EUPL License Adoption: 2.5x Run-Rate on GitHub in 2026
Henri Bergius, creator of Midgard CMS and NoFlo, published "I changed my license" yesterday: after 28 years across LGPLv2 and MIT, his default is now EUPL-1.2, the EU's strong copyleft. His reason holds one word that matters. Corporations that dislike his terms are "free to spend the effort or tokens to build their own." Tokens. The permissive bargain assumed rebuilding was expensive; LLMs collapsed that price. One anecdote — so I measured it: 2,704 EUPL-1.2-licensed repos hit GitHub in 2026 through Sep 6, versus 1,569 in all of last year — an annualized run-rate near 3,980, 2.5x 2025.
How fast is EUPL adoption growing on GitHub?
New EUPL-1.2 repos by creation year (GitHub Search API, Sep 6, 2026):
| Year | New EUPL-1.2 repos | Change |
|---|---|---|
| 2023 | 1,060 | — |
| 2024 | 1,200 | +13% |
| 2025 | 1,569 | +31% |
| 2026 (248 days, annualized) | 2,704 → ~3,980 | +154% |
The controls: GPL-3.0, the ordinary strong copyleft, annualizes to +24% over 2025 — baseline growth. AGPL-3.0, which closes the SaaS loophole, annualizes to +134%; EUPL closes the same loophole and adds official translations in 23 EU languages, annualizing to +154%. The accelerating licenses are the ones built to bind network users, not distributors.
Why are permissive-license developers switching to strong copyleft?
Bergius puts the mechanism plainly: open source won, and "all that our efforts did was to make it easier for big corporations build things more cheaply." Cheap adoption was the 2010s story; cheap generation is the 2026 one — a substitute for adoption entirely. When a model produces a workalike for the price of tokens, a license keeps only what follows the software into deployment: copyleft that survives the network boundary. His EUPL portfolio already covers a Reticulum mesh stack and a decentralized authorization system — where SaaS enclosure is the default exit.
Is strong copyleft actually winning, or is this a small-base effect?
Both, and the receipts cut both ways. Against 3.4 million MIT repos created in 2025, 2,704 is a rounding error, and my count includes only new repos — Bergius-style flips of existing projects are invisible to creation-date filters. But direction and slope are the signal: from a small base, SaaS-closing licenses compound five to six times faster than the copyleft middle. Watch the AGPL and EUPL lines, not the GPL line. When lifelong MIT authors start counting tokens, license choice becomes the leading indicator of how much of the commons still believes the old bargain.