On 25 August 2006, Amazon announced Elastic Compute Cloud: one instance type, one region, one operating system, which was Linux, running on the Xen hypervisor. About fourteen engineers built it in Cape Town, under a manager named Chris Pinkham, off an idea he had proposed at an executive offsite three years earlier. Nobody in that room asked Amazon to pay Linux's community for the privilege. Nobody thought to.
Eighteen years later, the arithmetic finally moved the other way. Between 2018 and 2024, four of the companies whose software had made the modern cloud possible, MongoDB, Elastic, HashiCorp and Redis, rewrote their own licences specifically to make that free ride harder to take. Each time, a fork answered within weeks. The commons that built the stack your AI systems now sit on top of has started taking its licence back.
[BOOK IMAGE: row 90 - the open source foundation]
That is the claim this article tests against the record: not a prediction, a pattern already run four times in six years, in the exact layer, cloud infrastructure, databases and orchestration tooling, that every large AI deployment depends on.
The foundation was never a secret
Go back further, because the foundation was never hidden, only invisible in the way infrastructure usually is. Amazon S3 had already launched on 14 March 2006. By the time EC2 shipped that August, the pattern was set: build the paid layer on the free layer underneath. Microsoft followed more slowly. Azure launched in 2010 supporting only Windows workloads, but by 2019 more than half of its virtual machines were running Linux instead. Google Cloud never needed the transition; it was Linux from the day it opened. The reason was never ideology. Open source removed the per-instance licensing cost at the infrastructure layer, which freed every provider to charge for something else: the service wrapped around it.
That something else arrived with Amazon RDS in 2009: managed PostgreSQL and managed MySQL, sold as a service. The database itself stayed open source. The part a customer actually paid for, provisioning, patching, backup, the two a.m. phone call, was proprietary. The pattern repeated across the estate over the following decade: Redis became Amazon ElastiCache, Elasticsearch became Amazon Elasticsearch Service and later OpenSearch, Kafka became Amazon MSK, Cassandra became Amazon Keyspaces, MongoDB became Amazon DocumentDB. In every case a community wrote and maintained the software. A hyperscaler sold the operational layer sitting on top of it, and kept the revenue that layer generated.
[BOOK IMAGE: row 94 - the invisibility trap diagram]
Call this the invisibility trap. A team running PostgreSQL on Amazon RDS has a commercial relationship with Amazon. It does not have one with the PostgreSQL community, and there is no obvious reason it should, until the community that keeps writing the software the team depends on gets less visible, attracts fewer contributors, receives less funding and draws less attention, precisely because it succeeded. None of this was illegal. None of it required the source code to close. Open source was not an obstacle to the extraction; it was the mechanism that made the extraction possible, because a hyperscaler could offer the exact software a community had built, at scale, without asking permission or paying a licence fee.
By 2022 the shape of the gap was visible in the accounts. The Linux Foundation, which hosts a large share of the infrastructure the cloud runs on, reported revenue of 311 million dollars that year. Amazon Web Services alone generated more than 90 billion dollars in the same twelve months. Neither figure maps cleanly onto the other's balance sheet, but the order of magnitude is the point: the community that built the substrate operates several orders of magnitude below the company monetising it.
[BOOK IMAGE: row 93 - the value gap chart]
The rebellion that answered it
The rebellion did not begin as a movement. It began as one licence change, made by one company, aimed at one competitor.
[BOOK IMAGE: row 115 - the licence rebellion begins]
On 16 October 2018, MongoDB's chief executive Dev Ittycheria announced that the company was moving its database from the GNU Affero General Public License to a new licence it had written itself, the Server Side Public License. The SSPL's operative clause was narrow and deliberate: anyone offering MongoDB as a managed service had to open-source the entire stack running around it, not just MongoDB itself. No cloud provider was going to do that. The Open Source Initiative declined to recognise the SSPL as open source in March 2019. MongoDB adopted it anyway. Redis Labs had already tried a milder version of the same manoeuvre, a "Commons Clause" added to its own licence in August 2018.
Elastic followed the same logic in January 2021, dual-licensing Elasticsearch and Kibana under the SSPL and a new Elastic License v2, aimed squarely at Amazon's competing managed offering. Amazon's answer arrived within three months: in April 2021 it announced OpenSearch, a corporate fork of the last Apache-licensed version of Elasticsearch, backed by dedicated engineering rather than a community side project. Within two years OpenSearch had grown its own community and its own commercial trajectory, independent of Elastic entirely. Elastic itself reversed course on 29 August 2024, returning Elasticsearch to the fully open AGPLv3. The fork it had provoked did not reverse with it.
[BOOK IMAGE: row 92 - Elasticsearch, the licence and the fork it provoked]
The clearest case is also the most recent. HashiCorp built Terraform into the standard tool for infrastructure as code over more than a decade, and on 10 August 2023 it relicensed its entire product line, Terraform included, from the Mozilla Public License 2.0 to the Business Source License 1.1, a licence restricting competing commercial use for four years from each release. HashiCorp's own decade of integrations, providers and trained practitioners was placed in play within a single announcement.
[BOOK IMAGE: row 118 - HashiCorp's developer network, and the licence that put it in play]
Within days, a coalition of infrastructure companies published the OpenTofu Manifesto. By September 2023 the Linux Foundation had adopted the project. OpenTofu 1.6.0 shipped in January 2024, four months after the fork was announced, as a drop-in Terraform replacement compatible with more than 3,000 existing providers. The founding coalition, Gruntwork, Spacelift, env0, Scalr, Harness and Oracle Cloud Infrastructure, deliberately governed the project so that no single vendor could hold more than half the seats at the table. That governance structure, not the code itself, is the part built to survive the next relicensing attempt.
[BOOK IMAGE: row 119 - OpenTofu, governed so no vendor holds the seats]
HashiCorp itself did not stay independent long enough to answer the question its own move had raised. IBM announced its acquisition of HashiCorp on 24 April 2024, a 6.4 billion dollar deal that closed in February 2025. As of writing, Terraform remains under the BSL with IBM as licensor, and OpenTofu continues to gain adoption regardless of who owns the original.
Redis followed the identical arc on a shorter timeline. In March 2024 Redis moved from a permissive BSD licence to a dual SSPL and RSALv2 model. Within weeks the Linux Foundation established Valkey, a fork backed by Amazon Web Services, Google, Oracle and Alibaba simultaneously, the same hyperscalers the licence change had targeted, now cooperating to route around it. A CHAOSS study found that external contributors making five or more commits to Redis dropped to zero within six months of the relicensing. Redis reversed its own decision on 1 May 2025, shipping Redis 8 back under the fully open AGPLv3. Valkey, like OpenSearch, kept the contributors and the momentum it had already gathered.
Four companies. Four licence changes aimed at the same target, a hyperscale layer consuming the software without paying into it. Four forks that answered within weeks, not years. And where the original company later reversed course, Elastic and Redis did, the fork it had provoked did not reverse with it. That is the lesson Chapter 20 draws from the sequence: in open source, a fork is permanent. A company can return to an open licence. It cannot get its old commons back. The cycle that took a decade to run its course for Unix, and five years for the cloud, ran its full course for HashiCorp and Redis in under two.
The board consequence
None of this is a story about database vendors settling scores with a hyperscaler. It is a story about where the power to change terms actually sits, and it sits closer to your own AI stack than the vendor names suggest. A generative AI platform is not one licence; it is a stack of them, an operating system, a container orchestrator, a vector database, a caching layer, an infrastructure-as-code tool wiring the deployment together, and any one of those licences can move exactly the way MongoDB's, Elastic's, HashiCorp's and Redis's did, on a timeline set by someone else's board, not yours.
I spent years on the enterprise side of exactly this kind of stack, in architecture reviews where cloud strategy got described entirely in vendor names while the open source layer underneath went unmentioned by everyone in the room, myself included, until it became a governance question rather than a technical footnote. The practical task this history sets is unglamorous and specific. A board relying on managed cloud services should be able to name what licence each critical open-source dependency actually runs under today, know who holds the legal power to change it, and know whether a fork is a real, tested option or only a theoretical one nobody has exercised. Permissive yesterday is not a guarantee of permissive tomorrow. The terms move; the only defence is knowing, before they do, which dependency could move next.
The open-source dimension is the reason a company like MongoDB could not simply forbid a fork outright, and it is worth naming directly. The Linux Foundation's role in this cycle was never to write code. It was to host governance that no single company could capture. When the OpenTofu coalition needed a neutral home within days of HashiCorp's announcement, the Foundation gave the project a board seated so that Gruntwork, Oracle Cloud Infrastructure and five other backers shared power with none holding a majority, before anyone had written a single new line of code. Valkey's four-hyperscaler backing repeated the pattern eight months later. This is the mechanism the SSPL and the BSL were built to route around and could not: an open licence lets anyone copy code, but a neutral governance host is what lets a fork actually hold a community together long enough to matter.
The sovereignty question sits one layer below the commercial one. Amazon Web Services built its cloud on Linux and Xen from 2006, and by 2019 more than half of Microsoft's own cloud ran Linux as well. The operational layer that decides whether an organisation's systems keep running is concentrated in a small number of hyperscale providers, whoever wrote the software underneath. When a licence can move the way MongoDB's, Elastic's, HashiCorp's and Redis's moved between 2018 and 2024, an organisation depending on that layer is depending on terms it does not set and cannot unilaterally fix. The licence rebellion proved the fork remains a working backstop; OpenTofu and Valkey both run in production today. Whether a given organisation could actually exercise that backstop under pressure, rather than merely cite it as available, is the assurance question a board still has to answer for itself.
Pull up the licence file for the open-source dependency your AI stack leans on hardest, and read the clause that says what happens if the vendor changes the terms. When did you last do that, and what did it actually say?
This is one of seven weekly series in The Hamberger Report. Subscribe on LinkedIn and the next one arrives in your feed.
The views expressed in this article are entirely my own, informed by morethan 30 years of professional experience in architecture, security, andtechnology leadership in New Zealand. I write as director of Te PonoLimited; the views are personal and do not represent the position of anyclient, any government agency, or the New Zealand government. My commentaryon legislation and policy is analytical, drawing on publicly availablesources and my professional expertise in architecture, security, and AIgovernance, and it is politically neutral.
Andreas Hamberger is a New Zealand-based enterprise architect and technology strategist. Over 30 years, he has moved from compiling kernels on a 486 to leading cloud, cyber, and AI transformation programmes across government, banking, transport, and aviation. He founded Yoper Linux, served as a technology specialist for Novell during the Linux Wars, and is the author of "Generative AI: Skynet or Heaven" and "Space Mafia." He can be reached at andreas@thehambergerreport.com. His latest book, Free as in Theft: The Hidden History of Open Source Software, traces the cycle of openness, adoption, enclosure and resistance that runs from IBM's 1956 consent decree to the AI licensing disputes of today.
This article was produced with AI assistance under my direction. Research,drafting and images pass through a pipeline I built and govern: automatedgates for source verification, forbidden language and political neutrality,and my own review before anything is published. The tools include Claude,Gemini and Openart. The frameworks, arguments and editorial judgements aremine and are the same discipline I apply to the AI systems I audit forclients. AI accelerated the work; the thinking, and the responsibility forit, are mine.
References
[1] Hamberger, Andreas. "Free as in Theft: The Hidden History of Open Source Software." Te Pono Limited, 2026. ISBN 978-0-473-78455-3. (Primary source for all Chapter 15 and Chapter 20 facts: the EC2 launch and Cape Town team, the "as a service" pattern, the invisibility trap, the Linux Foundation and Amazon Web Services revenue figures, the MongoDB, Elastic, HashiCorp and Redis licence changes, and the OpenSearch, OpenTofu and Valkey forks.)
[2] Wikipedia contributors. "OpenTofu." Wikipedia, accessed 3 September 2026. https://en.wikipedia.org/wiki/OpenTofu (Independent currency check: confirms Terraform remains under the Business Source License with IBM as licensor, and that OpenTofu continues to gain adoption, consistent with the manuscript's "as of early 2026" status.)