The cloud is very easy and cheap to migrate TO, but once you get to a certain scale, the cloud is not very cost effective.
There are a couple of hidden cloud costs - post migration - that cloud customers aren't aware of until after the fact. It might be inexpensive for you to upload your data to the cloud, but what about when you want your data back? What is the cost to pull your data back down off of public cloud?
That's one aspect of it. Secondly, we need to think about next generation workloads. If you're doing a standard workload that uses a simple X86 processor with a little bit of storage and a little bit of networking, those types of instances can be available on the cloud in a relatively cost-effective manner.When we start thinking about next generation workloads - workloads that leverage GPUs, workloads that leverage FPGAs, workloads that leverage high connectivity of six or eight, 16 GPUs inside of a single cluster.
Now, those types of machines are not very cost effective to rent or lease on the cloud.
So what we're realizing is that for certain workloads, it's actually much more cost-efficient to do it on-prem than it is to do it in-cloud. In summary: the hidden costs of migrating to the cloud are: 1) the expense associated to getting off of the cloud and bringing your data back home, 2) scalable costs associated with dynamic workloads.
You need to ask yourself what the workload is that you actually need to deploy Some workloads are well-suited for cloud environments whereas other workloads - next generation GPU centric workloads - are going to be very expensive to run in the public cloud.