Your Newest Customer Is an AI. Supabase Built for It First

Seventy percent. That is the share of new Supabase databases created by an agent or an AI tool, not a person. Supabase gives app makers a ready made backend, the behind the scenes machinery that stores data and signs users in, so they do not have to build it themselves. Eighteen months ago Supabase had two million developers. Today it has thirteen million, and the fastest growing user on the platform is not human. This month it raised $150 million and bought Turso to serve that user better.

Disclosure up front. UpRound holds indirect exposure to Supabase through USVC. Everything below is public information.

Software used to be measured in developers, one human signing up at a time. That count now misses most of what is happening, because software increasingly starts itself through agents. The more revealing number is databases, one created for every job the agent takes on. Supabase now creates four million a month. In June agents were already starting the majority. Today they start seven in ten.

A developer starts a few projects. An agent starts a database every time it tests an idea, runs a task, or builds a dashboard. Demand stops scaling with people and starts scaling with compute.

That is why the curve keeps steepening. Supabase went from four million developers to nine million in eight months, then to thirteen million in four. Venture capital repriced it fast. The valuation went from $2 billion to $10.5 billion in fourteen months.

The most important customer Supabase has today is not a developer. It is another platform. AI app builders turn a sentence into software for people who have never written code. Every one of those apps needs somewhere to store data and handle logins, and the builder picks who provides it.

Lovable, one of the biggest of these app builders, picked Supabase. When Lovable builds you an app, the backend underneath is Supabase, and you never chose it. Most of the big builders made the same pick, and the Supabase product that serves them grew its customers 370% in the six months to June. That is distribution no sales team can buy. One integration decision upstream becomes millions of databases downstream, created by users who may never know Supabase exists.

The product makes the choice easy to repeat. One call gives an agent everything that machinery includes: a database, user logins, file storage, live updates and search. A standard connector, called an MCP server, lets other agents read and write to it directly. Defaults compound. Every app an agent ships on Supabase adds to the pattern the next agent follows.

Supabase bought Turso to add a capability it can build the next decade on. The company already launches more than a million databases a week, most of them on Postgres, the industrial grade database that serious companies run on. That works today. It gets expensive at the next jump in scale, when every agent task wants its own database and most of them live for minutes.

Turso is built for that world. It took SQLite, a lightweight database, and made it run millions at once on a single server, waking each one only when it is used and sleeping it when it is not. Companies like Superhuman, the email app, and Mastra, an agent building framework, already give each agent its own database this way, in Turso's cloud or their own.

The fit is clean. SQLite handles the cheap, disposable database. Postgres handles the app that survives. A builder starts on one and graduates to the other without leaving Supabase.

On the same day, Supabase previewed Compute, a place where an agent can keep working over time instead of running once and stopping. Those agents need somewhere to store what they produce, and I expect that to be Turso. Running it all is Glauber Costa, Turso’s founder, who joined Supabase and wants to serve a billion databases. Supabase just gave him the demand to try.

The strongest argument against Supabase is that its moat belongs to someone else. The growth runs through AI app builders, and the distribution that drives it is also the exposure. A builder that chose Supabase can choose again. Bolt, a builder that once relied on backends like this, already ships its own database. Every builder at scale will ask whether the backend is too valuable to rent.

The agent behavior is not unique either. When Databricks bought Neon for about $1 billion, more than 80% of Neon's databases were being created by agents. Agents create databases everywhere. The question is where they get created, and that is the fight Supabase is in.

And the hyperscalers are moving up. Google Cloud spent October pushing managed hosting for AI agents, infrastructure that lands squarely against Supabase Compute.

Supabase's answer is product depth and open source. A builder can leave, but it leaves behind a full backend, a developer community and a migration bill. That is a real switching cost. It is not a permanent one.

I am watching three things from here. The first is the agent share of new databases, now seventy percent. The second is how many disposable Turso databases graduate into Postgres apps that pay. The third is whether the builders that chose Supabase keep choosing it as they scale.

The second matters most. Agents bring the volume. Graduation is where volume turns into revenue. For twenty years, software was sold to the people who built it. That era is ending. The buyer is now the builder's agent. Supabase saw it before anyone else.

Bashar Aboudaoud
Managing Member, UpRound

Live Webinar: Tuesday, October 27th: You leave with three things. Where private valuations actually sit right now. Which sectors are repricing fastest. How I am reading prices into 2027. 20 minutes, then live Q&A. 20 slots only for UpRound readers.

Note: Due to overwhelming demand, we’ve increased the number of seats. Click below to join the webinar this Tuesday.

Share with 10 people to receive a quarterly call with Bashar.