Nigeria's GitHub Developers Are 7.8x More Likely to Tag 'Web3' Than Anyone Else

Nigeria makes up half a percent of the geo'd developers in our GitHub index. Among developers who tag themselves as working in web3, it's nearly one in twenty — a 7.8x over-index, the sharpest skew we found anywhere in the dataset. That's a bigger number than the "India over-indexes in AI" story you may have already heard — including from us, in an earlier pass on this same dataset that got the math wrong.

We tagged 190,589 GitHub developers by self-reported specialty — backend, frontend, devops, mobile, security, data, ml-ai, web3, gamedev — and cross-referenced each one's geocoded location, self-reported employer, follower count, and "open to work" flag. The map that comes out is not the one you'd guess from headlines about any single country's AI boom.

A small, self-tagged slice of GitHub

Only 190,589 of the 1,553,663 developers in our index — 12.3% — set an interest-domain tag on their profile at all. Every number below describes that self-selected slice, not GitHub as a whole. Within it, backend (46,940) and ml-ai (40,544) are the two largest tags, ahead of frontend (38,468), devops (20,773), mobile (14,853), security (11,627), data (7,388), web3 (6,106), and gamedev (3,890).

Geo resolution is partial and uneven by domain too — 55–72% of domain-tagged developers resolve to a country (web3 lowest at 55.1%, devops highest at 71.9%), so every share below is "of the developers we could place," not of the tag as a whole.

India rules the generalist categories. The US rules everywhere else.

Here's the full picture: each domain's over-index ratio for seven countries, where 1.0x means that country's share of the domain matches its share of the general geo'd population, and anything above 1.0x means it's punching above its weight.

Domain ↓ / country over-index →	US	India	Germany	UK	Brazil	China	Nigeria
Backend	0.53x	2.83x	0.73x	0.67x	2.1x	0.51x	2.71x
Frontend	0.57x	1.97x	0.67x	0.74x	1.9x	0.38x	4.02x
ml-ai	0.97x	2.4x	0.82x	0.9x	0.67x	0.75x	1.51x
Devops	0.86x	1.83x	1.05x	1.02x	1.12x	0.54x	2x
Data	1.08x	2.09x	0.74x	0.92x	1.32x	0.33x	1.76x
Security	1x	1.48x	0.95x	1.02x	0.81x	0.47x	2.52x
Mobile	0.64x	1.53x	0.91x	0.83x	1.39x	0.83x	3.11x
Web3biggest skew	0.77x	1.47x	0.7x	0.89x	0.84x	0.62x	7.82x
Gamedev	0.84x	0.53x	0.89x	1.21x	1.17x	0.72x	0.44x

Two patterns jump out. First: India is #1 by raw share only in backend and frontend — the highest-volume, least-specialized tags — at 23.1% and 16.1% of each domain's geo'd population respectively. In every other specialty, the US leads, and India's own over-index ratio drops well below its backend/frontend numbers; in gamedev it dips under 1.0x entirely (0.53x), the one domain where India is under-represented relative to its own baseline.

Second, and less expected: Nigeria over-indexes in eight of nine domains, not just web3 — frontend (4.02x), mobile (3.11x), security (2.52x) and backend (2.71x) all clear 2.5x. Web3 (7.82x) is still the outlier by a wide margin — more than 90% higher than Nigeria's next-highest domain — but the pattern isn't "Nigeria is a web3 story." It's a country whose open-source presence skews younger and more broadly specialized than its general GitHub footprint, with one domain, gamedev (0.44x), as the sole exception where it under-indexes like India does.

Here's the same grid as raw share-of-domain percentages, for anyone who wants the receipts without hovering a chart:

Domain	US	India	Germany	UK	Brazil	China	Nigeria
Backend	13.6%	23.1%	4.6%	3.9%	8.7%	2.8%	1.5%
Frontend	14.7%	16.1%	4.3%	4.3%	7.9%	2.1%	2.2%
ml-ai	24.8%	19.6%	5.3%	5.2%	2.8%	4.2%	0.8%
Devops	22.1%	15.0%	6.7%	5.9%	4.7%	3.0%	1.1%
Data	27.6%	17.1%	4.7%	5.4%	5.5%	1.8%	1.0%
Security	25.6%	12.1%	6.1%	5.9%	3.4%	2.6%	1.4%
Mobile	16.4%	12.5%	5.8%	4.9%	5.8%	4.6%	1.7%
Web3	19.7%	12.0%	4.5%	5.2%	3.5%	3.4%	4.3%
Gamedev	21.5%	4.3%	5.7%	7.0%	4.9%	4.0%	0.2%
General baseline	25.6%	8.2%	6.4%	5.8%	4.2%	5.5%	0.6%

The single biggest skew: Nigeria and web3

Nigeria's web3 number deserves its own look, because it isn't a small-sample fluke. Of 3,364 web3-tagged developers we could geolocate, 145 are in Nigeria — 4.31%, against Nigeria's 0.55% share of the general geo'd population (2,818 of 511,379). That's a 7.8x over-index on a base large enough to trust, and the US — which leads every other specialty in this dataset — only over-indexes web3 at 0.77x, meaning web3 is actually under-represented in the US relative to the US's own general GitHub footprint.

This isn't a pattern invented by our crawl. Chainalysis, which tracks on-chain transaction volume by country, ranked Nigeria #2 worldwide in its 2024 Global Crypto Adoption Index (behind India), driven substantially by peer-to-peer trading volume — Nigeria's ranking has since moved (6th in the 2025 edition, as regulatory tightening reshaped how P2P volume gets measured), but the underlying grassroots adoption story is well documented independently of GitHub. What our dataset adds is a second, unrelated signal pointing the same direction: it isn't just trading activity — Nigerian developers are disproportionately building web3 tooling in the open, not only using the assets.

We caught our own mistake

An earlier pass at this same dataset claimed India over-indexes 4.8x in ml-ai — comparing India's 13.0% share of ml-ai-tagged developers against India's 2.7% share of all 1,545,080 developers in the index, geocoded or not. That's not a fair comparison: most of that whole-population denominator was never geocoded in the first place, so it understates every country's "true" share and inflates every ratio built on top of it.

Originally reported (wrong denominator)4.8x

Corrected (geo'd-population denominator)2.4x

Rebuilt with matching denominators — India's 19.6% share of geo'd ml-ai developers (5,267 of 26,873) against India's 8.2% share of the geo'd general population (41,773 of 511,379) — the ratio is 2.4x, exactly half the original claim. It's still a real, citable over-index. It just isn't twice as large as we first said. We're flagging our own correction here rather than quietly fixing it, because a plausible-looking ratio can be exactly wrong for a boring reason, and the fix is worth showing, not just applying.

Employers flip character by specialty

Self-reported employer tells a sharper story than country does. In some specialties, GitHub's top self-reported employers are almost entirely hyperscalers; in others, Big Tech barely shows up at all.

ml-ai77.6%Microsoft, Google, Nvidia, Amazon, Meta, Apple

Security68.5%

Devops46.8%

Web327.9%

Gamedev17.3%Unity dominates instead

ml-ai's top employers are Microsoft (243 self-reported mentions), Google (159), Nvidia (135), and Amazon (109) — dedicated AI labs like Hugging Face, Mistral AI, and Scale AI all sit in the teens, well behind the hyperscalers. Devops looks similar: Microsoft (250) and Red Hat (154, versus only #6 in the general-population employer list) lead, alongside Cloudflare (115).

Web3 and gamedev break the pattern entirely. Once free-text variants of the same name are merged — "ethereum" and "ethereum foundation," "unity technologies," "unity," and "unity-technologies" all get typed as separate strings by GitHub's own profile field — a different picture emerges:

#	Web3	Mentions	Gamedev	Mentions
1	Ethereum (+ Ethereum Foundation)	50	Unity (all variants)	54
2	Amazon	13	Google	7
3	Microsoft	13	Microsoft	7
4	Consensys	12	NetEase	4
5	Google	10	Epic Games	3

"Ethereum," combined, outnumbers Amazon and Microsoft combined (50 vs. 26). Unity's own name variants outnumber Google and Microsoft combined by roughly 3.9x (54 vs. 14). Two of nine domains almost entirely escape Big Tech's employer gravity, and — tellingly — they're the two most community- and studio-driven ones, not the two smallest.

Specialists build bigger followings, and everyone's more "hireable"

Two smaller, related findings round this out. First: developers who tag a specialty skew toward larger followings than the general population, and it isn't flat across specialties — ml-ai's "macro" influencer tier (10,000+ followers) is 3.3x over-represented relative to the general population (0.158% vs. 0.048%), and web3 skews furthest of any domain we checked: only 74.6% of web3-tagged developers sit in the bottom "nano" tier, versus 90.3% generally, and its non-nano share (25.4%) is more than 2.5x the general population's 9.7%.

Second, and more counterintuitive: we expected web3's higher "hireable" rate to be a web3-specific contractor-culture signal. It isn't.

General population10.8%

ml-ai23.8%

Gamedev24.1%

Web324.2%

Devops27.2%

Every domain we checked sits in a tight 23.8%–27.2% band — more than double the general population's 10.8%, but essentially flat across specialties. Devops (27.2%) edges out web3 (24.2%). The interesting story isn't "web3 developers are more available for hire" — it's that self-tagging any specialty at all is a marker of active career-branding behavior on GitHub, roughly uniformly, regardless of which specialty you pick.

What this shows, and what it doesn't

What we think holds up: which country leads a GitHub specialty, and by how much, varies sharply and non-randomly by domain — India's strength concentrates in the highest-volume, least-specialized tags; the US leads every genuinely specialized one; and Nigeria over-indexes broadly, peaking hardest in web3, in a pattern independently corroborated by Chainalysis's crypto-adoption tracking.

What limits it:

• This is a self-selected 12.3% slice — only 190,589 of 1,553,663 developers in the index set any interest-domain tag. Every finding describes "developers who self-tag a specialty," not GitHub or open source as a whole.
• Geo resolution is partial and domain-uneven (55–72% by domain; ~32.9% for the general population, and that figure is itself a lower bound — see the next point).
• Our own country facet is capped at roughly the top 50 countries by volume, confirmed directly: Iceland has 272 geocoded developers in our index but never appears in any top-50 country list we pulled. Every percentage above is built against that top-50 sum, not a true total. This biases raw shares slightly upward and consistently on both sides of every ratio, so the ratios hold; the standalone percentages are approximate.
• Free-text employer fields aren't deduplicated by GitHub — near-duplicate strings for the same company (huggingface/hugging face, unity/unity technologies/unity3d, ethereum/ethereum foundation) have to be merged by hand, as we did above, before ranking them.
• Domain tag, employer, and "hireable" are all unverified, self-reported, voluntary fields — subject to staleness and to the selection bias of who bothers to fill out a GitHub profile at all.
• We publish aggregates only — domain-by-country and domain-by-employer counts, not individually identifying developer records.

None of this claims any single country "owns" open source, or that Instagram-style verification bias exists on GitHub. It says something narrower and measurable: self-reported specialty tags cluster geographically in ways that track real-world signals we didn't manufacture — and a ratio built on a mismatched denominator can look twice as dramatic as it should.

Sources

Methodology

We built a search index of 1,553,663 GitHub users, each carrying geocoded location (where resolvable), self-reported company, follower count, an "open to work" flag, and — for the 190,589 who set one — an interest-domain tag (backend, frontend, devops, mobile, security, data, ml-ai, web3, or gamedev). The full corpus is queryable through the GitHub users dataset via its search and facets endpoints; every cross-tab in this post reconciles its country, company, influence-tier, or hireable counts back to that domain's flat facet total.

Country and employer shares were computed by filtering the facets endpoint to each of the nine domain tags in turn, then dividing by the sum of that domain's country (or company) facet response. Because the facets endpoint caps at roughly the top 50 buckets by volume regardless of requested page size, two specific cells (Nigeria in ml-ai and in gamedev, where its count fell outside that top-50 window) were confirmed with a direct filtered search call instead, using its exact total field. Over-index ratios divide a domain-specific share by the equivalent general-population share, built the same way, so both sides of every ratio share the same denominator methodology. We publish aggregates only — domain-by-country and domain-by-employer counts, no individual developer records.

Want to run your own cut? The GitHub users dataset is queryable by domain tag, country, company, influence tier, and hireable status, and pricing has the free tier. See also the companion study on how seed composition — not platform bias — explains verification-rate gaps: a politician on Instagram is 3x more likely to be verified than a photographer.