Paul Copplestone says Supabase is now a default database inside Claude Code and Codex, with about 13m sign-ups as builders outside traditional engineering join.
Paul Copplestone predicts enterprises will increasingly demand to inspect source code before adopting a vendor, favoring open platforms that agents and buyers can verify.
Garry says the hard part of building software is now knowing what to build, while vertical SaaS founders struggle to raise Series A rounds at five to ten million dollars ARR.
Why AI agents are launching a million Supabase databases weekly
AI is turning everyone into a software builder, and every new app needs a database. Supabase is betting that agents and non-engineers will pick an open, code-first Postgres platform by default. Because databases almost never migrate once chosen, winning these builders on day one could make Supabase the next AWS-scale backend.
Takeaways
Agents and non-engineers are driving a database explosion. Supabase has about 13m sign-ups and nearly 50m databases under management, up from 8m a year earlier. Over a million new databases launch weekly, and it is now a default database inside Claude Code and Codex.
Openness makes Supabase trustworthy to agents and enterprises. Supabase is open source and helps maintain Postgres itself. Paul Copplestone predicts enterprises will increasingly insist on inspecting source code to confirm security and will avoid closed-source vendors.
Database customers must be won on day one. Migrating databases between providers is nearly impossible, so whoever hosts the first database keeps it. Paul has heard roughly 40 percent of AWS revenue comes from startups that grew up there, a figure he flagged as unverified.
This scale demands architectures databases never needed. Paul imagines ChatGPT's 1.2bn weekly users each launching a database. Most workloads do not need a Ferrari-class database. The Turso acquisition and the Supabase Spark tier aim to carry prototypes all the way to petabytes.
Personal software is already replacing consumer apps. Harj built a control tower for email, finances, health and workouts that he rates at parity with Mint or Monarch. Garry runs GBrain over half a million Markdown files, and a Brazilian taxi driver built his own taxi app.
Knowing what to build is now the bottleneck. Garry argues anyone can build, so judgment about what to make matters most. Boris Cherny envisions simply talking to Claude like a friend. Paul frames the underlying technology as retrieval and calls Supabase an index company.
Vertical SaaS faces a harsh funding market. Garry says vertical SaaS founders at $5 to $10m ARR struggle to raise Series A rounds while horizontal and AI bets get funded. Paul notes a 15-person company at $2m ARR per head and 50 percent margins is very profitable.
Supabase's three chapters
A dashboard company, starting from a product built in about a hundred days during Y Combinator and aimed only at YC startups.
An API company from around 2021, after an integrating customer asked for an API to launch many databases. This suited the 2024 wave of builders such as Bolt and Lovable.
Tap a name to go there · tap a line to see why they are linked
On this episode 3
Show 1
A code-first company, with command-line tools rebuilt to run in sandboxes because agents mostly want CLIs. The target has widened from YC companies to every builder.
How Supabase plans to stay adaptable
Paul argues databases are a durable category but developer tools often stagnate as they grow corporate.
The company dogfoods its own tools, partners across the ecosystem and keeps spare team capacity for bets that may flop.
There is no AI policy. Staff use whatever tools they want, and internal use of coding agents rose sharply within six months.
Much engineering effort now goes into testing specific use cases rather than building new features.