Migrating to a public cloud infrastructure such as AWS, Google or Azure has become increasingly popular these days. The ability to effectively out-source/off-load the cost of maintaining and upgrading an IT infrastructure provides tremendous benefit to an organization. Plus, the speed by which IT admins can migrate various work-loads to public cloud, without having to opening a server rack, or turn a screwdriver, makes public cloud migration seem like a "no-brainer".
However, enterprises are quickly realizing two very valid risks associated with migrating 100% of their IT infrastructure to public cloud.
The first risk pertains to today's AI-intensive work-loads. AI-intensive work-loads tend to perform better on GPU-centric infrastructure. But when clustered GPU's are required to run AI-intensive work-loads, and those clustered GPUs are in the cloud, the cost of migrating and scaling these work-loads can become cost prohibitive.
More and more companies are realizing that the more dynamic their compute work-loads - especially GPU-centric workloads - the less sense it makes to migrate 100% of their IT infrastructure to the public cloud.
Secondly, companies are discovering that migrating FROM the cloud is far more difficult than migrating TO public cloud. When migrating OFF the cloud, enterprises are left with an "iceberg" of data that they then have to sort through.
A number of case studies exist re: Fortune 500 companies who have chosen to migrate off the cloud, after having discovered the cloud provided neither the flexibility nor scalability they in fact needed.