You probably don't need Amazon Web Services or Google Cloud Platform for your business...
We're in 2026 as I'm writing these lines, and the trend right now is Next.js.
Every indie developer wants to build a SaaS, so they launch tons of projects, forget to do any marketing, and move on to the next one.
And? It's a financial black hole.
For each new project, they deploy on Vercel because "it's simple" (and it's true, it is simple).
But here's the thing: for every new project, even without traffic, even without visibility, you still have to pay Vercel.
Whether it's Vercel or any other PaaS for that matter — in France we have Clever Cloud, which works really well — it's just not suited to what I do. I'd be paying 26 euros for every new project, and right now I have 18 active projects (domains).
That would cost me 468 euros per month for projects that, for the most part, are free.
It's not viable. The only ones lining their pockets in this story are the platforms themselves.
And... imagine the stress when traffic starts ramping up: "How much am I going to have to pay?"
As for me, since 2020 I've been running on a bare-metal server infrastructure.
It's great and it's worked well so far. In 5 years I've had one crash — an issue when restarting the machine, the network port didn't come back up properly.
It's getting more and more traffic, and as you may know, it has a free version.
So what do I do — kill the free version and hand a pile of cash to Amazon?
No, that's not really my style... it's not in my DNA.
I mentioned Coolify earlier, but after some analysis and stepping back, I don't like it.
Coolify is yet another big web interface on top, and it can have CVEs
The more exposed I am online, the more security becomes a real concern. I receive phishing emails I'd describe as "highly sophisticated" — they genuinely appear to come from Meta, Apple, or Google (the stores, TestFlight, etc.).
To scale, I'm going to stick with the same type of infrastructure I already have: servers. And to handle deployment — the "orchestrator" — I'll use Kamal.
That will spare me all the complexity of Kubernetes, and it'll let me deploy.
The single application runs on IIS and hosts around 200 sites. It handles more than 6,000 requests per second and 2 billion page views per month, with a render time of about 12 milliseconds per page.
The nine IIS web servers are deployed across nine virtual machines, each with 64 GB of memory. Average CPU load is only 5%, peaking at 12%, leaving plenty of headroom.
HAProxy distributes traffic across the nine web servers and automatically detects servers that stop responding.
On top of that: one primary SQL server (with a hot standby and 1.5 TB of RAM), two Redis servers (master/replica), three servers for the tag engine, and three Elasticsearch servers handling 34 million searches a day.
Well... what can I say. We're talking about a site that was in the global top 50 (back then — not anymore, because of AI, of course).
