Your Cloud Bill Is A Tax On Someone Else's Resume
Kubernetes adoption is often a tax on someone else's resume. How solo founders end up with $600 AWS bills — and the one-VPS stack that replaces it.
There’s an insurance company somewhere — real, working, profitable — with 100,000 monthly users and a peak concurrent load of about 5,000.
They spend high six figures a month on Kubernetes.
They employ twenty people to keep it running.
This story surfaced this week in the Hacker News thread on David Crawshaw’s cloud essay, and the comments section turned into a confessional. Engineer after engineer describing the same pattern: cluster adopted, cluster “optimized,” cloud spend doubled, incidents doubled, and somehow the only thing anyone can agree on is that they need to hire a platform engineer.
You don’t. You never did. Your entire application would run on a laptop.
The incentive nobody likes to say out loud
Here’s the quiet part: your DevOps team does not choose infrastructure based on what your application needs.
They choose it based on what their next job will pay for.
Kubernetes on a resume is worth more than Docker Compose on a resume. Terraform on a resume is worth more than “I SSH’d into the box.” Managed EKS on a resume is worth more than “I run a VM.” Every procurement decision in a modern engineering org is being made by someone who, at some level, is also writing the next page of their LinkedIn.
And management, god bless them, trusts the sales and marketing departments of Datadog and AWS and HashiCorp more than they trust their own engineers. So when someone internally says “we could do this on one server,” and someone externally sends a deck titled Scaling Your Platform For The Future, guess which one wins the meeting.
The decision was never technical. You just paid the technical price for it.
Kubernetes is not the villain. The scale is.
Let’s be precise, because “Kubernetes” is doing a lot of work in this essay.
Full enterprise Kubernetes — managed control planes, service meshes, operators for everything, a dedicated platform team, Helm charts nested inside Helm charts like Russian dolls of YAML — that thing was built for Google’s problem. Multi-tenant, multi-region, thousands of services, teams that don’t talk to each other.
If your org does not look like that, you are wearing a costume.
K3s on a single VPS is not the same animal. Docker Compose on a single VPS is not the same animal. Kamal shipping containers to one Debian box is not the same animal. Those are orchestration for people who want one sane way to deploy a container, not a career in platform engineering.
The HN thread is full of engineers who moved from full K8s to one of these simpler setups. The reports are boringly consistent: costs collapsed, incidents dropped, debugging became possible again. Nobody was shocked. Everyone had been waiting for permission to say it.
The solo founder’s version of this trap
You are not the insurance company. You do not have twenty people. You have you, and maybe a contractor, and a credit card that is getting nervous.
And yet — you will read the AWS Well-Architected Framework. You will follow a tutorial that starts with “first, let’s set up your VPC.” You will pay $80/month for a managed database to store 200 rows. You will provision a load balancer in front of one server. You will copy the shape of infrastructure you saw at your day job, because that shape felt legitimate, and you want to feel legitimate too.
This is how solo founders end up with a $600/month AWS bill for an app that has six users.
The shape of legitimacy is the trap. Nobody cares what your infrastructure looks like until you have customers, and once you have customers, “my app runs on one $12 VPS” is a story people love. It’s the opposite of suspicious. It’s proof that the thing works.
What to actually do
One machine until you can’t. One VPS. One Postgres on that VPS. One reverse proxy. Docker Compose or Kamal to deploy. You are allowed to stop here for years.
Scale vertically first. Hetzner will rent you a 48-core EPYC machine with 256 GB of RAM for €199/month. A mid-tier managed Kubernetes cluster on AWS starts at more than that before you’ve run a single pod. Most apps die from bad unit economics, not from running out of CPU.
When you outgrow that — and you might not — K3s on a few boxes gives you orchestration without the org chart. This is the actual sweet spot for a solo operator who needs more than one machine but less than a platform team.
Treat every infrastructure recommendation as a resume artifact until proven otherwise. Ask who benefits if you adopt this. If the answer is “the person telling me to adopt it,” weigh accordingly.
Your cloud bill is a leading indicator of how much time you are spending on things that do not make your product better. Watch it like you watch your weight.
The cloud was supposed to be leverage. For most people, most of the time, it has become the opposite: a recurring invoice for someone else’s credibility.
You are allowed to just run the server.