AI Tools Projects Directory
🏆 Leaderboard 📡 News ✨ Prompts ⚖️ Compare Models 🔧 Tools 📦 Projects 🚀 Spaces
Blog
Catalogue data

289 accounts hold half the stars in open-source AI

A ranking of repositories hides the question you actually want answered: not which project, but whose. Rolled up by owner, 11,394 projects become 8,533 accounts — and the shape is not the one the headlines suggest.

Updated 1 November 2026·6 min read·No paywall
A grand hall of identical doors where three are enormous and the rest are ordinary, with a crowd flowing evenly through all of them
Three doors are enormous. Most people still come through the others.

TL;DR — the short version

289 accounts out of 8,533 hold half the stars in the catalogue. That is 3.4 % of the owners. The top 1 % — 85 accounts — hold 29.2 %.

And yet the long tail is the bigger half. 7,215 owners — 84.6 % of them — have exactly one project in the catalogue, and those one-repo accounts hold 34,892,935 stars between them: 54.7 % of everything.

No single owner dominates. The largest, Microsoft, holds 1,301,852 stars across 95 repositories — 2.04 % of the catalogue. The ten largest together hold 8.4 %.

What follows for you: “backed by a big lab” is a weaker signal than it sounds. The median owner in this catalogue holds 1,311 stars, and more than half of all the attention in open-source AI sits with people who published exactly one thing.

Every ranking of open-source AI is a ranking of repositories. That hides the question most people actually want answered, which is not which project but whose. A repository is a thing; an owner is a party you are trusting, a party that can archive, relicense, rename or abandon everything it holds in one afternoon.

So we rolled the catalogue up by owner. 11,394 projects resolve to 8,533 distinct accounts, and the shape that comes out is not the one the headlines suggest.

Share of all stars held by the largest owners Top 10 8.4% 5,347,131 Top 25 15.1% 9,611,502 Top 50 22.2% 14,148,458 Top 100 31.5% 20,126,727 Top 289 50.0% 31,911,195 All 8,533 100% 63,822,390 Cumulative share of 63,822,390 stars · olud.ai catalogue · 1 November 2026
Cumulative share of the catalogue’s 63,822,390 stars held by the largest owners, ranked by total stars. The ten largest hold 8.4 % across 205 repositories; the hundred largest hold 31.5 % across 684. It takes 289 owners to reach half.

Half the stars sit with 3.4 % of the accounts

289 owners hold half of everything. Put the other way: 8,244 accounts share the other half. That is concentration, but it is a remarkably flat kind of concentration compared with what people assume about this field. There is no owner holding a tenth of open-source AI. The largest holds one fiftieth.

The ranking is also not the one you would guess from press coverage. Microsoft leads on 95 repositories, but second place is a single person — Sindre Sorhus, with five repositories and 594,376 stars — and fifth is a project account with exactly one repository to its name.

The twelve largest owners

OwnerStarsReposShare of catalogue
microsoft1,301,852952.04 %
sindresorhus594,37650.93 %
openai568,142240.89 %
huggingface501,403390.79 %
public-apis493,25810.77 %
morluto416,86320.65 %
openclaw390,94140.61 %
nilbuild370,74010.58 %
HKUDS355,761250.56 %
deepseek-ai353,79590.55 %
datawhalechina346,287300.54 %
anthropics338,65120.53 %

Four of those twelve are companies with an AI product to sell. The rest are research groups, community accounts, a Chinese open-courseware collective, and one individual developer. If you built a mental model of open-source AI from funding announcements, this table is not it.

The one-repo majority

Here is the finding we did not expect. 7,215 of the 8,533 owners have exactly one project in the catalogue — 84.6 % of all accounts — and between them they hold 34,892,935 stars, 54.7 % of the total.

Most of the attention in open-source AI does not sit inside portfolios. It sits with people and groups who shipped one thing that worked. That is partly an artefact of how big single hits get: a list, a framework, a model runner can carry half a million stars on its own. But it holds across the distribution, not just at the top, and it is the opposite of the platform-consolidation story usually told about this field.

What to do with this. When you are judging whether a dependency will still be there in two years, the owner’s other repositories tell you something a star count cannot: whether anyone there has done this before. An account with one project is not a warning — it is most of the catalogue — but it does mean the project’s future and the author’s interest are the same variable. Read it next to the contributor count.

Method

All figures come from one snapshot of the olud.ai catalogue taken on 1 November 2026 at 00:03 CET: 11,394 projects carrying 63,822,390 stars between them. The catalogue is a curated subset of open-source AI on GitHub, not a census of it, so every share quoted is a share of that catalogue and not of the whole forge. Star counts move daily and the figures here will drift; the live numbers are on each project’s page in the catalogue.

Owners are the account part of the repository path exactly as the forge reports it, with no attempt to merge accounts that belong to the same organisation — a company publishing under two handles counts as two owners here, which makes every concentration figure in this article a slight under-statement. Shares are of total stars, not of repository counts. The “289 owners” figure is the smallest number of owners, taken in descending order of stars, whose combined total reaches 31,911,195 stars.

Data: olud.ai catalogue, refreshed nightly — method on /methodology.html.

Keep reading