public.diagnostictable: diagnostic_questions
0011playbook
Is your tool Supabase-shaped?
Eight questions. Answer yes or no, and see which lessons apply to you.
next: public.origins
public.origins
0001RLS on6 rows
A workaround that became a company
Supabase started as a fix for one CTO's chat app.
Paul Copplestone was co-founder and CTO of Nimbus for Work, an office-management startup in Singapore. When its chat feature hit Firestore's limits (too slow, queries too limited), he moved the data to Postgres and wrote an open source real-time layer for it. Developers wanted it far more than he expected. He read Hacker News about three times a day, watched the Show HNs, and saw Postgres trending. In January 2020 he and Ant Wilson spent the month interviewing developers before building Supabase.
· loading
storage / evidence/hacker-news-2020.jpgimage/jpeg · 117 KB
NOTICE: Hacker News thread, May 2020Firebase has no true competitor, let alone an open source one.Hacker News commenter
next: public.taglines
public.taglines
0002The one that mattered
Five words and a launch
The first pitch was "real-time Postgres." Accurate. Technical. Ignored.
8
hosted databases before the change
800
within three days
"The open source Firebase alternative" named a tool developers already used and the thing they feared about it: lock-in. Copplestone changed the line, and the next day an early user posted Supabase to Hacker News, where it reached the top. The two can't be separated, but Copplestone credits the positioning: "product-market fit often is just like a product-positioning fit."
· loading
storage / evidence/tagline-switch-may-2020.jpgimage/jpeg · 134 KB
Why not Appwrite or Nhost?
Supabase wasn't the only open source Firebase alternative. Appwrite's repo is six months older and pitched itself the same way, Nhost did too, and PocketBase went after the same developers. All four are open source, and Claude names all four. Only one became the default. The differences in the data: Supabase sat on a full Postgres database, had 61 Hacker News stories over 100 points to its rivals' 2 to 5, and is the only one Claude picks first.
· loading
One snapshot, so read it as correlation, not proof: timing, funding and teams differ too. PocketBase shows a different path: 62K stars for a single-file backend aimed at solo developers.
Borrowing a category gets you considered. Becoming the default took a standard people already trusted and years of showing up in public.
next: public.postgres_bet
public.postgres_bettable: hn_stories
0003MIT / Apache 2 / PostgreSQL
Bet on the boring standard
No new database. Every project is a full Postgres database.
· loading
Copplestone timed AWS himself at the start, and never said the number out loud.
Why Postgres: the game theory
Nobody owns Postgres, so everyone contributes to it. It has been the most-used database in the Stack Overflow survey every year since 2023, so every hyperscaler has to offer the best Postgres it can, which makes Postgres better still. Copplestone calls it game theory. Supabase bet on that flywheel instead of building a database of its own.
NOTICE: YC Startup School Paris, 2026No one owns Postgres, which means that everyone contributes to Postgres.Paul Copplestone, CEO
storage / evidence/github-2026.jpgimage/jpeg · 127 KB
Win Hacker News by teaching Postgres
We pulled every Hacker News story with Supabase in the title or link since 2020, about 700 of them. Of the 61 that reached 100+ points, 29 never say Supabase in the headline. They teach Postgres: a language server, Postgres in the browser, full-text search. Copplestone posted 18 of the 61 himself, including two of the top three.
· loading
· loading
next: public.launch_weeks
public.launch_weekstable: launch_week_star_lift
00043.8x → 1.0x
Launch weeks worked. Then they stopped.
Supabase popularized launch weeks in dev tools; Resend, Trigger.dev and Dub now run their own. We measured each one's effect on GitHub stars against the four weeks before it, skipping the week just before. Average lift fell every year, from 3.8x in 2021 to 1.0x in 2025.
· loading
-- the detail
The first Launch Week started on 29 March 2021: one launch a day for five days, a way to recreate the pressure of an accelerator inside the company. Copplestone saw "these 20% bumps each time." Stars flatten as a repo gets big, so the honest reading is: the ritual kept paying; it stopped moving stars.
· loading
storage / evidence/launch-week-2021.jpgimage/jpeg · 74 KB
· loading
NOTICE: for founders copying thisLaunch weeks win attention early. At scale, they serve the users you have.From star data, 2021 to 2025
next: public.revenue
public.revenue
0005
CHARGE WHEN IT'S REAL
· loading
next: public.ai_waves
public.ai_waves
000670% agent-created
The machines picked it too
Supabase didn't pivot to AI. AI app builders and coding agents picked it, after years of being the fastest way to start a Postgres project.
· loading
In late December 2024, Bolt and Lovable were creating databases so fast that Supabase said it was getting DDoSed. The team debated building its own app builder to own the distribution, and doubled down on Supabase for Platforms instead. Copplestone says over 50 companies now build on it.
NOTICE: Series F announcement, June 2026Agents are now deploying the majority of databases on our platform.Paul Copplestone, CEO
· loading
Then the unusual part: after the holidays, with Claude Code taking off, users accelerated while activation and conversion went up too. Copplestone says they can measure 60% of databases as agent-launched and suspects the real figure is closer to 90%. In October 2026 Supabase put it at 70% of new databases.
· loading
We asked Claude 1,020 times
We asked three Claude models (Opus, Sonnet and Haiku 5.5) 34 questions a developer might ask, 10 times each, then had a judge model mark the first recommendation. The study was pre-registered before the first call. Supabase wins the questions it built its name on and loses the ones it didn't: AI apps get plain Postgres, and Postgres hosting goes to Neon.
· loading
Then we gave Claude Code five apps to build in an empty folder, with no network and no accounts.
· loading
next: public.rounds
public.rounds
0007$1B+ raised
The receipts
Over $1B raised since 2020. The valuation went from $2B to $10.5B in about 13 months.
· loading
Sacra estimates ARR grew from about $30M at the end of 2024 to about $170M in May 2026 (third-party estimates; Supabase doesn't disclose revenue). Copplestone says over 60% of each YC batch now builds on Supabase. In October 2026 Supabase announced it is acquiring Turso, terms undisclosed.
Next bet, per Copplestone: self-driving databases for the operate stage, so nobody wakes up at night to a database that's down. Building is easy to copy; operating is the hard part.
on why cash isn't the moatWe should not exist if AWS are so good.Paul Copplestone, on why cash isn't the moat
next: public.principles
public.principlestable: lessons
00090/… shipped
The playbook, by stage
None of these lessons costs much to run. Every lesson, in the order you'll need it. Mark one as shipped in its block and the counter here updates.
-- show the query
· loading
next: public.dont_copy
public.dont_copy
00116 warnings
What not to copy
Every schema has its warnings. These are the ones nobody puts on a slide.
HINT: make the safe path the fast path. Enable RLS on every table.
next: public.templates
public.templates
0011playbook
Steal the templates
Copy these into Notion, Linear or a doc. They're MIT licensed, like everything here.
next: public.sources
public.sources + faq
0010seed
Frequently asked
Three compounding moves. A new tagline, "the open source Firebase alternative", went live the day before an early user's Hacker News post reached the top, and hosted databases went from 8 to 800 in three days. Relentless shipping, packaged as Launch Weeks, which Copplestone saw give "20% bumps each time" (their effect on GitHub stars faded every year). Then AI app builders and coding agents adopted Supabase as their default backend; by October 2026, 70% of new databases were created by agents or AI tools.
If developers already know the incumbent and dislike something specific about it, yes. "X alternative" borrows a category people already understand and names the reason to switch in the same breath. Supabase later outgrew the comparison and became "the Postgres development platform". Treat your tagline as an experiment, not a brand asset.
The product did the marketing. Copplestone timed AWS (about 8.5 minutes from a new database to a first row) and got Supabase to about 5 seconds. The team polled developers on Hacker News, Reddit and Twitter, and turned releases into Launch Weeks. In his words: "if I could give a free database, I'd rather give that to developers than spending on marketing."
Not for Supabase. Everything it builds is open source except the platform and billing code, the part it takes to host millions of databases. Customers pay for hosting, scale and not being on call. By Copplestone's rough recollection, revenue took about 18 months to start, reached roughly $1M ARR by the end of the first year of charging, then about $7.5M the year after.
One setup step gives a full Postgres database with auth, APIs and storage, and years of open source code, Reddit threads and GitHub discussions meant models already knew it well. Supabase then built for that traffic: Supabase for Platforms for app builders, and a CLI and MCP server for agents. By October 2026, Supabase said 70% of new databases were created by agents or AI tools.
There's no trick, and Supabase's CEO says part of it is luck: if agents settle on about three defaults per category, most tools won't be one. What you control: get developers to a result fast with as little input as possible, make the product fully operable from a CLI and MCP, publish clean docs, and answer questions in public where models can learn from them.
Supabase's CEO puts it bluntly: "we should not exist if AWS are so good. They've got way more cash than us." Cash wasn't the deciding factor. Developer experience, open source trust and community were. When Copplestone timed it at the start, AWS took about 8.5 minutes from a new database to a first row; Supabase now takes about 5 seconds.
Time your onboarding against your main competitor with a stopwatch, then cut it until you're an order of magnitude faster. It's cheap, it compounds, and Supabase never even advertised theirs. Second: describe the ideal customer whose usage grows into big bills, and name 20 of those teams. Launch weeks come later, and measure them in signups, not stars.
Sources
Usage metrics are company-reported. Revenue figures are third-party estimates. Star, Hacker News and homepage data is our own analysis of public sources, collected 17 Oct 2026.
- 01Contrary Research: Supabase
- 02First Round Review: Paul Copplestone (Jun 2026)
- 03YC Startup School Paris: Paul Copplestone (2026)
- 04Hacker News: the May 2020 post
- 05Supabase blog: Alpha launch postmortem (Jul 2020)
- 06Supabase blog: Launch Week, March 2021
- 07Supabase blog: How we launch
- 08Supabase blog: 1,000 YC companies (Oct 2025)
- 09TechCrunch: $6M seed (Dec 2020)
- 10TechCrunch: Series C (Sep 2024)
- 11Fortune: Series D (Apr 2025)
- 12PR Newswire: Series E (Oct 2025)
- 13PR Newswire: Series F (Jun 2026)
- 14TechCrunch: Series F and Claude Code (Jun 2026)
- 15PR Newswire: $150M and Turso (Oct 2026)
- 16Sacra: revenue estimates
- 17Matt Palmer: statement on CVE-2025-48757
- 18Stack Overflow Developer Survey 2026
- 19OSS Insight: supabase/supabase star history
- 20Hacker News search API (Algolia)
- 21Wayback Machine: supabase.io, 2020
- 22Supabase blog RSS (launch week dates)
- 23github.com/supabase/supabase
next: you
Don't trust our numbers? Check them yourself. Real Postgres, in your browser.
public.psql · case_study
PGliteruns in your browser
public.authors
1 row