Cloud computing is about centralization, the ability to push workloads that run on expensive on-premises systems to cheaper systems on public clouds.
This centralization lets us share resources with other tenants to reduce costs, thus increasing overall efficiency. Centralization provides a single location for data that may have once existed in many enterprise silos. Plus, it lets you centrally manage processes and systems with a single set of tools.
Essentially, we're moving to cloud-based platforms to take back control of our systems and do a much better job of managing those systems centrally. The cloud's cost advantage is really a bonus; the cloud would be a better option even if its costs were the same as the traditional on-premises data center.
You can think of this centralization as Phase 1 of the great cloud migration: getting all those apps and data out of their datacenters and client silos into a single location, the cloud.
Despite all the advantages of centralization, the cloud will likely shift to a distributed approach over the next decade. It will move the data processing back to the ultimate consumers of those processes and data. That's Phase 2 of the great cloud migration: re-creating the notion of client/server computing, with the cloud being the server and the workloads (but not their applications or data) moving back to the clients.
This is not a repudiation of centralization or a shift from centralization. Yes, workloads will move closer to those who use them. But the workloads will remain centrally managed and controlled, even if they execute across a widely distributed architecture. They remain logically centralized whether or not their bits and processing are physically distributed.
The cloud platform automatically distributes the workloads on private cloud instances, which are tightly coupled with the centralized public clouds. To those who manage and use the systems, the workloads are in the public cloud, but deployed in a hybridlike distributed architecture.
Current technology can already do this. Ironically, once we move to a centralized cloud-based architecture, all possibilities are open -- including putting the workloads back on-premises, minus the silos. IT is a funny world.