Supabase Raises $150M and Acquires Turso for Agent Databases

Supabase has raised $150 million in a round led by Singapore's GIC and agreed to acquire Turso, the SQLite-compatible database company, SiliconANGLE reported on October 2, 2026. The deal arrives with a striking usage claim: Supabase says 70% of new databases on its platform are now created by agents or AI-driven tools, not by people. This piece checks the announced facts against Supabase's own posts, works through what the numbers imply, and compares the move with the agent database strategies of its closest rivals, treating "agent databases" as a distinct infrastructure category.

$150MNew roundLed by GIC, announced Oct 2, 2026

70%Agent-createdShare of new Supabase databases

4MDatabases a monthReported, plus 1M+ new users

4 monthsAfter Series F$500M raised June 4, 2026

What was announced

According to Supabase's own announcement, the company is acquiring Turso, with Turso founder Glauber Costa and co-founder Pekka Enberg joining Supabase and Costa leading the agentic infrastructure effort. The post says Supabase is launching more than one million databases per week and that "agents are spinning up millions of databases to power the prototypes, explorations, dashboards, and apps they're building." It also says that for existing users "nothing changes" and that Turso will keep operating.

SiliconANGLE's report adds the funding details: the $150 million round was led by GIC, with Alphabet's CapitalG, IronArc and SquarePeg participating, and the proceeds are earmarked for employee liquidity and new agent features. The same article describes Supabase Compute, launched alongside the deal as "hosted sandboxes for long-running AI agents." Supabase's own product post says Compute runs long-lived services and agents next to a project's database, with no limit on run time, configurable memory and CPU, and a full Linux environment. The same page says Compute is in private alpha, with a waitlist for access. The purchase price and closing date have not been disclosed, according to RuntimeWire.

Why agents create so many databases

The 70% figure is easier to interpret with the underlying volume. Reported coverage puts Supabase's intake at more than 1 million users and 4 million databases a month. If 70% of 4 million are agent-created, that implies roughly 2.8 million databases a month, or around 90,000 a day. That is an inference from two reported numbers, not a figure Supabase published, and counting methods were not defined in the coverage.

The trend has a visible trajectory. In its Series F post in June 2026, Supabase said database launches had grown 600% over the prior year and that over 60% of new databases were launched by AI tools. Four months later the stated share is 70%. Neon, the serverless Postgres company, reported an even higher share earlier: Databricks said in May 2025 that over 80% of databases provisioned on Neon were created automatically by AI agents. The percentages are not directly comparable, since each company measures its own telemetry, but all three point the same way.

One structural cause is the way app-building tools work. Lovable's documentation describes a native Supabase integration in which its agent designs the schema, runs migrations and wires the interface to a project's database from a chat prompt. When a coding agent builds an app for someone who never opens a database console, the agent is the entity that provisions the database. Each prototype, dashboard or experiment can become its own database.

Agents are spinning up millions of databases to power the prototypes, explorations, dashboards, and apps they're building.
Supabase, announcing the Turso acquisition

Postgres versus SQLite per agent

Supabase's product is managed Postgres. Turso's is a Rust rewrite of SQLite. The two have different cost shapes, and the acquisition reads as an attempt to cover both. Supabase describes Turso's cloud platform as one where "a single server can manage millions of databases, loading them when needed and suspending them when they're not." Turso's own post describes a "diskless, WAL-on-S3 architecture," removal of SQLite's single-writer limitation, and a goal Costa states as serving "upwards of a billion databases." It also says users get a path to Postgres through Supabase when an app outgrows SQLite.

The SQLite logo. SQLite is the embedded database engine that Turso's open-source database reimplements in Rust. Image: SQLite project, public domain, SVG conversion by Mike Toews.

The economic logic, offered here as analysis rather than reported fact, comes down to what a database costs while idle. A database that is a file loaded on demand has a near-zero cost when nobody is using it, while a database that is a running server process carries a baseline cost per instance. When the number of databases is in the millions and most are touched rarely, that baseline dominates. Supabase's stated design goal, in the same announcement, is that databases should be "cheap to create, available on demand, and have a clear path to production when needed." Read together, the plan is SQLite-style databases for the exploratory long tail and Postgres for what graduates to production.

This is the database-per-agent pattern: rather than many agents sharing one schema with row-level separation, each agent, task or user gets an isolated database. Isolation simplifies permissions and cleanup, and a bad write affects only that database. The trade-off is operational: fleets of small databases need provisioning, migration and observability tooling that single large databases do not.

The funding trajectory

Supabase's valuation has risen steeply across four rounds. Per Orrick's note on the Series E, it raised $200 million at a $2 billion valuation in its Series D earlier in 2025, then $100 million at $5 billion in the Series E, led by Accel and Peak XV. Supabase's Series E post dates that round to October 3, 2025. Its Series F post dates the $500 million round to June 4, 2026, led by GIC at a $10 billion pre-money valuation.

Supabase valuation by round

Reported valuation in billions of US dollars at each of the last three priced rounds. The October 2026 round of $150M did not disclose a valuation and is not shown.

Series E and F figures are pre-money as stated by Supabase; the Series D figure is as reported by Orrick, which does not specify the basis. Series F is $10.5B post-money per other coverage.

Some arithmetic helps. The Series F valuation is double the Series E in about eight months. The new $150 million is 30% of the Series F size and landed 120 days after it, with employee liquidity as a stated use. The company did not disclose a valuation for this round, so no step-up can be computed. The repeat lead from GIC, which SiliconANGLE describes as Singapore's sovereign wealth fund, signals continued support from an existing investor, though the terms are not public.

The competitive landscape

The shape of the market became clearer a year earlier. Databricks agreed in May 2025 to acquire Neon for about $1 billion, citing the agent-driven share of Neon's databases. PlanetScale, long associated with MySQL, announced general availability of PlanetScale for Postgres, putting a third managed Postgres option in front of the same developers. Postgres itself is the common denominator: it was used by 55.6% of respondents in Stack Overflow's 2025 Developer Survey, per a secondary summary.

Supabase's acquisition differs from Databricks' in one respect worth noting. Neon extended a Postgres architecture with fast provisioning. Supabase bought a different engine, SQLite, and kept Postgres as its production tier. That gives it two storage shapes under one developer experience, at the cost of maintaining two engines.

What to watch

Closing and pricing. Neither company disclosed a price or closing date, and the open-source status of Turso Database was reaffirmed in Turso's post.
Compute availability. Supabase Compute is in private alpha, so general pricing and limits are not yet public.
Definition of "agent-created." The 70% share depends on how Supabase attributes creation to agents or AI tools, which was not specified in the coverage.
Security hygiene. A fleet of machine-created databases raises the stakes of defaults such as access rules, since nobody reviews each one by hand.
Convergence. Whether Neon, PlanetScale and others add lightweight per-agent tiers, and whether SQLite-style engines gain a production foothold.

For teams building multi-model agent workflows, including those on a model-agnostic workspace such as Metir, the practical takeaway is that the data layer is becoming as swappable and agent-driven as the model layer.

Sources:

Image credits

Hero: "Skylines of the Central Business District, Singapore at dusk" by Basile Morin, licensed CC BY-SA 4.0, via Wikimedia Commons. It shows the Singapore skyline and Marina Bay, and is used as a representation of GIC's home city; it does not depict GIC, Supabase or Turso.
In-body: "SQLite370.svg," part of the SQLite documentation released to the public domain by D. Richard Hipp, SVG conversion by Mike Toews, via Wikimedia Commons. Public domain.