Most companies that have a seven-year-old server sitting in a closet next to accounting talk about the cloud in one of two ways. Either "cloud is just someone else's computer and it's not actually cheaper" — and they push the decision out another year. Or "let's move everything to the cloud, it'll pay for itself" — and they end up with a monthly bill that's twice the electricity cost of that old server within six months. Both attitudes come from the same root: nobody ever sat down and did the math properly.
The server-vs-cloud decision in an SME is not ideological. It's an accounting problem with concrete line items: hardware, licenses, electricity, admin time, hardware failure risk, ransomware risk, migration cost. Some of these show up on invoices; some are hidden in the yearly bonus of the bookkeeper who loses half a day every month to "weird stuff" on the server. The goal of this article is one thing: to show how to count these properly, before you sign anything.
What's actually sitting in that server room
Before the cloud topic even shows up, take inventory. Sounds trivial — in half the companies we visit, no one has an up-to-date list of what runs on the physical server and what for. "Oh, probably some accounting database. And the old file share. And it does the backups to that NAS in the other cabinet, I think."
Typical contents of an office server in a 20-person manufacturing or trading SME:
- ERP database (Comarch Optima, Subiekt GT, enova365 or Microsoft Dynamics),
- file shares — accounting, HR, sales documents,
- some industry-specific application (quote calculator, warehouse management, CAD license server),
- Active Directory / domain controller,
- print server, sometimes scan-to-folder,
- local backup — most often a USB disk or a NAS in the adjacent room.
Each of those elements can have a different answer to "cloud or not". An ERP database with 15 people working concurrently behaves differently in the cloud than a file share that three departments touch once a day. That's why the decision "we're moving everything to the cloud" is just as foolish as "we're staying on the server until the end of time". You have to split it into layers.
Three realistic paths, not two
In practice for an SME there are three paths, not two:
- Everything on-premise — you buy a new server every 5-7 years, you have full control, you pay a known one-off amount for hardware and licenses.
- Hybrid — what needs low latency (ERP, WMS label printing) stays local; the rest (email, documents, off-site backup) goes to the cloud.
- Everything in the cloud — Azure / AWS / OVH Cloud + Microsoft 365 or Google Workspace; only the router, switch and Wi-Fi stay on site.
In Polish SME reality, path 2 is most often the most sensible. Not because it's "the trendiest", but because when you count it honestly, it usually ends up cheapest and with the lowest risk. All-cloud makes sense for companies without a desktop-class ERP (services, software houses). All-on-premise makes sense when your internet link is only good on paper and a 3-second lag on a warehouse transaction means warehouse staff goes back to paper slips.
How to count it properly — the 5-year TCO
The most common mistake when comparing server to cloud: comparing hardware price to the monthly Azure bill. These are two different units and the conclusion will be nonsense. Compute the 5-year TCO (Total Cost of Ownership) for both options. That's one hour with a spreadsheet, not a 30-thousand consulting engagement.
What goes on the on-premise side (5 years):
- server + disk array (15-30k PLN net for 20-50 users),
- UPS plus optionally server room AC (3-8k),
- Windows Server + CAL licenses, possibly SQL Server (10-25k depending on user count),
- electricity (server + cooling, typically 1500-3000 PLN per year),
- local + off-site backup (NAS + cloud backup, 3-5k per year),
- admin time (external or internal) for maintenance — often 4-8 hours per month,
- every 2-3 years: disk replacements, UPS batteries, possibly memory upgrades.
What goes on the cloud side (5 years):
- virtual machines or PaaS services (typically 800-3000 PLN per month for an SME, depending on ERP),
- per-user licenses (Microsoft 365 Business Standard ~ 60 PLN/user/month),
- transfer and cloud backup,
- admin time for maintenance — usually smaller, but not zero,
- one-off: migration cost (15-60k PLN depending on scale).
After running the numbers, the 5-year delta is usually not dramatic — a dozen-or-so percent in either direction. But that's not the whole story. The rest of the story is risk.
The line items nobody puts in the spreadsheet
There are four costs that don't appear on any invoice, yet they often dominate in reality.
Hardware failure risk. A 2017 server running your ERP will, at some point, fail. The question is not "if" but "when" and "what day of the week". At one of our clients — a 28-person building-materials wholesaler — the RAID controller died on Thursday at 9 a.m. The server had been out of vendor support for 18 months. Time to source an identical second-hand component "right now": 6 days. Sales during that time — on paper slips and hand-written invoices. Nobody put that in the 2017 purchase spreadsheet.
Ransomware risk with local backup. The classic "server + NAS in the next room" setup goes down in a single incident. Cloud backup (or properly separated off-site backup) saves you, but it's an extra cost you need to add to the on-premise scenario.
Admin time during a crisis. The external admin doesn't pick up at 5 p.m. on Friday. On Monday, missed orders are waiting. Cloud doesn't eliminate this, but it shrinks the class of problems that need someone physically driving in.
Scaling up. When the company suddenly gains 8 new users (acquisition, new branch), on an on-prem server this often means "buy CALs and pray the RAM is enough". In the cloud, it means "click and pay more". The reverse direction is rarer — the cloud doesn't love "scaling down".
A concrete decision: 28-person wholesaler, Comarch Optima ERP
Let's show this with a real configuration. At one of our clients — a 28-person electrical-supply wholesaler — there was an HP ProLiant server from 2017, running Comarch Optima with about a 90 GB database, file shares and the domain controller. The hardware still worked, but it had survived two disk replacements and had been out of vendor support for 18 months.
We costed three options over 5 years:
- New on-premise server: 142k PLN TCO (hardware, licenses, electricity, admin, off-site backup).
- Everything to cloud (Azure VM + M365): 168k PLN TCO + 38k migration.
- Hybrid: Comarch Optima stays local (a stronger Dell mini-server for 28k + licenses), email, documents, AD and backup move to Microsoft 365 and Azure. TCO: 121k PLN + 22k migration.
They picked the hybrid. Seven months after rollout: full off-site backups (previously basically non-existent), MFA on email (three accounts had already been compromised once), the office team got back 6 hours per week previously spent manually tidying file shares. The biggest saving wasn't in money — it was in nerves during the last big storm, when the area lost power for 14 hours while the online store kept taking orders.
This isn't a story about cloud winning. It's a story about somebody actually doing the math.
What to do this week
If your server is older than 5 years, or vendor support is ending, don't wait for a failure in November. Do four things:
- Inventory what's on the server — services, databases, shares, versions, licenses, who uses what.
- Check the hardware contract and vendor support — when warranties end, when support is being retired.
- Run an honest 5-year TCO for at least two options (on-premise vs hybrid) — including electricity, admin time, off-site backup, migration.
- Verify the internet links in the office and warehouse — without a stable 100/100 Mbps connection plus a failover link, a cloud-hosted ERP in production won't fly.
Only with those four items on the table does talking to a vendor make sense. Without them, you're buying blind — whether you're buying a server or a cloud.
If you want someone to walk this spreadsheet through with you on a concrete example of your company — book 30 minutes. Free, with no product brochure on the other side of the call.
The server-vs-cloud decision is not ideological. It's a calculation you do once every five years — and no vendor will do it for you, because every vendor has skin in the answer.