Why devs prefer using VPS vs. Cloudflare or Supabase?
Use Cloudflare or Supabase for speed of launch and zero server management. Use a VPS or your own cluster for ownership of the source code, predictable costs, and the ability to move your app without rebuilding it from scratch.
Last updated
Use Cloudflare or Supabase for speed of launch and zero server management. Use a VPS or your own cluster for ownership of the source code, predictable costs, and the ability to move your app without rebuilding it from scratch.
What are the real limits of BaaS and Serverless?
The primary trade-off in platforms like Supabase or Cloudflare is the "lock-in" tax. These tools are designed to make the start effortless by hiding the infrastructure. You don't manage a database or a routing table. You use their proprietary APIs and dashboard. This works until you hit a limit that requires a higher tier or a feature that the platform doesn't support. Because your app is built specifically for their ecosystem, moving it means rewriting your persistence layer and your authentication logic.
Billing is the second limit. Serverless pricing often starts low but becomes unpredictable as you scale. You are paying for "credits" or "executions" rather than raw compute. For a solo founder, a sudden spike in traffic or a bug in a loop can lead to a bill that exceeds the monthly cost of a dedicated server. You are renting a sandbox, and the landlord controls the price of the air you breathe.
Why is a VPS or K3s better for ownership?
A VPS or a Kubernetes cluster (like K3s) gives you a raw slice of hardware. You aren't using a "platform" so much as you are using a machine. Your application is packaged as a Docker image and deployed via Helm. This means the source code is a real repository, not a proprietary configuration file in a cloud console. If your provider raises prices or goes under, you move your Docker image and your Postgres data to any other machine that runs Linux.
The cost is fixed. A VPS costs a set amount per month regardless of how many API calls your app makes. For those who want to drive costs toward zero, you can join your own hardware to a cluster. A machine at home becomes a node in the cluster over a private network, allowing you to host your own production workloads on hardware you already own.
What are the risks of managing your own infra?
When you move away from managed services, you inherit the "failure domain." In a managed environment, the provider handles the major version upgrades of your database. On your own infrastructure, a simple image tag change can break everything. For example, moving a Postgres image tag from 15 to 16 will cause the container to crash-loop because Postgres does not upgrade the data directory across major versions on start. You have to perform a deliberate migration.
Storage is another point of failure. If you configure storage replicas on a single node, the replication traffic can starve the control plane. We saw this when Longhorn was configured with multiple replicas on one node. The disk contention starved etcd until the cluster stopped responding. In a managed cloud, these details are hidden, but in your own cluster, they are your responsibility.
When is a hosted builder the wrong choice?
A hosted AI builder is the wrong choice if you intend to build a long-term business that you want to own. Many "vibe-coding" tools lock you into their specific cloud. If you want to learn what happens after your lovable plan, you will find that the path to ownership is often a dead end. You are paying for the convenience of not knowing how your app is deployed, but that ignorance becomes a liability the moment you want to scale or migrate.
If you only need a landing page and never want to see a terminal or a config file, stay with the hosted builders. They are faster for things that don't need to last. But for a product that requires a real database and predictable scaling, the machinery of a cluster is a necessary investment.
my-app.engineer allows you to describe an app in chat, but instead of hosting it in a proprietary sandbox, it deploys it as a Helm release onto a K3s cluster that you own. You can see your deployments and manage your tools from a single interface, but the underlying infrastructure remains standard Kubernetes, Docker, and Postgres.