Cloud infrastructure has long been designed around humans searching, clicking, scrolling, and streaming in a stable and predictable manner. AI agents behave differently. They unleash massive amounts of activity, querying hundreds of databases, searching documents, launching multiple subagents that call APIs in seconds, and disappearing as quickly as they arrive.
With that premise in mind, Amazon is redesigning the core of its cloud infrastructure. On Thursday, AWS announced the next generation of OpenSearch Serverless. It is a fully managed search and vector database, essentially a large-scale information storage and retrieval system designed specifically for agent workloads. AWS says the new system can scale up instantly when an agent triggers a task, and scale down to zero when idle.
The launch reflects growing awareness across the technology industry. The infrastructure that was originally designed for a human-driven Internet doesn’t work well in a world with more and more agents.
Although AI agents still make up a relatively small portion of internet activity, machine-generated traffic is already significant and will continue to grow. According to Cloudflare, bots accounted for 31% of all HTTP traffic over the past six months. AI crawlers, search engines, and assistants accounted for about a quarter of all bot requests during this period.
“Non-human traffic will exceed human traffic in the first half of 2027,” Lai Yi Ohlsen, senior product manager at Cloudflare, told TechCrunch.
At Google’s I/O developer conference last week, the company said users will be able to delegate tasks to AI systems, including researching purchases, booking travel, browsing the web, and navigating apps. But the toll goes beyond consumer-centric AI agents. Businesses are increasingly deploying agents internally and for their customers, creating new types of machine-generated traffic behind the scenes.
As a result, cloud providers and infrastructure companies have been considering how to adapt systems built for humans to a world of agents that constantly autonomously obtain information, invoke tools, and generate traffic between machines.
That’s where AWS’ new OpenSearch Serverless comes in.
“The timing is simple: agents are moving from experiment to production, creating traffic patterns for which the previous infrastructure was not designed,” Tia White, general manager of Amazon OpenSearch Service, told TechCrunch. “With spikes occurring without warning and idleness occurring without notice, enterprises need search that can accommodate empty or idle compute without paying for it.”
The main technical change in this new generation is that by decoupling compute from storage, compute can be scaled up in seconds to accommodate bursts of agent traffic and scaled down to zero, so customers pay $0 when the agent is idle.
“Previously, even serverless versions combined storage and compute, so at least one instance had to be up and running,” White says. “We couldn’t automatically spin up (compute) as fast as we wanted, so idle compute was always reserved for workloads, regardless of whether they were using them or not.”
Think of it like always paying for a parking space, even when you’re not using it. AWS’s upgraded serverless is like paying for metered parking.
At release, OpenSearch Serverless will natively integrate with AI development platforms like Vercel and Kiro, allowing developers to deploy production-ready vector backends for search and agents without managing infrastructure.
This shift is visible across the cloud industry. Databricks and Snowflake are repositioning themselves as AI memory and retrieval systems for enterprise data. Microsoft has rolled out an update to Azure designed to handle bursts of AI agents and share memory between agents. Cloudflare, like Amazon, introduced infrastructure last month aimed at providing a persistent environment and instant scalability for agents.
As more companies deploy AI agents, there will be more pressure to redesign their infrastructure around machine-generated workloads, potentially making agents cheaper and easier to deploy at scale.
If you buy through links in our articles, we may earn a small commission. This does not affect editorial independence.
