In 2023 37 Signals left the Cloud. In the process they saved upwards of $2 Million / year.

In this post I'm highlighting a few learnings I found interesting / insightful.

The Cloud is not Magic - I think a lot of people think the Cloud is magic. They pay money and it does the thing. But at the end of the day it's all just servers running software. The difference with Cloud vs on-prem is you're renting a server you cannot see which is good in the sense you don't have to physically plug it in but everything else is on you.

The Cloud is not more efficient - Okay so we live in a capitalist society so it should be clear that everyone is trying to make money off of their transactions. The Cloud is no different. They have access to the same servers we do, they just upcharge for renting them. All those cool serverless functions, world-distributed DBs, etc are an awesome idea but the more we get away from the base server the more it's going to cost and, often, the worse it's going to perform per unit of compute.

For most cos this doesn't matter - I still like and use the Cloud for all my sites. For small-medium cos the Cloud makes sense because you likely aren't paying that much to rent these servers compared to your revenue and you're not doing anything too complicated. So use the Cloud to quickly spin up a few dozen monolith replicas, horizontally replicated. Monoliths scale and will take you far.

But when efficiency does matter, get back to basics - At medium and large scale when you've presumably found product market fit and are now trying to optimize for delivery and efficiency, going back to basics is likely the best move. Over the years of iteration and building there's likely a lot of cruft that's built up - bad system designs, tons of wasted storage / compute, and likely lots of dead systems that were built for features that never launched. In this phase, moving out of the renting market often makes sense - you're ready to settle down and build a home base. You can likely do that way more efficiently on your own hardware.

Next

I'm particularly excited by these moves to on-prem because I think it will start building a small industry around cheaper cloud basics. The only things I really want are a server, db, cache, and queue that I can easily spin up, scale, and manage. If you give me that for cheaper than these big clouds, you've disrupted the whole industry. Hopefully with the hype around cloud exits this will come to fruition.

If you liked this, you might also like:

• Servers as Factories - a simple metaphor for thinking about web servers
• Why I'm moving from SvelteKit to F#
• Open Source Software is Unsustainable