In January, I wrote a sentence on the Magic Pages blog I might now need to take back.
"If I applied the standard of 'never use infrastructure from companies with ethically questionable customers,' I'd have to build my own data centres. And quite honestly, as fun as my home server project was, I would not want to do this on a bigger scale."
Five months later, I'm sitting in front of a spreadsheet doing exactly the maths I said wouldn't want to do. To build a data centre.
Let me explain how that happened, because it's a good example of something I think a lot of small SaaS businesses are about to run into in the next few weeks and months.
A server that cost €70 now costs €452
The single thing that changed was Hetzner. The server and infrastructure provider where all of Magic Pages' servers live. They adjusted their prices – and not by a little.
A dedicated server that I could rent for under €70 a month now starts at €452 a month for a comparable, available spec. Cloud instances went up by 2-3x as well.
So, adding a new Docker Swarm worker, with the RAM Magic Pages actually needs, went from a 10-20% price increase we can swallow (which happened earlier this year) to a potentially catastrophic line item.
I'm not writing this to complain about Hetzner. They've been a great host and the price increase is grandfathered for existing servers (a reader has pointed me towards the origin of the term "grandfathering", so I'll avoid using that in the future) not applied for existing servers. Frankly, the whole industry is dealing with the same RAM-price spike that's driving this. They're not being unreasonable.
But "not unreasonable" and "doesn't change my business" are two very different things.
There's an entire category of small software businesses – indie SaaS, bootstrapped tools, managed-hosting outfits like mine – who assumed cheap European dedicated servers would stay cheap. That assumption was well...optimistic. We all priced our products on these servers.
But it wasn't permanent. It was just stable for a while. A long, long while.
When the numbers underneath your business move, your whole model shifts on top of it. The margin that felt comfortable gets razor thin. The scaling plan that made sense becomes the thing that can bankrupt you.
The angle changed.
The good news: Magic Pages is fine
I want to be clear about where Magic Pages actually stands, because I don't want anyone reading this to worry.
We're fine. Genuinely.
The existing infrastructure is grandfathered kept at the old prices, and our current setup scales comfortably to somewhere around 2,500 customers without me needing to add a single new server. That's a lot of headroom from where we are now − a deliberate choice when setting up the new infrastructure at the end of last year.
Nothing about the day-to-day changes. Your sites don't get slower, your price doesn't go up, nothing breaks.
But we need to adjust the strategic outlook of where Magic Pages is heading.
Because every new server I add lands at Hetzner's new prices. And that's where the maths gets interesting.
Magic Pages initially started on Hetzner Cloud. Cloud was convenient and expensive, and once the business was real enough to care about margins, the numbers stopped making sense. So we moved to dedicated servers. Renting entire machines beat virtual ones.
But now the ground has shifted again. At the new prices, renting a new high-RAM dedicated server, every time we grow, starts to look a lot less obviously correct. And the option I dismissed out of hand in January – owning the hardware – suddenly pans out in a way it simply never did before.
I want to be precise about why, because it's not my nerdy urge to put together a bunch of servers. For that I have my homelab. It's pure math.
When a server you'd had to rent costs €452 a month, and the equivalent machine costs €1,000 - 2,000 euros to buy outright, the payback period collapses from "never" to "months", even when you factor in the costs of power, cooling, spare parts, and so on.
The RAM-price spike that made renting more expensive? You pay that once when you own the hardware. You pay it every single month, forever when you rent it.
I still don't want to build a data centre
Owning hardware and building a data centre are two very different commitments, though.
What I dismissed so far is colocation – or server housing.
You can own your servers and put them in someone else's facility. Power, cooling, connectivity, security, etc. is all handled by people who do that for a living. You get the asset and the economics without the 3am headache because the burglar alarm went off.
So the thing I said I'd "never want to do on a bigger scale" turns out to have a version that's actually sensible. Not building a room. Just owning the machines that go in one.
I haven't pulled the trigger on it because, even with this new math, there are real questions left. I need to find the right facility, procure the servers (which is a huge upfront investment), explore logistics of getting the servers from here to the data centre, figure out whether the operational overhead is worth it at our size. And I'm still working through them.
But I wanted to write this down now, while it's still a decision and not yet a done deal, because the reasoning is the interesting part, in my eyes – more interesting than whatever I eventually choose.
What I learned since yesterday is that the numbers underneath your business are not constants. They're someone else's pricing decision, and they have their own costs, their own pressures, and no obligation to keep your margins intact.
It's worth knowing, right now, which assumptions in your business are load-bearing. Which single number, if it doubled (or 6x 🙃), would change everything?
For Magic Pages it is the price of a dedicated server. For you it might be an API, a payment processor's cut, cloud fees, or a framework you don't control.
When you know that number, check your blind spots. You might be very, very surprised.