There is a moment in every hosting conversation where someone says the word cloud and everyone nods, because the cloud is the default now, the way electricity is a default. You do not choose it so much as fail to question it. We questioned it, ran the numbers for the kind of sites we actually look after, and came out the other side with efficient, low-density machines we own outright. This is the honest version of why.
The arithmetic of small sites
Most websites are small. Not small as in unimportant — small as in measurable. A bakery, a two-person law office, a photographer's portfolio, a village association: sites like these see hundreds of visits a day, a few thousand in a very good week. Served as rendered pages with sensible caching, that is a workload a modern laptop chip does not notice. The processor in a beacon node spends most of its life close to idle, and it is doing its job precisely then.
The cloud's great gift is elasticity — the ability to grow tenfold overnight. That gift costs money every month whether you use it or not, and a site that gets four hundred visits a day will never use it. Paying indefinitely for elasticity you will never need is not prudence. It is a subscription to someone else's business model.
What the cloud actually rents you
A cloud instance is a slice of a shared machine in a building you will never see, with performance that depends on your neighbours and a bill that depends on your attention. The headline price is rarely the real one. The real one arrives later, in egress fees, storage tiers, snapshot charges, and the support plan you buy after the first bad afternoon. And when something is slow, you file a ticket into a queue and wait for a stranger who has never seen your site.
None of this makes the cloud wrong. For bursty workloads, big teams, and audiences spread across continents, it is genuinely the right answer, and we will say so to your face. But we are not those, and neither are the clients we take. Renting a data centre to serve a bakery's opening hours is using a container ship to cross a pond.
Renting a data centre to serve a bakery's opening hours is using a container ship to cross a pond.
The 35-watt argument
Our production fleet is a set of beacon nodes — every machine here gets a name before it gets a workload — efficient, server-grade boxes drawing about 35 watts each under load. That is less than a good desk lamp. It sounds unserious until you remember what the workload actually is: for low-traffic sites, the visitor's experience is dominated by page weight, caching, TLS handshakes, and distance — things you fix with craft, not with scale.
A page that arrives in six hundred milliseconds from a well-tuned 35-watt box beats a page that arrives in two seconds from a hyperscaler, every single time, and no visitor has ever asked which one had more vCPUs behind it.
What ownership changes
The practical difference is speed of understanding. When a client's site misbehaves, we do not open a dashboard of abstractions and guess. We know exactly what else is running on the node, because we put it there. Diagnosis that takes a cloud support queue days takes us minutes, and the fix is applied by the same hands that found the fault.
The deeper difference is incentive. A host that oversells shared servers profits most when you use least of what you paid for. We cannot oversell even if we wanted to — our nodes' limits are our limits, in plain sight on our own racks. That is why the fleet runs low-density by policy: when capacity approaches its engineered limits, we add iron instead of a spreadsheet of excuses.
The honest trade-offs
Owning iron means carrying the risks the cloud abstracts away. Hardware fails. Power flickers. Disks age. We answer with the boring disciplines: nightly backups kept off-node, restores we actually rehearse, spare parts on the shelf, and a recovery plan that has been walked through rather than merely written down. Boring is the point — reliability is mostly the accumulation of unglamorous habits.
And there is a ceiling, which we name out loud: if your site becomes a national event, a low-density node is the wrong home for it, and we will help you move somewhere elastic and cheer you on. Until that happy emergency, your pages come off a node with your name in its notes, watched around the clock.
That is the whole argument. Not that owned hardware is right for everyone — it is that it is right for this: well-tended sites on iron with a name, served from a fleet that never oversells, looked after like it matters. Because to the people who own those sites, it does.
Written from the engine room. Questions about anything above? Write to us — answers usually land the same day.