Choosing a Multi-Tenant SaaS Architecture in Berlin in 2026

The verdict in three sentences

In 2026, 90% of early-stage B2B SaaS are right to start on a shared database with Postgres RLS: it's the cheapest to operate and the simplest to evolve. You only move to a schema per tenant or a dedicated database under real pressure: a compliance requirement, an enterprise customer, or a volume that unbalances other tenants. Each extra notch of isolation improves perceived security but adds 10 to 30% to run per customer and complicates migrations.

The three isolation models

Multi-tenancy runs on a slider: the more you isolate, the more you reassure, the more you pay. Here are the three reference models and their concrete impacts.

The shared database with Row-Level Security (RLS) isolates data via a tenant_id column and Postgres policies. Done right, it is robust and covers the vast majority of B2B cases. The schema per tenant offers cleaner logical separation for audits. The dedicated database is the only one guaranteeing physical isolation — often required by regulated sectors (healthcare, finance, public sector).

Cost, compliance and switch thresholds

The real trade-off isn't technical, it's economic and regulatory. Here are the thresholds that justify a switch in 2026.

One key point: you can combine models. The winning pattern in 2026 is hybrid multi-tenancy — a shared database for the mass of SMBs, a dedicated database for the 2 or 3 enterprise accounts that require it and pay for the overhead. Never pay for physical isolation for a customer who neither asks for it nor funds it.

Mini case study

Nadia, CTO of a document-management SaaS in Berlin, runs 60 SMBs on a shared RLS database for a run of 420 EUR/month. A large banking-sector account signs but demands a dedicated database with data residency. Moving all 60 customers to dedicated databases would cost +25% run, i.e. +105 EUR/month per tenant — unsustainable. Solution: she keeps the 60 SMBs on the shared database and provisions a dedicated database for that one customer, billed as an "Enterprise" tier at +400 EUR/month. The infra overhead (about 90 EUR/month) is absorbed, and margin on that account stays at +310 EUR/month. The hybrid model turns a constraint into a premium product.

FAQ

Is Postgres RLS really safe for B2B? Yes, provided you apply policies on every table and test for cross-tenant leaks. Thousands of regulated SaaS run on RLS; the risk comes from a forgotten policy, not the technology.

When should you move to a schema per tenant? When a customer demands contractual separation (DPA clause, audit) without requiring a physical database. Schema per tenant costs 10–20% more in run but stays manageable up to a few hundred tenants.

How far does a database per tenant scale? Technically thousands, but run cost and migration complexity explode beyond a few hundred. Reserve it for accounts that fund it.

Can you migrate between models later? Yes, but a shared-to-dedicated migration is a multi-week project. Design your data-access layer from the start to make the switch possible without a full rewrite.

Does hybrid multi-tenancy complicate the code? Moderately: an abstraction layer routes each query to the right backend based on the tenant. The initial dev overhead (5,000–10,000 EUR) pays off with the first enterprise account billed at premium.

Let's scope your project. Tell us your target volume, compliance constraints and SMB/enterprise mix, and we'll recommend the isolation model and cost the run. Detailed quote within 48 h. WhatsApp +221 77 596 93 33.