How can deployment platforms afford free tiers?

Deployment platforms can offer generous free tiers because they stopped renting you a machine and started renting you execution. A VPS reserves CPU and memory for you around the clock. A modern platform shares compute across thousands of workloads and allocates it only while your code is actually handling a request. Idle apps, which is most apps, cost almost nothing.

This is part 1 of my Modern Deployments series.

What did deployment look like before?

A few years ago, deploying a web app usually meant renting a VPS: 2 vCPUs, 4 GB of RAM, 80 GB of SSD. Then you configured Nginx, installed Node, set up PM2, sorted out SSL, opened firewall ports, and hoped nothing crashed at 3 AM.

Today most developers push to GitHub, click Deploy, and get a production URL in under a minute. It almost feels like servers disappeared. They didn’t. Someone is still paying for CPUs, RAM, storage, bandwidth and electricity.

Why is a dedicated server wasteful?

A VPS allocates infrastructure before your application receives any traffic. Whether your app serves one request a day or a million, that virtual machine stays powered on and keeps its CPU and memory reserved for you.

Most side projects, portfolios, hackathon apps and internal tools spend well over 99% of their lifetime idle. Provisioning them as if they were constantly busy is the waste.

What changed?

Platforms stopped giving every application its own long-lived server. Instead they built systems that:

• Pack many workloads onto shared machines, isolated by containers or micro-VMs
• Start compute on demand when a request arrives
• Reclaim it once the app has been idle for a while
• Cache static output at the edge, so most page views never touch compute at all

The economics changed because the allocation model changed. The free tier exists not because servers got cheap, but because an idle application no longer needs dedicated infrastructure.

What’s the catch?

Renting execution has real costs, and they show up as behaviour you need to understand:

• Cold starts when compute has to be created for a request. See what a serverless cold start really is.
• Usage limits on execution time, bandwidth and invocations, which is where the business model makes its money once you grow.
• Less control over the machine your code runs on. See why your app doesn’t run on a server anymore.

The rest of the series

1. How free tiers work (this post)
2. Your app runs on a desired state, not a server
3. What is a serverless cold start?
4. What actually gets deployed to production?
5. Where do your environment variables go?

Once you see “serverless” as renting execution instead of machines, the rest of the series starts to make sense.