JA EN
shiichan

The Next-Gen OpenSearch Serverless Is Here, Scaling to Zero for Up to 60% Savings!

AWS Blog ★★★★ View Markdown
#aws#serverless#ai

Hey there, it's me, Shiichan! Today I've got a big search-engine update for you. If you're building AI agents, this one's going to make you smile, so let's dig in.

AWS Blog aws.amazon.com

What was announced?

Over on the AWS Blog, AWS announced the next generation of Amazon OpenSearch Serverless. It's a fully managed search and vector engine designed for people building AI agents. When it's idle, capacity shrinks all the way down to zero; at peak it scales up to thousands of requests per second, and then drops back to zero when things quiet down. That flexibility is the star of the show.

The next generation of OpenSearch Serverless scales from zero to thousands of requests per second and back to zero when idle

The story so far

Until now, if you ran an OpenSearch Service cluster provisioned for peak capacity, you had to keep that capacity around even during quiet hours, and the cost added up. Creating and scaling resources also took time, so even standing up a small backend meant a bit of setup work up front.

What changes

With the next generation, resources are created in seconds, and capacity scales up to 20 times faster than the previous generation. On top of that, you get up to 60% cost savings compared to an OpenSearch Service cluster provisioned for peak capacity.

Because it shrinks all the way to zero when unused, you can spin up search and vector backends for your agents without worrying about idle cost. There are native integrations with AI development platforms like Vercel and Kiro too, so you can reach a production-ready setup in minutes without managing infrastructure.

Dive Deep

From the console, the fastest path is to pick Create collection in the Serverless menu, then use Express create. No configuration is required, and the default settings plus matching security policies are applied automatically. If you want to keep using your existing infrastructure, just choose Switch to Classic. At launch, the collection types you can pick are full-text search and vector search.

You can also create things from the CLI or SDKs. You start by creating a collection group, like this. The key is specifying --generation NEXTGEN, and you can set the minimum capacity to 0 so scale-to-zero kicks in.

aws opensearchserverless create-collection-group \
  --name channy-nextgen-group \
  --standby-replicas ENABLED \
  --generation NEXTGEN \
  --description "My NextGen collection group" \
  --capacity-limits '{
    "maxIndexingCapacityInOCU": 96,
    "maxSearchCapacityInOCU": 96,
    "minIndexingCapacityInOCU": 0,
    "minSearchCapacityInOCU": 0
  }' \
  --region "us-east-1"

Then when you create a collection, it inherits the generation from its parent collection group. The supported collection types are SEARCH and VECTORSEARCH.

aws opensearchserverless create-collection \
  --name channy-nextgen-collection \
  --type SEARCH \
  --collection-group-name channy-nextgen-group \
  --standby-replicas ENABLED \
  --description "My collection in NextGen group" \
  --region "us-east-1"

The developer experience is nice too. From the Vercel console you can create a new OpenSearch collection or connect an existing one. And from Claude Code, Cursor, and Kiro you can use OpenSearch Agent Skills, where each skill packs in domain knowledge, best practices, and multi-step execution logic. With the OpenSearch Launchpad in Kiro Powers, you can work through architecture planning with guided, end-to-end steps.

For pricing, compute is billed as OpenSearch Compute Units (OCUs) for indexing, search, and GPU acceleration, and storage is charged separately in GB-month. It's generally available today, in all AWS commercial Regions where Amazon OpenSearch Serverless is already available.

Wrap-up

  • The next generation of Amazon OpenSearch Serverless is GA: a fully managed search and vector engine for AI agents
  • Zero when idle, up to thousands of requests per second at peak, and up to 60% cost savings versus peak-provisioned clusters
  • Resources created in seconds, scaling up to 20 times faster than the previous generation
  • Native integrations with Vercel and Kiro; Express create needs no config and stands up in minutes
  • Billed as pay-as-you-go OCUs plus separate GB-month storage

If you're a developer who wants to add search or vector search to your agents while keeping idle costs low, this looks like a great fit.

Related posts