Times are tough, but that doesn’t mean chief information officers aspire to make them tougher. Yes, I’m talking about cloud repatriation, or the practice of moving workloads out of the cloud and back into private data centers.

But, no, you shouldn’t do it. For every 37signals that moves a workload back on-premises, there are tens of thousands more enterprises going the other way, and for good reason. In an uncertain economy, the last thing you want is to lock yourself into very inelastic hardware purchases and software licenses.

There are best practices, and there are worst practices. Cloud repatriation is the latter.

But what about Hey/Basecamp?

You can count on 37signals co-founder David Heinemeier Hansson to make controversial claims, some of which are true. Others, however, are controversial simply to make a stir. Most recently he spent nearly 1,000 words telling the world why his company is moving out of the cloud and back into an on-premises datacenter. His tl;dr? “Renting computers is (mostly) a bad deal for medium-sized companies like ours with stable growth.” 

To get there, he argues that cloud is good for small companies with simple applications, or for those with “highly irregular” loads — that is, “wild swings or towering peaks in usage.” For anyone operating during a recession (everyone reading this!) or even during boom times, you live in that second bucket, even if your load may not meet DHH’s bar for “highly irregular.” More on that in a minute.

Where DHH’s argument really falls apart is when he moves way off-script to denounce “entrenched cloud interests.” “It strikes me as downright tragic that this decentralized wonder of the world is now largely operating on computers owned by a handful of mega corporations,” he says.

OK… fine. But there are many reasons enterprises increasingly entrust their applications to the cloud providers, reasons that may well not apply to 37Signals’ Basecamp or Hey applications. 

Let’s be clear, though: Your workloads very likely don’t look like 37signals’ workloads. And for everyone, betting on predictability in the midst of a recession seems like poor policy indeed.

Cloud and predictability

“Every time I read about some service repatriating from the cloud it’s always (a) very stable consumption (b) thin margins (c) computer cost is the majority of cost of goods,” notes Modal Labs founder Erik Bernhardsson. Guess what? This almost certainly doesn’t describe your company. 

It’s also likely, even if you can make the math work, that your team likely can’t operate the infrastructure as well as the cloud providers. (DHH believes his team can. Bravo!) You employ great people but guess what? They’re not better than Amazon Web Services, Google Cloud or Microsoft at operating infrastructure, managing security and the like. They’re simply not, nor do they need to be. AWS calls it “taking the undifferentiated heavy lifting” off the shoulders of customers so that they can innovate in other areas. 

Especially in our current recession, this seems like a much smarter way to spend IT money than on re-creating a little AWS within your data center. Furthermore, in an uncertain economic environment, why would you want to lock yourself into inelastic hardware purchases and software licenses?

Much of DHH’s argument comes down to the predictability of workloads, but three months ago no one was predicting a global slowdown. It seems like madness to think it’s better to have spent a million dollars on servers and software with the very real possibility that what you bought for today won’t be what your business is actually using tomorrow (whether because customer demand is up or down).

In sum, sure, you’ll hear of companies moving workloads from the cloud back to private data centers. But before you start down that path, it’s worth considering all the flexibility and services you’ll abandon to save a few theoretical dollars. And, surprise! You likely won’t save those dollars.

Matt Asay is vice president of partner marketing at MongoDB Inc. He wrote this article for SiliconANGLE.

Image: DALL-E