Cloud Repatriation or Not?
Does cloud make sense for everyone? When should you run your own infrastructure?
It's complicated.
AWS operating margins range anywhere from 25-30%. And that might continue to be strong, even if infrastructure becomes more commoditized (as AWS moves up the stack into higher margin services).
And cloud infrastructure costs as a percentage of revenue have never been higher for SaaS companies. The bottoms-up model is extremely effective for onboarding companies to cloud – it's extremely cheap to get started (in some cases, AWS will even give you up to $100,000 in credits as a startup). It only begins to become more expensive as a company matures.
A few observations:
Infrastructure companies will be forced to think through this the most. It's no coincidence that Dropbox was one of the first to repatriate its services from the cloud (ironically, Dropbox was enabled by the cloud as one of the first customers of Amazon S3).
AWS will continue to dominate the market for IaaS and certain higher-level services. For every Dropbox, there's a Bank of America that doesn't have core competencies or products around low-level cloud services.
Software continues to get simpler. For example, open-source data center and workload management software continue to make repatriation an easier task.
Egress, among other things, prevents companies from partially outsourcing infrastructure.
Maybe there's a market for low-margin, high-volume infrastructure companies built on cloud.
A few startups hypothesize there's a competitive advantage in bypassing cloud: optimizations that AWS won't or can't make. I don't believe the benefits outweigh the costs in the vast majority of cases.
We're probably a decade out from this being a top-of-mind issue for most companies, if ever.