Skip to content

← Blog

Your App Works. Where Should You Host It Now?

November 12, 2026 · R. B. Atai

After vibe coding, hosting choices start to matter: cloud bills grow, small VPS machines get painful, and shared fixed infrastructure can become more rational for early projects.

Vibe coding has shortened the path from an idea to a working application. But once the first deployment succeeds, a different kind of work begins: the database must retain its data, services need updates, someone must notice failures, and costs have to stay reasonable.

If your application already works and needs a permanent home, message me through the chat on this site (bottom-right corner). Tell me what the application does, where it runs now, whether it has users, and what the infrastructure currently costs. You do not need an architecture diagram. We can first determine whether the existing setup can be optimized or whether moving it makes sense.

A rack of servers in a server room

A Free Start Does Not Stay Free for Long

My projects started spontaneously. Yet another attempt to build an application finally worked, and I needed somewhere to host it. I chose Google Cloud Run: deploying a container was quick, there was no server to manage, and a small application stayed within the free tier.

This is close to an ideal setup for a first launch. You do not need to buy capacity in advance or predict future traffic. The platform starts the instances it needs and can scale to zero when there are no requests.

Then the application starts living. Builds and deployments multiply. It needs a database with persistent storage. One application becomes several services. Built-in AI is added, and then users arrive.

No single component looks catastrophically expensive. But the total bill combines compute, databases, storage, builds, the container registry, traffic, and external APIs. In my case, a few dozen dollars quickly became more than $100 — my psychological limit for a side project that was not yet generating comparable revenue.

The main problem was not the $100 mark itself. Infrastructure costs and the number of projects were growing faster than revenue.

A Cloud Budget Is Not a Precise Spending Limit

Public cloud carries an uncomfortable risk for an early-stage project: an error in code, configuration, or scaling can trigger uncontrolled resource consumption. The next bill can be not just higher than expected, but several times higher.

Google Cloud lets you configure budgets and alerts. However, a regular budget alert only tells you that a threshold has been reached; it does not stop services by itself.1 Cloud Run also supports spend caps that pause new usage after the configured amount is reached.2

Even a spend cap does not turn the cloud into a prepaid card. Google explicitly warns that enforcement is not instantaneous, overages caused by reporting latency are billed as usual, and in-flight requests continue to completion. A cap also covers one eligible service in one project, while persistent resources can continue accruing costs.

For an established business, this is a manageable risk. For a small project without stable revenue, the possibility of an unexpected four-figure bill changes how attractive cloud elasticity looks.

Why a Small VPS Does Not Always Solve the Problem

The next obvious option is a VPS with a fixed monthly price. For one small API, a website, and a database, it can be a rational choice.

Problems begin when there are several projects. A virtual machine with 2 GB of memory quickly turns operations into a constant fight with the OOM killer. You have to decide which processes can coexist, watch disk usage, free memory, and hope the next build does not displace a running service.

You can rent a larger VPS, but then the savings become less obvious. If every application needs a separate machine, separate backups, and spare capacity, the fixed bill once again grows with the number of projects.

Your own time also has a price. A cheap virtual machine stops being cheap when it requires manual intervention every week.

What Changed with Rented Servers

For my portfolio of projects, the solution was to rent physical servers and run Kubernetes on them. There was enough capacity for development infrastructure and all production applications, while monthly costs became predictable.

The economics changed. Instead of paying for every small service separately, I pay for a shared pool of resources. A new project can use spare cluster capacity without creating another standalone bill for a minimal database, container, or virtual machine.

Kubernetes is not there for fashion. It provides one way to deploy applications, separate their resources, update containers, and move workloads between nodes. A container-based architecture also keeps the path open to a hybrid model or a return to public cloud without rewriting the application.

But servers with Kubernetes are not a free cloud. The cluster must be updated and monitored. You need backups, restore testing, certificate and secret management, disk monitoring, and a clear procedure for node failures. The setup took substantial work, but infrastructure is my field. A team without that experience must count this work alongside the cost of the hardware.

How to Choose Hosting for an Early-Stage Project

There is no universally cheap option. It is more useful to look at the whole application and at who will operate it after release than at the price of a single container.

A managed platform or public cloud is a good fit when:

  • you have one application and its architecture is still changing;
  • traffic is unpredictable or highly variable;
  • the team does not want to manage an operating system or cluster;
  • launch speed matters more than the lowest possible monthly price.

One VPS can be enough when:

  • there are only one or two simple services;
  • the load is understood and fits comfortably on one machine;
  • a basic, partly manual operating model is acceptable;
  • someone on the team owns updates and backups.

Shared fixed infrastructure starts to make sense when:

  • you already have several applications or services;
  • their combined cloud bill keeps growing;
  • development and production can share spare capacity safely;
  • someone can own the cluster, monitoring, backups, and incidents.

The fact that an application was built with AI does not change its production requirements. Users still expect their data to survive, releases to be reversible, and failures to be noticed. That is the same boundary between a demo and a production-ready MVP, but the infrastructure decision often arrives earlier than expected.

What to Calculate Before Moving

Do not compare the price of Cloud Run with the price of one server. Compare the full cost of operating the application:

  • compute, databases, storage, and outbound traffic;
  • builds, the container registry, and external APIs;
  • backups and restore testing;
  • monitoring, updates, and incident response;
  • the time of the person responsible for all of the above.

After this calculation, public cloud may still be the best option. Its elasticity, managed services, and mature security controls are often worth the price. But when an early-stage product is mostly paying for convenience it barely uses, fixed infrastructure can be more rational.

For me, this move was not a rejection of the cloud. It was a way to align infrastructure costs with the stage of my projects. First, they need a predictable platform and clear operations. If traffic and revenue grow, Kubernetes does not close the road back to the cloud.

Photo by Kevin Ache on Unsplash