Running a Business on a £3 Droplet: An Ode to the Most Fragile Infrastructure in Britain

There's a server somewhere in the UK right now — probably in a Frankfurt data centre, technically — processing card payments for a garden centre in Luton. It has one vCPU, 1GB of RAM, and a monthly invoice of about £2.40. It has been up for 400-odd days. Nobody has logged into it since March.

If you've spent your career in agencies or enterprise, this will horrify you. If you've ever run a small business's website for a mate of a mate, it will feel like reading your own diary.

The Economics of Barely Enough

The £3 tier exists for a reason: it's the price point where 'just get a server' becomes cheaper than arguing about it. At that size you're not buying redundancy, you're not buying headroom — you're buying a machine that works, on a good day, for a workload that's small enough that nobody notices the corners being cut.

And here's the uncomfortable truth: for a surprising number of British small businesses, that's fine. A booking form for a hairdresser in Leeds doesn't need Kubernetes. It needs nginx, a Postgres instance, a Let's Encrypt cert and someone who vaguely remembers the SSH password. The whole thing fits comfortably in a gigabyte of RAM if you tune your swap file and say a small prayer.

The problems start when "fine" becomes load-bearing.

How These Setups Actually Survive

Having poked around a fair few of these systems, the survival strategies are remarkably consistent:

Nobody touches anything. The server runs Debian 10 because upgrading to 11 once broke PHP and nobody has the appetite to relive that. It's frozen in amber. Is that good practice? Absolutely not. Does it work? Apparently, for years at a time.

Backups exist by accident. The hosting provider's snapshot runs nightly, and that snapshot — taken by a company, not a person — is the entire disaster recovery plan. It's never been tested. It's also, weirdly, often sufficient, because the data volume is tiny.

The workload never grew. Traffic is flat. The business serves 200 customers a month and will serve roughly 200 customers a month forever. Capacity planning for growth is a solution to a problem this server will never have.

There's a lesson in here for teams running far bigger estates: a lot of infrastructure complexity exists to handle variance, and if you genuinely have none, the complexity is dead weight.

The Day It Dies

But these setups do fail, and when they do, it's never graceful. The failure mode of a £3 droplet isn't a degraded experience or a partial outage — it's a hard stop. Disk fills up with logs nobody rotated. The kernel panics after a provider maintenance window. Postgres corrupts during an unclean shutdown because there was never enough memory for it to flush cleanly.

And then the real cost surfaces, and it's never the hosting bill. It's the accountant's Tuesday. It's the three days of a freelancer's time at £400 a day to reconstruct a server from memory and a stale backup. It's the GDPR conversation about whether the customer table in that last snapshot was encrypted (it was not).

The maths is brutal and one-sided. Saving £30 a month by downsizing from a sensible instance is a rounding error. One bad recovery eats a decade of those savings.

When Cheap Is Actually Right

I'm not here to tell you every small business needs managed infrastructure. Some honest thresholds:

• Cheap is fine when downtime costs nothing, data is reproducible, and someone — anyone — knows how the box is put together. A static marketing site with a contact form? Go wild.
• Cheap is dangerous when the server holds customer data, takes payments, or is the only copy of anything. At that point you're not saving money, you're writing yourself an unpriced liability.
• The middle ground is usually just better defaults: automated backups you've restored from at least once, disk and memory alerting (a free UptimeRobot monitor and a 90% disk alert cost nothing), and a one-page document describing the setup.

That document matters more than the hardware. The single biggest cost in these failures isn't the outage — it's the archaeology afterwards, because the one person who knew how it worked has moved to Bristol and changed phone numbers.

The Actual Lesson

The £3 droplet is a mirror. It shows you the minimum viable infrastructure, and by extension, it shows you what all your extra spending is for. Every pound above that baseline buys you resilience, headroom, or someone else's 3am — and it's worth auditing whether you're getting any of those things, or just paying for a nicer dashboard.

Most of us don't need to spend more. We need to spend the same, deliberately, with a backup we've actually tested and a runbook that isn't one person's memory.

And if you're reading this on a server you haven't logged into since spring: check the disk usage. Trust me on this one.