Open-source business models in 2026: what the dust finally settled on

Three years ago, open-source relicensing became a genre of its own in tech infrastructure. A database vendor, an infrastructure tool company, a search platform — all following the same arc: popular open-source tool grows big, cloud provider monetises it as a managed service, vendor changes the licence to cut the cloud provider out, community forks to preserve the original terms. By early 2026, four of those stories have enough distance to read clearly.

Redis reversed itself. Elastic reversed itself. HashiCorp got acquired by IBM for $6.4 billion before the outcome was fully settled. Valkey, OpenSearch, and OpenTofu proved far healthier than the “they will die in a year” camp predicted.

Here is what actually happened.

The open-source relicensing pattern, briefly

Skip this section if you followed the events closely. If not: permissive open-source infrastructure tool gets popular → cloud provider offers a managed version and makes money on it → tool vendor relicences to a source-available variant (SSPL, BSL, or something bespoke) → community forks under the original terms → cloud provider backs the fork → vendor faces the same economics again, now competing with an open-source project it cannot control. The commercial logic for each change was understandable. The community reaction to each was also understandable. What is interesting now is the outcomes.

Redis: the reversal

Redis changed from a BSD licence to a dual RSAPL and SSPL licence in March 2024. Within weeks, the Linux Foundation announced Valkey — a community fork backed at launch by Amazon Web Services, Google Cloud, Alibaba Cloud, Ericsson, and Oracle. By May 2025, Redis had reversed again, adding AGPLv3 as an option alongside its commercial licences. The stated reason: the landscape had changed.

The reversal was real, but it arrived after the fork had built momentum that a licence file cannot undo. Valkey’s GitHub now averages 80 pull requests per month, nearly double Redis’s 42. Aiven completed what is documented as the largest Redis-to-Valkey migration: 15,000 servers across three months. A 2025 Percona survey found that 83% of large enterprises were either testing or running Valkey. AWS ElastiCache now offers Valkey as a first-class option. The adoption curve was already set.

This does not mean Redis is finished. The commercial product has features that Valkey does not, and enterprises on managed Redis from Redis Ltd. are not in a hurry to migrate. But the community centre of gravity shifted the moment the Linux Foundation and three major cloud providers committed to the fork, and that shift was not reversed by the AGPLv3 announcement fourteen months later.

Elastic: the first one to go back

Elastic changed from Apache 2.0 to a dual SSPL and Elastic Licence in January 2021, citing AWS’s Elasticsearch managed service as the direct cause. AWS forked as OpenSearch in April 2021, under Apache 2.0. Three and a half years later, in September 2024, Elastic added AGPLv3 as an option alongside its proprietary licences. Elastic’s CTO described the original move as having worked: market confusion resolved, the relationship with AWS “stronger than ever”, and the commercial product on stable footing.

OpenSearch, meanwhile, has over 3,300 contributors. AWS handed governance to the Linux Foundation in September 2024 — the same month Elastic reversed. OpenSearch had over 100 million downloads in its first year. The Elastic reversal did not noticeably dent OpenSearch’s trajectory. Developers burned by the original licence change were not going back simply because the licence changed; the trust was spent, and OpenSearch had its own momentum by then.

The Elastic case is the most plausible framing for a commercial vendor: the AGPL creates a legal obligation for cloud providers to share source code or negotiate a commercial relationship, the reversal repaired some community goodwill, and the fork — while real and healthy — did not kill the commercial product. The vendor got a better outcome than it would have had without the original licence change. The community got a fork that is not going away.

HashiCorp: the acquisition before the outcome

HashiCorp changed Terraform from MPL 2.0 to BSL 1.1 in August 2023. The Linux Foundation announced OpenTofu the following month, with backing from Gruntwork, Spacelift, Harness, Env0, and Scalr. IBM acquired HashiCorp for $6.4 billion in February 2025. IBM has not reversed the BSL. Terraform remains BSL 1.1 under IBM’s ownership.

OpenTofu, now past version 1.8, is approaching 10 million downloads from GitHub releases. Fidelity Investments migrated 50,000 state files across more than 2,000 applications and 4 million managed resources. Boeing, Capital One, and AMD run OpenTofu in production. OpenTofu’s contributor base tripled since launch to over 160 active contributors. Overall IaC practitioner adoption sits around 12%, with 27% of teams planning to evaluate or expand it — while Terraform still holds a larger share.

The HashiCorp story differs from Redis and Elastic in one structural way: the commercial vendor is now an IBM product, not an independent company navigating developer goodwill. IBM does not face the same pressure that typically produces reversals. The acquisition froze the licence situation on the vendor side, which means OpenTofu’s growth will compound without the usual “will they reverse it?” uncertainty hanging over the decision.

The forks: a genuine status check in 2026

The standard reaction to a licence-change announcement in 2023 was that the community fork would die within two years, as most community forks of well-resourced vendor projects do. That prior was not unreasonable. It has not been accurate for any of the three forks discussed here.

What made these forks survive where others did not:

• Cloud-provider backing at launch. AWS, Google, and Alibaba committing engineering resources to Valkey from day one gave it the maintenance headcount to keep pace with upstream features. OpenSearch had AWS engineering from the start. OpenTofu had Gruntwork and Spacelift — not hyperscalers, but infrastructure companies with direct financial interest in keeping the tool alive.
• Linux Foundation governance. Each fork is either a Linux Foundation project or closely affiliated with one. That matters beyond legitimacy signalling: it means vendor acquisition events or leadership changes do not automatically destabilise the fork’s governance structure.
• OSI-approved licences. All three forks use Apache 2.0. That keeps them distributable inside enterprise vendor products without legal complications, which matters for the Gruntworks and Aivens that build on top of infrastructure primitives.
• Real enterprise migration at scale. When Fidelity migrates 4 million managed Terraform resources to OpenTofu, the fork has a different staying-power argument than a few hundred GitHub stars. Enterprise adoption creates its own gravity.

The honest caveat: these are the three most prominent forks from the 2023–24 wave. The pattern does not generalise to every project that gets forked after a licence change. Most do not get hyperscaler backing. Most do not land at the Linux Foundation. Most do not have a multi-billion-dollar acquisition to underscore the stakes. Smaller projects forked under similar pressure have mostly gone quiet.

Who actually won

The cloud providers. AWS, Google, and Alibaba are running managed versions of Valkey, OpenSearch, and OpenTofu (or supporting them actively). Their managed-service businesses were structurally unaffected by the licence changes. The changes forced them to contribute engineering hours to forks rather than running the original projects, but the service offerings continued without interruption. This is the clearest single outcome.

The Linux Foundation. It now hosts Valkey, the OpenSearch Foundation, and OpenTofu. The 2023–24 licensing wave handed it three significant infrastructure projects with cloud-provider-backed contributor bases. That is a durable strengthening of its position as the neutral home for infrastructure that commercial vendors and cloud providers can co-develop without fighting over ownership.

The vendors, partially. Elastic’s outcome is the most defensible: the AWS relationship improved, the AGPL gives it a legal basis to require source-sharing from cloud deployments, and the commercial product still has a future. Redis’s reversal is harder to call a win — it came late, and Valkey’s momentum was already established. HashiCorp’s revenue was growing 15% year-over-year at acquisition time, so BSL did not kill the business; but IBM bought it, not the original founders or their shareholders at a valuation that reflected the licensing gamble paying off.

Engineering teams that migrated. If you moved infrastructure to Valkey, OpenSearch, or OpenTofu, the outlook is good. The forks are healthy, migration tooling improved considerably through 2025, and the Apache 2.0 licences create no downstream legal complexity. The risk of adopting a fork that loses momentum or ends up unmaintained has not materialised for any of these three.

Who did not win. Developer trust in commercial open-source vendors as a category. Burned enough times, a meaningful segment of the infrastructure-engineering community updated its priors: if a tool is popular and a cloud provider is managing it, watch the licence page. The Redis reversal did not reverse Valkey’s momentum because trust is not restored by a licence file change.

What this means for infrastructure decisions

A few practical notes for engineering and platform teams evaluating infrastructure in 2026.

Your infrastructure choice is now a governance decision. A project’s licence, its governance structure (foundation versus vendor-controlled), and its cloud-provider support are the first things to audit, not afterthoughts. Apache 2.0 plus Linux Foundation plus cloud-provider contributors means a project can survive vendor distress events — acquisitions, pivots, shutdown. Everything else is a spectrum of risk.

The migration window matters more than you might expect. Teams that moved to Valkey or OpenTofu within six months of the fork announcements are in better shape than teams still evaluating in 2026 — not because the forks are fragile, but because the cloud-provider engineering hours compound over time. The longer you wait, the larger the feature-parity gap you are comparing against an older baseline.

Do not count on reversals. Redis’s AGPLv3 return did not pull Valkey adopters back. Elastic’s reversal did not dent OpenSearch. If your team has already migrated, the commercial vendor’s licence change is mostly irrelevant. If you have not migrated, a reversal reduces urgency — but it does not make the BSL or SSPL-era vendor equivalent in risk to the Apache 2.0 fork. Read the actual licence before deciding.

The open-source infrastructure licensing debate of 2023–24 is not over. More vendors face the same economics, and the cloud-provider playbook is now well-established. But for the three cases examined here, the picture is clear enough to act on: the forks are real, the reversals were real but partial, and the cloud providers kept their managed-service businesses regardless of what the vendor’s licence said.