A free server — really?

The word "free" in cloud pricing has a specific technical meaning, and it is not the one you think. I learned this the way most people do — from an invoice. Not a catastrophic one, but large enough that I sat down and built a spreadsheet, because a surprise bill is a data problem before it is a money problem.

I spend my working life modelling costs and behaviour from numbers, so the instinct is automatic: do not argue with the marketing, reconstruct the meter. What is actually being counted, at what rate, and what happens when the counter rolls over. Apply that to a free tier and the picture gets clear very quickly.

What "free" is counting

Every free cloud offer has the same three shapes, in some combination. It is free for a period of time. It is free up to a quantity. Or it is free for some resources while adjacent ones bill normally. The third is where people get caught, because it is the one nobody reads carefully.

A server instance can be genuinely free while the storage attached to it is not. The machine can cost nothing while the data leaving it is billed per gigabyte. A backup snapshot taken once a night is trivially cheap and completely invisible until ninety of them exist. None of this is a trick — it is all documented — but the documentation is priced in the language of a metering system and read by people thinking in the language of a shop.

The second thing free tiers count is time, and this is the one that actually matters. Introductory terms end. Usually after twelve months, sometimes on a schedule you agreed to and forgot. The machine does not stop; the meter simply starts. Whatever you built during the free period is now a running cost, and by then it is doing something you depend on, which is precisely the point of the offer.

Building the comparison honestly

When I compare a rented server against something standing in my own room, I refuse to let either side cheat. That means counting things people usually leave out of whichever column they already prefer.

On the rented side: the instance, the storage, the data leaving it, the backups, the address, and the renewal price after the introductory period rather than during it. On the owned side: the hardware, spread across the years you will realistically use it; the electricity, which for a small low-power machine running continuously is a genuine line and a smaller one than most people guess; the internet connection you were paying for anyway; and — this is the one everybody forgets — your own time.

That last line is where a lot of enthusiastic self-hosting arguments quietly fall apart. A machine at home is not maintenance-free. It needs updating, its backups need testing, and when it breaks it breaks on your Saturday. If you value your own hours at anything at all, put a number in that cell. In my case the number is small because this is my trade and the work is already happening. For someone whose trade is something else entirely, the same spreadsheet gives a different answer, and that answer is legitimate.

Where the break-even actually sits

The shape of the answer, rather than the exact figures, is what transfers between situations. Rented infrastructure is cheap at the start and linear forever. Owned infrastructure is expensive once and then nearly flat. The two lines cross, and everything depends on where.

For anything short-lived, rented wins and it is not close. A project that runs for three months, an experiment, a client demo, something you are not sure you will keep — paying by the hour for something you can delete is exactly the right instrument. I use rented servers constantly for precisely this, and anybody telling you self-hosting is always the answer is selling hardware.

The picture inverts on two axes. The first is time: anything you intend to run for years crosses the line, usually sooner than expected. The second is volume, and this is the sharper one. Cloud pricing punishes large quantities of data — especially data that moves. A family photo archive or a company's document history is exactly the shape of workload that looks trivial in month one and becomes the dominant line on the invoice by year three. That is not a flaw in cloud pricing. It is what the pricing is designed to reflect.

The part that is not about money

Having done the arithmetic honestly, I will admit the arithmetic is not why I run my own machines. It is a supporting argument, not the reason.

The reason is that a rented server's price is a variable somebody else controls. Terms change, tiers are restructured, a service is discontinued with twelve months' notice, an account is suspended by an automated system with no human to appeal to. None of those are hypothetical; all of them happen regularly to people who did nothing wrong. A machine in my own room has a fixed set of failure modes, and every one of them is mine to fix.

So the advice I actually give, when somebody asks whether to rent or own, is neither. It is: build the spreadsheet before you build the server. Put the renewal price in, not the promotional one. Put your own hours in. Then look at where the lines cross and decide with your eyes open. The answer is genuinely different for different people, and anyone who gives you a universal one has not done the arithmetic.