Single-Tenant vs Multi-Tenant: A 2026 Decision Table
Single-tenant gives every customer their own database and one deploy each. Multi-tenant puts all of them in one database separated by row-level security, so a single deploy reaches everyone. The choice is usually made by accident, and it sets your cost per customer for years.
The actual difference
Both models serve many customers from one codebase. The difference is where the wall between customers is.
Single-tenant gives each customer their own database and their own deployment. Two customers cannot see each other because they are not in the same place. Multi-tenant puts every customer in one database with an organisation column on each row, and a policy that filters by who is asking. Two customers cannot see each other because a rule says so.
That distinction produces every other trade-off on this page.
The decision table
| Single-tenant | Multi-tenant | |
|---|---|---|
| Database | one per customer | one, shared |
| Deploy | one per customer | one, everybody |
| Isolation | physical — separate credentials | logical — row-level security |
| Upgrade | N deploys, versions can be pinned | one deploy, all move together |
| Blast radius of a bug | one customer | all of them |
| Infrastructure cost | scales with customers | close to flat |
| Per-customer change | possible, and a trap | configuration only |
| Backup and restore | per customer, simple | whole database, awkward |
| Suits | few, high value, divergent | many, similar, price-sensitive |
The one that surprises people: version pinning
In single-tenant, a customer can stay on the version they are on. You ship the update when they are ready, and a customer mid-audit or mid-season can wait. In multi-tenant, everyone moves when you deploy. There is no version to pin, because there is one running copy.
That is an operational difference more than a technical one, and it decides what you can promise in a contract. Enterprise buyers ask about it. Small businesses never do.
The per-customer change trap
Single-tenant makes a customer-specific change easy, which is exactly why it is dangerous. The request arrives, the change is small, the deployment is theirs, and nobody is harmed. Do that five times and a fix written once no longer applies anywhere.
Single-tenant does not cause forking. It removes the thing that would have stopped you.
Multi-tenant forces the discipline by making per-customer code physically awkward. Whether that is a benefit depends on whether you were going to hold the line yourself.
How to choose
- Pick single-tenant if you expect fewer than about twenty customers, they differ meaningfully, they care about where their data sits, or one of them is large enough to want their own upgrade schedule.
- Pick multi-tenant if you expect many similar customers at a price point where one hosting bill each does not work, and configuration can genuinely cover the differences between them.
- Do not pick on elegance. Both are correct architectures. The wrong one is the one that does not match how many customers you will have and how alike they are.
Staying able to choose later
You do not have to decide today, but you do have to keep the option. The way to keep it is to put an organisation key on every table from the beginning, even while you have one customer, and scope every access policy by membership rather than by a hard-coded identifier.
Do that and the schema runs either way — one database holding a single organisation, or one holding many — without a code change. Skip it, and the retrofit later is the most expensive change on the whole road.
When this decision is not your problem yet
If you have one customer and nobody has asked to be the second, this is not your question. Choosing an architecture for customers you do not have is how internal systems grow features nobody uses. The question becomes real the day somebody outside asks to use what you built.
Which rung is your system actually on?
The Platform Readiness Review is seven days and $2,500. A map of what you have built, where the core-versus-configuration boundary really falls, the assumptions that break when a second business signs, and a 90-day plan including what to postpone. If it is not ready to be productized, we say so and you keep the map.