Build1 publisher3 min readPublished
Voss rests the case against license-level charging on three reversals and one buyout between 2017 and 2025
The post's hawks-and-doves model explains why MIT-style terms keep winning. Each of the four cases behind the claim is an infrastructure project with a downstream party rich enough to fund a fork of its own.
The Engineer · Build desk
What happened
- Voss published a post on seldo.com titled 'Nobody Pays for Open Source: We Can Force Them To', arguing that the free-licensing outcome is stable and that payment has to be forced, since asking does not work.
- The frame is hawks and doves from evolutionary biology, where closed source withholds code and open source gives it away, and the stable mixture is the one no player improves on by switching behaviour.
- Voss puts software's stable strategy at permissive terms, quoted as 'anybody may use this for anything, including commercially, for free', the bargain MIT, BSD and Apache licenses make.
- Four attempts to charge at the license level carry the argument: React's BSD+Patents in 2017, Elasticsearch in 2021, Terraform in 2023 and Redis in March 2024.
- Three of the four ended with an open source license restored, and the exception, HashiCorp, kept its Terraform change and was bought by IBM.
Compiled by The EngineerSomething wrong?How this is made
Why it matters
- decision A vendor weighing a source-available move is choosing how long it can hold one. The Redis case sets the recent benchmark at about fourteen months before the old license came back.
- constraint When a downstream user has the budget to staff a competing copy of the code, license terms stop working as a revenue lever for the maintainer.
- precedent Two of the forks were adopted by the Linux Foundation. The next company that relicenses should plan against a foundation-hosted competitor with governance.
Relicensing fails in a specific sequence. The cases in the post show it. Someone downstream has to want the old terms enough to pay engineers. The code has to land somewhere neutral, so users are not swapping one vendor's risk for another's. Adoption then has to be fast enough that the upstream company's own customers never have to choose. Amazon forked Elasticsearch as OpenSearch after Elastic went source-available in 2021, and the fork ended up at the Linux Foundation with thousands of contributors [14]. OpenTofu went to the Linux Foundation as well [15]. Valkey picked up every cloud provider in about a week [16].
React is the odd one out here, because no fork shipped. The Apache Software Foundation banned the BSD+Patents license from Apache projects in 2017, WordPress announced it was dropping React, and Facebook relicensed under MIT within weeks [13]. The leverage came from large downstream consumers deciding what they would ship.
The reversals are also getting faster. Elastic moved in 2021 and was back on an open source license in 2024, roughly three years. Redis moved in March 2024 and restored the AGPL in May 2025, about fourteen months [3]. Voss writes that the Redis CEO admitted the change had cost the company enormously [16].
"Every project that has tried to be a slightly less generous dove has lost to a project that stayed a full dove," Voss wrote [12]. The four cases support something narrower. Each of them had a downstream party with the money and staff to run a competing codebase: Amazon was selling Elasticsearch as a service before it forked it [14], and Valkey had every cloud provider inside a week [16]. A package with one maintainer and no commercial downstream has nobody to fund the fork. The post does not include a case like that.
On cause, Voss rules out ideology. "Nobody who has tried to charge for open source at the license level has held the line against a competently run fork, and I don't think that's because of ideology," he wrote [17]. He puts it down to market structure, where disruption drags the price toward zero and leaves a large cheap half of the market with nearly all the users and a small expensive half with nearly all the money, Android and iOS [18][19].
The remedy is at the registry layer, and the post tells anyone short on time to skip to that part [5]. The text available here stops mid-sentence in the section that follows, headed "The stable outcome runs on people burning out" [20][21]. So the diagnosis can be tested against four dated cases. The proposal cannot be tested at all. Voss first thought about writing on the economics of open source in 2013, and by his own dating, a 2022 note that then sat for four more years, the post is from 2026, thirteen years later [2][3][4][1][2].
What to watch
- The registries section itself: what it asks a registry operator to do, and on whose authority.
- A relicense that no funded downstream bothers to fork would break the pattern these four cases establish.
- Any disclosure from Elastic, Redis or HashiCorp putting a number on what the license change cost.