Why one CTO moved away from the cloud

While market-intelligence firm Gartner predicts a 20.4% yearly growth in public-cloud spending, and an early 2023 survey from the media group TechTarget showed that 46% of organizations would identify as “cloud first,” 37signals CTO David Heinemeier Hansson is very happy with his purchase of giant, old-fashioned servers; he may even consider himself cloud-last.

After tracking costs of cloud services with another classic IT tool—the spreadsheet!—the software company executive determined they’d save more money with a cloud exit and a physical data center entrance. A series of posts from Hansson detail the company’s move from services like AWS to the hefty servers, and a savings they recently calculated at $1 million a year. (They now just have eight petabytes worth of data in an Amazon S3.)

“Getting our applications out of the cloud provided the main celebration for our exit, but seeing the actual spend tumble is the prize,” Hansson wrote in September.

Hansson spoke with IT Brew about cloud’s promise, if it’s living up to the deal, and if there’s a better, on-prem deal out there.

The following interview is edited for length and clarity..

Many orgs consider themselves “cloud-first,” and they’re going to dedicate a lot of IT spending to the cloud. And perhaps they’re feeling they have to move the cloud. What advice do you have for them?

If you don’t care what it costs, I think cloud is fine. I think the entire economy has pivoted on that. At 5.5% interest rates from the Fed, it’s just completely different. Money now has value on its own. Every dollar saved now has tremendous value.

Startups used to be all “cloud, cloud, cloud, because that will allow us to move the fastest.” Now all of them are looking at runway. How do we make sure that the money we’ve raised extends as far as possible, and cloud is the biggest sinkhole they have on their entire P&L. Nothing sucks in more money than cloud, outside of payroll.

What’s nice about owning your own hardware?

The experience of dealing with modern data-center and white glove-service operators is almost indistinguishable from the cloud. We never touch physical hardware. We don’t buy data centers, we rent space in the data centers. We dropship Dell orders, straight up pallets, to those data centers. We never touch the boxes! We never open the boxes, we never see the boxes. There’s a white glove service at the data center that will rack the machine, connect it to power over ethernet, what we see is an IP address appear online.

Is there a drawback?

The only drawback in my estimation of this whole setup is that there’s a time lag of about two weeks from when we put our purchase order in with Dell to when we see the IP address online takes—about two weeks. In the cloud, it takes five minutes. So, if you have those needs that you need five minutes response time: Wonderful. Cloud is so great. We don’t; we never did. We always had two weeks to wait for new hardware to arrive.

Was there a moment that you can think of where you knew it was time to switch and go from cloud to on-prem?

If you look at what you can buy from Dell today, or HP or whoever, versus what you could buy five years ago, it’s dramatic how much more you get for your dollar. The opposite has happened with cloud; cloud has gotten more expensive, in part because managed services have taken over more of it, which has the promise of it being easier. We were never able to realize that. So, we finally just came to the conclusion, you know, that this isn’t adding up.

The only one of the three fundamental premises of the cloud that’s true, and has been true for us, is that it’s faster to procure a lot of servers really quickly. If you need 100 new machines online in five minutes, there’s no alternative to the cloud. We just never need that. We virtually never have needed that. My opinion is that virtually no company needs that.