Amazon has overhauled its serverless OpenSearch database, separating storage from compute to better handle the unpredictable demands of agentic AI. The company claims the update prevents users from paying for idle compute capacity during the intermittent bursts of activity typical of AI agents.
According to an article by The Register, the new architecture allows search collections to scale down to zero when inactive. AWS says it has solved the technical hurdles associated with "cold starts," enabling the system to resume operations within seconds. The company claims the updated service auto-scales 20 times faster than its previous iteration.
"Collections can shrink all the way to zero when nothing's happening," said Tia White, Director of OpenSearch at AWS. "We have mitigated the cold start problem, so they spin back up in seconds when traffic is needed as agents restart. It auto-scales 20 times faster than before."
AWS is positioning the managed search and vector engine as a primary tool for AI development. The service now integrates directly into the Vercel console, allowing developers to deploy search backends without leaving their workflows. It also powers the "OpenSearch Launchpad" within Kiro, the company's new agentic coding IDE, which provides architectural guidance for search-heavy applications.
Competitive shifts in the search market
The move comes as AWS attempts to capture more of the AI agent market, which White suggests requires a different infrastructure approach than traditional enterprise search. "Historically, search has not had to decouple storage and compute, because the traffic was pretty predictable," White told The Register. "Now with agentic workloads, even the most sophisticated technical teams need to use a serverless offering."
While the update is a technical win for AWS, it intensifies competition with Elastic. Elastic launched its own serverless offering in 2024 and recently reported performance gains using AWS Graviton instances. The rivalry dates back to 2021, when Elastic changed its software licensing, prompting AWS to fork the code and launch the Linux Foundation-backed OpenSearch project.
AWS acknowledged that while the new serverless logic shares some features with the open-source project, it relies on a proprietary storage layer that remains closed-source. White stated that while she could not rule out opening the technology in the future, there are currently no plans to do so. Despite the growth of OpenSearch, DB-Engines rankings still place ElasticSearch in 11th place globally, while OpenSearch currently sits in 31st.
