A data centre burned down in 2021 and took thousands of businesses with it. The lesson was not "avoid that host" — it was that almost nobody had a backup they controlled.
In March 2021, one of OVHcloud's Strasbourg data centres caught fire and was destroyed. A second was damaged. Around 3.6 million websites went offline, and a substantial number never came back — because their only backups were in the same building, or in the adjacent building, or in a service the customer had assumed was running and had never tested.
The instructive part is not that OVH had a fire. Every provider will eventually have an incident: a fire, a flood, a ransomware event, a fibre cut, a botched migration, or an acquisition that quietly ends the product. The instructive part is how many professional operators discovered on the same morning that their disaster recovery consisted of hope.
What actually goes wrong, ranked by likelihood
- You break it yourself. A plugin update, a bad deploy, a mistyped database query. By far the most common cause of needing a restore, and the reason retention period matters more than backup frequency.
- Your account is suspended. An expired card, a chargeback, a false-positive abuse report. Your site and your host's backups become simultaneously inaccessible, and you are negotiating with a billing department rather than a support engineer.
- Compromise. Ransomware or a defacement, discovered days after the fact — which is why a backup from yesterday may already contain the intrusion.
- Provider incident. Hardware failure, a mis-executed migration, a data-loss bug. Rarer, and usually recoverable from the provider's own copies.
- Provider ceases to exist. Fire, insolvency, acquisition and wind-down. Rare, and total.
Notice that in cases 1, 2 and 3 — the likely ones — the host's own backups are either insufficient or unreachable. A backup you cannot access without your host's cooperation is not a disaster recovery plan. It is a convenience feature.
The minimum viable plan for a small site
- An off-site copy in storage you own. Backblaze B2 or Cloudflare R2, in your own account, paid on your own card. About 50p/month for a small site. UpdraftPlus or a cron script will push to either. This is the single item that converts a catastrophe into an afternoon.
- Retention of at least 30 days. Long enough to predate a compromise you noticed late. Seven days of backups is seven days of possibly-already-infected backups.
- A tested restore. Once a quarter, restore into a staging site and click around. Backups fail silently far more often than anyone expects — a database export that stopped completing three months ago looks identical to a healthy one in a file listing.
- Your domain at a separate registrar. If the host is unreachable, you can still repoint DNS to a new one. If the domain is held there too, you cannot.
- A written recovery note. Where the backups are, what the credentials are, what the DNS records should be, in an offline password manager entry. Write it now; you will not be thinking clearly on the day.
What good hosts do, and what to ask
Ask any prospective host three questions: where are backups physically stored relative to the live data, how long is retention, and can I restore without opening a ticket? The good answers are "a separate facility", "30 days or more", and "yes, from the panel". Krystal, WP Engine and Kinsta all answer well. Budget shared hosting frequently cannot answer the first question at all, and some providers charge for restores — a policy that tells you exactly how they view the service.
Note too that on unmanaged infrastructure — Hetzner, DigitalOcean, any VPS — backups are off by default and entirely your responsibility. The 20% snapshot surcharge is worth paying and is still not sufficient alone: snapshots sit with the provider, so they cover cause 1 and cause 4, and nothing else on the list.
The proportionate response
You do not need multi-region replication for a plumbing company's website. You need one off-site copy in your own storage account, thirty days of retention, and one restore test a year. That is about an hour of setup and £6 a year, and it is the difference between a bad morning and a business that no longer has a website.
The people who lost sites in Strasbourg were not careless. They were doing the normal thing, which is to assume the host has it covered. The hour spent proving that assumption is the cheapest insurance in this entire industry.
Published 26 June 2026. Prices and ownership change — check the comparison table for current figures, and the provider's own site before buying. No affiliate links anywhere on this site.