Build Your Own Multi-Tenant Stack, or Buy a Platform?

Building means standing up your own provisioning, patching, backup and monitoring for client instances; buying means paying a platform to run each client’s instance for you. Building wins on cash cost per client and total control; buying wins on time, on not depending on one engineer’s availability, and on the tenth client costing the same to add as the first. For a practice without dedicated infrastructure engineers, buying is almost always right. If you have them, building is a legitimate choice, not a mistake.

Both sides of that answer deserve to be taken seriously. Here’s the comparison, played straight.

What building actually involves

Renting servers costs $10 to $50 a month each, and installing open-source software on one is a well-documented afternoon for a competent engineer. That is the easy 10%.

The other 90% is what turns one working server into a fleet you can sell: automated provisioning so a new client is minutes rather than a ticket, patching that reaches every client without drift, per-client backups that are actually restorable, certificate issuance and renewal at scale, monitoring with an escalation path, and a runbook that survives the engineer who wrote it leaving. Our cost guide puts numbers on the ongoing time: typically 3 to 5 hours a month per client, plus the interruptions, plus the build.

The build is the part that gets underestimated, because it is the part that feels like progress.

Where building genuinely wins

• Cash cost per client. Nobody will run a client’s instance for less than you can at the margin. If the engineering labour genuinely exists and is otherwise idle, building is cheapest in money.
• Total control. Every version, every configuration, every integration decision is yours. Unusual requirements such as strict data-location rules, an exotic compliance regime or a bespoke deployment often only fit something you built.
• No provider to depend on. There is no platform to outgrow, be acquired, or reprice. (A good platform blunts this by making leaving easy, but building removes the question.)
• It can become a product. If hosting is a capability you intend to sell as such, the stack is an asset rather than a cost centre.

Where buying wins

• The work happens even when everyone’s busy. Patching, backups and monitoring don’t depend on one engineer’s diligence surviving holidays, crunch weeks and resignations. This is the single biggest practical difference, and it is a client-facing risk, not just an internal one.
• Out-of-hours belongs to someone else. When a client instance breaks at 6pm on a Friday, a team whose whole job is running it is already on it, and you are not paying for a rota to cover a service that is one line on the invoice.
• The tenth client costs what the first did. Bought capacity scales by arithmetic. A stack you built scales by whatever you automated, and the gaps show up exactly when you are winning work.
• Predictable cost you can price against. A flat monthly figure per client, instead of a variable tax on your most capable engineer’s attention.
• Speed to live. Minutes instead of a provisioning ticket, with the client’s domain and certificate done properly on day one.

Side by side

Per client:

	Build it yourself	Buy a platform
Cash cost	$10 to $50/month	From $49/month
Time cost	3 to 5 hours/month, forever	None
Up-front engineering	Real: provisioning, backups, monitoring	None
Skill needed	Infrastructure, ongoing	None
Control	Total	High, within the platform’s shape
Depends on	Your engineer staying available	The platform doing its job
When it breaks at 2am	Your rota	The platform’s problem
Client data ownership	Theirs	Theirs, if exports are free and standard

How to decide

Build if you employ infrastructure engineers with genuine spare capacity, have requirements a platform cannot meet, or intend to sell hosting as a capability in its own right. Budget the build and the per-client hours honestly, and write down what happens when the engineer who knows it leaves.

Buy if client software matters to your practice but running servers isn’t the practice. You keep everything that made open source worth choosing, no seat licence, no vendor between you and your client, a real exit, without staffing it. And check the platform gives you real isolation per client, not a shared database with a client column; buying the wrong kind of multi-tenancy trades an operational problem for a security one.

Kosnomos is the buy option: every client app on its own virtual machine, live in about ten minutes, patched and backed up daily, watched around the clock, priced by instance size, from $49/month per client, never per seat, with volume discounts from 10 clients.