running into the same problem right now - we're running in ECS, which doesn't provide a k8s-like service discovery mechanism. theoretically as long as you can provide an XDS endpoint that tells each instance about its neighbors, it doesn't matter where you're running, but I don't know what environments can and do provide that. k8s + the operator is definitely the most straightforward way to implement that piece (ignoring all of the other complexity that k8s brings).