Two Hosts, Four, Ten: What Your Business Actually Needs at Each Stage

Most businesses don’t get quoted for the stage they’re in. They get quoted for the stage a vendor would like them to reach.

That’s how a company with twenty people ends up licensed for capability built to run a hospital, and how a company with two hundred ends up running its entire operation on equipment that should have been replaced two years ago. Both are paying the wrong amount for the wrong thing, and neither one knows it, because nobody ever laid out what the path is supposed to look like.

So here it is. Two hosts to ten, what changes at each step, and what to skip.

First, the variable that actually decides everything

Before host counts, one thing matters more than any of them: who runs this when it breaks.

There is an enormous difference between a business where one person or an outside provider covers everything, and a business with a real IT staff — administrators and engineers who own separate pieces. Those two companies can have identical server counts and should not buy identical systems.

The first company needs technology it can operate and recover with the people it has. Sophistication it can’t support isn’t protection — it’s a longer outage with a support contract attached. The second company can absorb complexity, because someone on staff understands it at two in the morning.

Server count tells you what to buy. Staff depth tells you what you can safely operate. When those two disagree, staff depth wins. I’ve watched more businesses hurt by that mismatch than by buying too little.

Stage one: two hosts

Roughly twenty to fifty people. Often around twenty virtual servers.

Twenty virtual servers sounds like a lot to an owner and isn’t. Domain controllers, a file server, a print server, an application server, a database, remote access, a couple of vendor-specific systems nobody remembers installing. It adds up quickly in any business running real software.

What this business needs is genuinely modest: two hosts, storage that lives inside those hosts and replicates between them, a backup separated from production, and one server offsite that could carry the load if the building went dark.

That’s it. And two details in there matter more than everything else.

The storage should live in the hosts, not beside them. The common design is two servers connected to one small shared storage array. It looks redundant — two servers! — but both of them depend on that single array. Lose it and you’ve lost everything, and you’ve paid for redundancy you don’t have. Storage that lives inside each host and replicates to the other survives losing either one. Same budget range, fundamentally different outcome. This is the most common expensive mistake I find at this stage.

Offsite doesn’t mean a second data center. One server, somewhere else, capable of running your critical systems in a degraded state. Not full failover. Not an identical environment. Just enough that if the building floods, orders still get taken and work still gets scheduled while you sort out the rest. Owners are routinely quoted full-failover architecture when degraded but running is what the business actually needs, and the gap between those two numbers is large.

What this stage gets oversold: platform licensing. Enterprise-tier virtualization is priced for capability this business will never consume — automated workload balancing across large clusters, deep network segmentation, orchestration tooling for environments many times this size. None of it is bad. It’s just aimed somewhere else, and it gets quoted here by default. That’s a platform-fit question, not a brand question.

Stage two: three to four hosts

Roughly seventy-five to a hundred fifty people, single location.

Density is what changes, not the shape. More virtual servers, bigger databases, more people hitting them at once. Usually a third host arrives because the first two genuinely ran out of memory — that’s a legitimate trigger and it’s fine.

Two things start to matter that didn’t before.

Your backup window stops fitting. The overnight job that used to finish by 4am now runs into the workday. This is the earliest honest signal that a business has outgrown its setup, and it almost always shows up before anything else breaks. Owners rarely hear about it because it looks like a technical annoyance rather than a business risk.

Downtime tolerance gets real. At this size there’s usually a system where an hour down costs actual money. That’s the moment to measure it rather than guess — and it’s also the moment most businesses buy redundancy sized to a tolerance nobody ever measured.

What this stage gets oversold: redundancy. Automatic failover for systems that could tolerate a two-hour recovery. Clustering configured at install, never tested, and quietly misconfigured for years. The question that prevents it is boring and nobody asks it: how long can this actually be down before it costs us? Answer that honestly and you’ll usually buy less than the quote in front of you.

Stage three: multiple locations

This is where the path stops going up and starts going sideways.

A business with four service centers doesn’t need one large cluster at headquarters. It needs compute at each service center — usually two hosts, the same pattern as stage one — because when the network link to a location drops, that location has to keep working.

That’s how a business ends up at eight or ten hosts without ever building anything complicated. It’s not one big system. It’s the same small system, replicated, four or five times.

This runs against the instinct vendors sell, which is to consolidate everything centrally and connect the branches back to it. Consolidation looks efficient on a diagram and it makes every location dependent on a network link. For construction, field services, distribution with branches, healthcare with satellite clinics, and energy operations with yards and shops, that dependency is the thing that takes you down.

Growth at this stage means repeating a design you already understand. That’s cheaper, easier to staff, and far easier to recover than anything centralized.

The other kind of growth: one site, high density

Roughly two hundred people and up, single location. Usually manufacturing.

Different reason for the same host count. One building, but a real ERP, a substantial database, systems that talk to equipment on the floor, and workloads that don’t tolerate delay. Four to six hosts in one rack because the work is concentrated, not distributed. This is the manufacturing profile more than any other.

This is the one stage where a genuinely sophisticated platform starts to earn its price — and only if the staffing question from the top of this article comes out the right way. Concentrated workload plus real IT staff is the profile where advanced capability gets used. Concentrated workload plus one overloaded person is the profile where it becomes a liability.

Where cloud actually enters

Here’s the part that gets written dishonestly most often, so let me be direct about what I actually see.

Email and files left the building years ago. That migration is finished for nearly everyone and isn’t worth discussing.

The system your business actually runs on — the ERP, the job scheduling, the industry-specific software your whole operation lives inside — usually has no credible cloud version. Sometimes the vendor offers a hosted option that’s slower and costs more. Sometimes there’s nothing at all. Telling an owner to move it anyway is advice that serves the advisor.

Where cloud genuinely fits in this growth path is recovery. That offsite server from stage one is an excellent cloud workload. It sits idle most of the time, which is exactly what cloud is priced well for, and it removes the need for a second physical location you’d otherwise have to build and maintain.

So the honest path for most of these businesses isn’t a march toward the cloud. It’s on-premises compute with cloud recovery behind it, plus the productivity tools you already moved. Hybrid isn’t a compromise here. It’s the correct architecture, and it stays correct for years.

Backup, sized to what you can afford to lose

One sizing decision that spans every stage. Backup separated from production — genuinely separated, on different credentials and different infrastructure, so an attacker who takes production can’t reach the copy — comes in different sizes, and the size is set by one question: how much work can you afford to redo?

A business that can reconstruct a day from paperwork needs something very different from a shop that just shipped product against data it can’t recreate. Both answers are legitimate. What isn’t legitimate is never having asked, which is the default. That separation is the thing that actually saves you — it’s the reason a ransomware event I lived through cost hours instead of weeks.

Where businesses get stuck

The stall is quieter than the growth. What I notice walking in:

The backup window no longer fits and everyone’s stopped mentioning it. Performance complaints that get explained away for a year. One person carrying the entire environment in their head with nothing written down. Equipment past its supported life still running something the business can’t operate without. And a support renewal that arrives annually while nobody can explain what’s inside it — that’s the upper-stage version of overpaying, and it persists longest because it’s automatic.

None of those cause an outage on their own. Together they’re a business waiting for one.

The point

Most businesses I look at aren’t spending too much on technology. They’re spending it in the wrong places — over-licensed at the bottom, over-engineered in the middle, over-contracted at the top, and under-protected the whole way through.

The path itself is unglamorous. Two hosts with storage inside them. A separated backup sized to what you can afford to lose. One server somewhere else that keeps you operating. Then repeat that per location, or make it denser in one place, depending on which direction your business grows.

Know which stage you’re in and you’ll stop buying for the one someone else wants to sell you. That’s most of the savings, and nearly all of the right-sizing work.

← All articles