Supabase buys Turso to give AI agents cheap SQLite databases on demand
Victoria Shinder2 min readSupabase, the hosted Postgres platform, is acquiring Turso, the company behind a Rust reimplementation of SQLite and a cloud service built on it. The announcement on Supabase's blog frames the deal around one workload: software written by AI agents, which create databases in far larger numbers and for far shorter lives than human developers do.
Supabase puts a number on that. It says it already launches more than one million databases per week, and that it expects demand driven by agents to outrun the capacity the industry has today. Its argument is that a database should be about as cheap and as quick to create as a file, and that a small prototype should not need a dedicated machine of its own.
That is where Turso fits. According to Supabase, Turso's platform lets one server manage millions of SQLite databases, loading each one when a request arrives and suspending it when it goes quiet. The post names Superhuman, Sauna.ai, CTO.new and Mastra as customers that give each agent its own database this way, either on Turso Cloud or inside their own cloud accounts.
The division of labour is stated plainly. Supabase stays built around Postgres; Turso continues its SQLite work and keeps operating as a service. Existing customers of both are told nothing changes for them. What the two will build together is described only in outline: the Supabase developer experience for agents, with SQLite for small, on-demand workloads and a route to Postgres once an application grows. Turso co-founders Glauber Costa and Pekka Enberg join Supabase, and Costa will lead the agent infrastructure effort. Financial terms were not disclosed in the post.
What it means
The interesting part is not the purchase but the assumption behind it: that the number of databases, rather than their size, becomes the scaling problem. Per-instance pricing, provisioning APIs and monitoring built for a few long-lived databases do not fit millions of short-lived ones. Teams that let agents create infrastructure should look at what happens to the databases nobody deletes, and whether the promised path from SQLite to Postgres is a migration they would actually run.