Multi-Tenant System
A multi-tenant system is software that serves many customers simultaneously on the same infrastructure, while keeping their data strictly separated. Almost all online services used in a browser work on this principle.
When a company offers a program over the internet, it has to decide how many customers will share one piece of infrastructure. In a multi-tenant system, only a single copy of the program runs on the same machines for all customers. In this context, each customer is called a tenant, roughly translating to lodger or client in German. Each one sees only their own data and notices nothing of the others. The comparison with an apartment building fits well: everyone lives in the same building, sharing the foundation, heating, and stairwell, but each apartment has its own lock. The counterpart is called single-tenant: there, each customer gets their own, completely separate installation.
Why cloud services can be so cheap
The main reason for this design is money. A server costs the same whether one customer uses it or a thousand do. Those who share the infrastructure also share the costs. This is why a provider can sell software for a few euros a month, even though operating it for a single customer alone would never be affordable.
On top of that comes the maintenance effort. When a security hole needs to be patched, a multi-tenant system only has one place to fix. With a thousand separate installations, that would mean a thousand updates. That’s why new features appear for all customers on the same day.
The price for this is less freedom of choice. An individual customer cannot keep an older version if they don’t like the new one. For such reasons, banks, hospitals, and government agencies sometimes explicitly demand their own, separate installation. This is then significantly more expensive.
How tenants' data stays separated
The central technical task is data separation. In the database, the storage for all the service’s information, almost every entry is tagged with a customer ID number. With every query, the system checks who is asking and returns only entries with the matching number. Other providers go further and set up a separate database for each customer, while still operating just one program.
This is exactly where the greatest danger lies. If a developer forgets to check the ID number in a single query, a customer suddenly sees someone else’s data. Such mistakes are called data leaks and have already gotten major providers into trouble. That’s why modern systems build the check in so deeply that it cannot be accidentally left out.
A second problem is known as the noisy neighbor. If a single customer suddenly consumes an enormous amount of computing power, the service becomes slow for everyone else. To counter this, providers set upper limits per customer, so-called limits. Anyone who needs more has to pay extra.
From Google Docs to the ChatGPT interface
Practically every service used in a browser without installation is a multi-tenant system. Google Docs, Spotify, Microsoft 365, or school platforms run for millions of users on shared infrastructure. The cloud itself also works this way: Amazon and Microsoft rent out the same data centers to countless companies simultaneously.
In the AI industry, this principle is especially visible. When a company uses a language model from OpenAI or Anthropic via an interface, it doesn’t get its own model. Its requests run over the same graphics cards as those of all other customers. This is the reason why such requests cost only fractions of a cent.
In the news, the term usually appears in two contexts. Either it’s about data privacy, because a company doesn’t want its data to reside on a machine alongside data from unrelated companies. Or it’s about pricing, when a provider with shared infrastructure undercuts a competitor.
