If you keep up with technology news, you’ve been hearing a lot lately about how enterprises are moving more and more key workloads from their own on-premise data centers to the public cloud. In fact, it happens so much that even the biggest transitions are hardly news anymore.

Man bites dog?

What does make people pay attention, though, is when the opposite happens: When a major product or service moves from running in the public cloud to an on-premise data center. And that’s exactly what happened last week when CNBC broke the news that Facebook plans to move its WhatsApp service from IBM’s SoftLayer public cloud service to its proprietary data centers.

This is indeed a big deal in some ways, but not for the reasons you might think.

Most of the coverage played up the angle that this is a competitive blow for IBM, losing what was reportedly once one of its five biggest cloud customers spending $2 million per month. And, sure, in the short term, losing a high-profile whale like WhatsApp—used by 1.2 billion people around the world—certainly won’t help IBM in its cloud wars with Amazon Web Services, Microsoft Azure and Google Cloud Platform.

But given the overall growth momentum for the public cloud, there are likely plenty of new customers out there to help fill the void at IBM.

WhatsApp isn’t just leaving IBM, it’s leaving the cloud

No, the real significance here isn’t that WhatsApp is leaving IBM SoftLayer, but that the move isn’t to one of IBM’s cloud competitors—to Facebook’s own data centers. From Facebook’s perspective, that’s not surprising, as the company has already migrated its Instagram app, which it bought in 2012, from AWS to its own data centers. Indeed, some observers had wondered what was taking Facebook so long to move WhatsApp, which it acquired in 2014 for a whopping $19 billion.

The bigger issue, then, is that very trend of Facebook acquiring huge cloud-based services and migrating them—sooner or later—to its own data centers.

The WhatsApp move is a further demonstration that in some instances—especially at this kind of extreme web scale—the economics of the public cloud still may not match the advantages of running your own data centers.

Who else is big enough for a move like this?

Conventional wisdom holds—and I agree—that the number of those situations is dwindling as public cloud providers continue to expand and prove themselves in ever larger and more mission-critical implementations. But this move clearly demonstrates that major, big-name exceptions to that trend remain, and that no matter how much the public cloud grows, the biggest enterprises operating the highest-volume services may always have good reasons to operate their own data centers.

Finally, it’s worth noting that SoftLayer ran WhatsApp on bare-metal servers, in search of the best possible performance. And that since Facebook began building its

own data centers in 2010, it’s been a leader of the Open Compute Project, which aims to set open source standards for low-cost data center equipment. At Facebook’s size, its economies of scale are likely similar to those enjoyed by the public cloud providers themselves, which would mean the company can save money running things in house. That simply may not be true for many other enterprises.