Why I Still Run My Own Infrastructure
I have spent enough time in infrastructure and application support to know that cloud is brilliant at some things and a very expensive way to do others. If you need unpredictable scale, fast experimentation, managed services, and the ability to spin environments up and down constantly, public cloud is completely rational. If your workload is stable, always on, and fairly well understood, the economics can look very different.
That is the category I mostly live in. I run production websites, business applications, and scheduled analytics jobs. I am not trying to burst to millions of users overnight. I am trying to run dependable systems at a predictable monthly cost.
What the AWS Baseline Looks Like Right Now
Using Amazon's current public EC2 price list for US East (N. Virginia) on April 8, 2026, a simple always-on three-instance baseline looks like this:
- t3.large: $0.0832/hour
- m6i.4xlarge: $0.768/hour
- r6i.xlarge: $0.252/hour
If you leave those running for a typical 730-hour month, the compute alone comes to roughly $805.34/month. That is before EBS volumes, snapshots, public IP charges, load balancing, and internet transfer. AWS's own On-Demand pricing page is very clear that EC2 is only one part of the bill, and that transfer and adjacent services are priced separately.
That does not mean AWS is doing anything wrong. It means you are paying for flexibility, service integration, and the ability to compose infrastructure as rented building blocks rather than one fixed machine.
The key point is not that AWS is "too expensive" in the abstract. The key point is that always-on workloads can end up paying a premium for flexibility they barely use.
Why Dedicated Infrastructure Feels Different
Dedicated providers such as Hetzner, OVHcloud, and Scaleway sell a very different proposition: fixed monthly hardware, fixed bandwidth policies, and far fewer moving billing parts. OVHcloud's current UK bare-metal ranges, for example, advertise fixed monthly server tiers rather than a stack of separate compute, storage, balancing, and transfer line items.
That billing model matters. When I rent one solid box and run multiple services on it, I know roughly what the month will cost before I start. I am not doing mental arithmetic about per-hour instance rates, per-gigabyte storage, snapshots, data transfer, or whether a quiet month and a busy month will look radically different on the invoice.
For my sort of workload, that predictability is worth a lot. The savings are not theoretical either. Once your cloud baseline is several hundred dollars per month before extras, the gap to a self-hosted or fixed-price dedicated setup can easily become thousands per year.
What Makes Self-Hosting Viable
Self-hosting only works if you actually know how to operate systems. That part matters more than the monthly price. You need to be comfortable with Linux, backups, monitoring, TLS, service restarts, patching, failure recovery, and the general reality that there is no managed-service fairy tidying up behind you.
That is precisely why I am comfortable doing it. I do not have to imagine what happens when a disk fills up, a worker queue jams, a reverse proxy is misconfigured, or a deployment goes sideways. I have seen all of that in production. That operational knowledge is what turns self-hosting from a false economy into a sensible option.
I still outsource the things I do not want to run myself. Mail is the obvious example. Deliverability, spam handling, reputation, and compliance are a whole discipline in their own right. I am happy to pay for that to be somebody else's problem.
What You Give Up
The trade-offs are real. If the hardware fails, you are dealing with a provider ticket instead of abstract cloud redundancy. If traffic spikes unexpectedly, you do not magically scale out with a toggle. If your team does not have operations skills, the cheap server can become an expensive mistake very quickly.
That is why I do not treat self-hosting as a universal recommendation. It is a recommendation for predictable workloads plus operational competence. If you have both, the economics can be excellent. If you do not, managed cloud is usually worth the premium.
Bottom Line
I still use cloud services when they are the right tool. But for the kind of steady, known, always-on work I run, self-hosting remains one of the most financially sensible choices I can make.
AWS's own current list prices make the shape of the problem obvious: even a modest three-instance baseline is already around $805/month before the rest of the infrastructure bill appears. That is exactly why I keep choosing fixed-cost servers for stable workloads. The operational burden is real, but so are the savings.
Cover image sources: AWS EC2 On-Demand Pricing, Hetzner Dedicated Root Servers, and OVHcloud Bare Metal pages, captured on April 8, 2026.
