OpenSearch Builds an Independent Future as AI Search Demand Grows

OpenSearch’s most consequential shift may be happening around the software, not inside it. Five years after AWS launched the project in response to Elastic’s licensing changes, the search and analytics platform is trying to establish that it can attract users, contributors, and companies on its own terms. The evidence is mixed: deployments are spreading beyond Amazon’s managed service, but AWS still supplies most recorded contributions and retains a visible role in project leadership.
That tension matters to technology teams assessing OpenSearch as infrastructure for search, analytics, and AI applications. A project’s code can be open source while its direction remains closely associated with one company. OpenSearch’s challenge is to show that its governance and contributor base translate into practical independence for the organizations building on it.
Governance has changed; perception takes longer
OpenSearch began in January 2021, after Elastic moved Elasticsearch and Kibana from Apache 2.0 to a mix of proprietary and source available licenses. Elastic said the change was meant to stop large cloud providers from selling the software as a service without contributing back, and accused AWS of infringing the Elasticsearch trademark. AWS responded by creating OpenSearch, which initially shared much of Elasticsearch’s technology and documentation.
Elastic later added AGPL licensing options for Elasticsearch and Kibana, and its relationship with AWS improved. OpenSearch, meanwhile, continued as a separate project. Its most meaningful governance change came in September 2024, when AWS donated it to the Linux Foundation and the OpenSearch Software Foundation was established.
The transfer was intended to give other organizations confidence that they could participate without AWS retaining ultimate control. Bianca Lewis, executive director of the OpenSearch Software Foundation, called the donation “a watershed moment.” In her view, AWS could no longer close the project, change its license, or claim ownership of the application.
Lewis also sees a business rationale for AWS stepping back from direct ownership. “AWS don’t make money from applications,” she said. “AWS make money from compute.” The company can earn revenue from the infrastructure customers use to run OpenSearch without being the organization that governs the project itself.
Formal independence, though, doesn’t erase the practical signs of AWS’s influence. Figures supplied by the foundation show that AWS’s share of contribution activity across project repositories fell from 91% in OpenSearch’s first year to 72% in 2025–26. In the core repository, Amazon-affiliated engineers made up 95% of maintainers in April 2022; they account for 71% today.
AWS also holds four of the 15 voting seats on the Technical Steering Committee. Yet an AWS employee, Andrew Ross, was reelected committee chair in October. These figures tell a more nuanced story than either full independence or AWS control: participation has broadened, but the original backer remains the project’s largest contributor and a central presence in its technical leadership.
Deployment patterns point beyond Amazon’s service
Users’ infrastructure choices offer another measure of how far OpenSearch has traveled. The Linux Foundation and OpenSearch Software Foundation’s 2026 Open Data Infrastructure Report found that self-managed deployments have grown more quickly than Amazon’s managed OpenSearch Service since 2024.
Among surveyed OpenSearch users, 60% now run self-managed deployments in the cloud, up from 37% in 2024. Use of Amazon’s managed service rose from 44% to 48% over the same period. The data suggests that organizations are increasingly willing to operate the software themselves, giving them more control over hosting and operations.
That growth hasn’t removed doubts about the project’s independence. Nearly a quarter of the 71 surveyed non-adopters—24%—said concerns about vendor neutrality and OpenSearch’s historical ties to AWS were reasons they had not adopted it. For prospective users, the Linux Foundation transfer is a governance milestone; the composition of contributors and the project’s long-term decision-making are the evidence that must sustain it.
Those concerns also have a practical dimension. Companies that choose a platform partly to reduce dependence on a single supplier will want confidence that the project can keep developing if one vendor’s priorities change. Broader participation is therefore more than a governance signal: it can help shape the pace, scope, and continuity of the technology.
Search features put OpenSearch in the AI stack
OpenSearch has also moved beyond its early resemblance to Elasticsearch. Lewis acknowledged that the first versions were close: they had the same plugins, and much of the documentation came from Elasticsearch. By the time of the Linux Foundation donation, she said, the codebase had already diverged substantially.
The project is now positioning its search and analytics capabilities for applications that need to retrieve and analyze information, including systems built around large language models. Vector and hybrid search are particularly relevant to retrieval-augmented generation, or RAG, where an application fetches information from external sources to provide context for a model’s response.
OpenSearch 3.9, released in September, adds work aimed at helping developers search and retrieve data for AI applications. The release also includes changes to performance, monitoring, and alerting. These capabilities place OpenSearch in a part of the AI infrastructure stack where teams need to connect models with organizational data, rather than treating the model as the whole application.
Lewis sees an opening for an open platform in that role. “If you decide that you wanted your AI infrastructure on an open platform, I think [OpenSearch] today is just the default choice,” she said. That’s a confident claim, but the project’s prospects will depend on how well its search features fit real production needs and whether organizations can operate it at scale.
Enterprise adoption depends on time and portability
Interest in OpenSearch doesn’t necessarily mean that companies are ready to replace established platforms. The foundation’s report indicates that many deployments are still experimental or serve applications that could be replaced. Lewis points to long contracts as one reason for the gap between evaluation and migration: a company with a three-year Splunk contract or a five-year Elastic license may continue using the product it has already committed to.
That makes adoption a longer process than a comparison of features or license terms. Teams also have to account for migration work, data movement, and the engineering effort needed to keep applications running while infrastructure changes. A trial can establish technical fit; it can’t, on its own, show that an organization is ready to retire an incumbent system.
Portability is part of OpenSearch’s appeal. Lewis framed the issue as a question companies should ask before they become dependent on a platform: if they later want to leave, will they need to redesign their applications? OpenSearch’s open source model may give teams more options over where and how they run search workloads, but those options are valuable only if the project remains viable and the software meets their operational requirements.
That brings the adoption case back to contributor diversity. If organizations are choosing OpenSearch to keep their options open, they’ll look for signs that no single vendor can set its direction alone. The project has made measurable progress: AWS’s share of maintainers and contribution activity has declined, self-managed cloud use has grown, and the software has developed beyond its Elasticsearch beginnings. The remaining test is whether those trends continue as OpenSearch competes for production workloads and a larger role in enterprise AI systems.
Lewis’s five-year ambition is a platform that is “not dominated or heavily influenced by any single vendor in the world.” OpenSearch isn’t there yet. Its next phase will turn on whether more organizations contribute to its development as well as deploy it, making vendor neutrality visible in the work that shapes the software.