First, a fact-check. The source material claimed Supabase had 90,000 GitHub stars. We queried the GitHub API: 111,076 stars, measured October 4, 2026. The 90,000 figure was last year's snapshot—the project passed 100,000 stars this April and now stands at 111K, with 15,000 forks and 1,893 contributors (April count). The repository was created in October 2019, is seven years old, and is Apache 2.0 licensed.
But the star count isn't this project's biggest story. The biggest story hides in the self-introduction it changed.
The Retirement of a Slogan
When Supabase launched in 2020, its official positioning was "open-source Firebase alternative." Firebase is Google's backend suite; developers disliked its lock-in and proprietary NoSQL. Supabase said: same out-of-the-box backend, but built on Postgres, with open protocols—leave anytime. A smart piggyback strategy.
Visit the repository homepage today and that phrase is gone. In its place: "The Postgres development platform." Firebase is no longer its reference point; in comparison pieces, Firebase is now the one being compared. A September 2026 landscape summary put it neatly: Firebase owns mobile, Supabase owns SQL-first web SaaS, Neon owns serverless Postgres, Appwrite owns self-hosting.
When an alternative grows big enough, it removes "alternative" from its name. This is the endpoint of all piggyback narratives: the reference point becomes a peer.
What It Actually Wraps
Back to the BaaS itself. Supabase's wrapping philosophy deserves mention because it barely reinvents anything.
The database is Postgres itself, one dedicated instance per project. The REST API comes from PostgREST, a decade-old standalone Haskell project that turns any PG database into an API automatically. Auth uses GoTrue, adapted from Netlify's open-source project. Realtime push is an Elixir service that listens to Postgres's built-in replication stream, converting every row insert/update/delete into websocket messages. File storage connects to S3-compatible object storage. The edge functions engine, Edge Runtime, is written in Rust with Deno at its core, running TypeScript and WASM.
The philosophy in one line: if it exists in the Postgres ecosystem, don't rewrite it. "Powerful PostgreSQL foundation" isn't marketing—it's the load-bearing wall of the entire architecture. Data isn't stored in a proprietary format; it's just standard PG tables. This is the biggest difference from Firebase, and the technical backing for the "leave anytime" promise.
Why 2025
Going viral needs a recipe. The 2025 recipe: AI broke through the cost of writing frontends. Lovable, Bolt.new, v0, Cursor—type a sentence, get an app interface. The frontend became free, and the backend became the bottleneck—behind every interface there's login, a database, file uploads. Building that by hand takes days, and AI apps can't wait.
Supabase caught it. Nearly all these AI site builders integrated one-click Supabase. AI generates the frontend, Supabase carries the backend—idea to launch with zero backend code written. This combination became the standard recipe of 2025.
The ledger followed: October 2025, Series E, $100 million, led by Accel and Peak XV, at a $5 billion valuation. Nine months later, June 4, 2026, Series F, $500 million, led by Singapore sovereign fund GIC with Stripe participating, $10.5 billion post-money. The official figure: nearly 10 million developers. An open-source project taking sovereign fund money is rare.
The Customer Changed
The most informative line in the Series F coverage came from a startup media outlet: AI agents, especially Claude Code, now deploy the majority of databases on Supabase. CEO Copplestone gave an even starker number in a public interview: more than 1 million new Postgres databases deployed per week, with manually created ones approaching zero (interview claim, single source—note that).
This line deserves a ten-second pause. Databases used to be standard equipment for a "project"—one project, one database, and creating one was the opening ceremony. Now the unit of database creation is the agent session: an agent receives a requirement, spins up a database, runs it, the session ends, and the database is kept or discarded. A million a week—infrastructure becomes a consumable.
That week's Series F coverage headline put it more bluntly—on Supabase, the customer is now an agent.
Counter-Evidence Takes the Stage, As Usual
37 hours. An August 2026 Hacker News thread: a company's production database was down for 37 hours due to a billing error, with no self-service recovery option—no restart, no restore. One-sided account, unverifiable, but plenty of resonance in the comments. As scale grows, operational weaknesses are starting to show.
The second issue is code quality. A frequent reminder in the vibe coding community: in AI-generated apps, auth is the component most prone to silent failure—it runs, but the permission model may be wrong. Supabase's Row Level Security (RLS) puts permission policies at the database layer, with every row constrained by policy. Good design—but only if the policies are written correctly. AI is great at creating tables, not so great at writing RLS policies. The gap between declared permissions and row-level enforcement is where the next batch of incidents will come from.
Back to the Main Thread
This is a new tier on the deployment-collapse spectrum. The collapses we've documented before were on the inference and training sides; this one is on the infrastructure side: from idea to "app with a backend" used to be weeks of engineering—now it's part of an agent session.
What hasn't collapsed is equally clear: verification. The 37-hour outage, miswritten RLS, silently broken auth—all bills charged to verification bandwidth. In the era of free databases, operations and data contracts are the new bottleneck.
The star count will keep climbing. But what to watch next isn't stars—it's whether this project can handle the operational bill of a million databases a week.
---
Customs Verdict Table
| Source Claim | Verification | Verdict | |---|---|---| | 90K stars | GitHub API measured 111,076 (2026-10-04); 100K milestone reached this year | Old snapshot, conservative | | Auth/storage/dashboard wrapped on PostgreSQL | Matches official architecture docs item by item (GoTrue/Storage/Studio) | Confirmed | | One of GitHub's top open-source projects | Now in GitHub's Top 100 most-starred | Confirmed | | Hottest backend open-source project of 2025 | Series E $5B (2025-10) → Series F $10.5B (2026-06); default integration by AI site builders | Direction confirmed, momentum continuing into 2026 |
Sources: GitHub API; TechCrunch/e27/techfundingnews (funding); Supabase official architecture docs and Changelog; HN August 2026 outage thread (one-sided); ai.engineer (developer count). Series F details—GIC lead + Stripe participation + $10.5B post-money—cross-confirmed by four sources. "1 million databases per week" is the CEO's interview claim, flagged as single-source.