I’m proud of the work I’ve done in and on data centers. I can talk about data centers all day. You already know that all of your favorite cloud services run in data centers somewhere. But why would you ever want to deal with running one of your own?

I’d say there are four broad reasons why operating a service from a local data center is a good idea, especially if you deal with scientific instrumentation and associated data flows, or have life-critical services as part of your environment.

1. The service requires very high bandwidth. If you have a few hundred devices generating terabytes of large-file data per day, this could be you.
2. The service requires very low latency. VDI works surprisingly well in the cloud, but latency accumulates, and people like it when the computer feels fast. Bonus observation: if you have terabyte of little tiny files, I bet you have a latency problem more than a bandwidth problem when you want to move those files around.
3. The service has a tight integration with specific local devices or components. Maybe a nurse call system, or facilities management system, or some kind of specialized networking to a lab instrument or processing facility.
4. The service has specific security and/or regulatory constraints. Many could be merely perceived, but some are real and still exist, I promise. Cloud services can and do provide excellent frameworks for security monitoring and compliance, but these constraints can still be important to the business.

I’m not saying a local data center is always required for workloads like these. Running a data center is hard and expensive, and so is running some of these workloads in the first place. Cheap things are cheap in the cloud, but some things aren’t cheap to do anywhere.

If you take a service through the above list and find you hit at least one of them, it might mean that you actually “should” run it locally, and it’s worth thinking about.