AWS Rebuilds Cloud Search for an Agent-Driven Internet

Cloud infrastructure was built for human behavior—steady clicks, scrolls, and streams. But AI agents don’t behave like people. They can spawn dozens of sub-agents in seconds, hit hundreds of databases and APIs, then vanish. That mismatch is forcing cloud providers to rethink core services.

This week, AWS launched a new generation of OpenSearch Serverless, a fully managed search and vector database designed specifically for agent workloads. The system scales compute instantly when agents trigger tasks and drops to zero when idle. The key change: decoupling compute from storage. Previously, even idle workloads required at least one running instance—like paying for a parking spot you never use. Now, customers pay only for active compute.

“Agents are moving from experiments into production, creating traffic patterns that old infrastructure wasn’t built for,” said Tia White, GM of Amazon OpenSearch Service. “Spikes happen without warning. Idle periods are sudden. Enterprise search needs to keep up without charging for empty compute.”

The timing reflects a broader shift. Cloudflare reports bots accounted for 31% of HTTP traffic over the past six months, with AI crawlers and assistants making up a quarter of that. Cloudflare’s Lai Yi Ohlsen predicts non-human traffic will exceed human traffic by early 2027.

At Google I/O, the company outlined plans for consumer AI agents that handle purchases, travel, and app interactions. Enterprises are deploying agents internally and for customers, generating new machine-to-machine traffic behind the scenes.

AWS’s OpenSearch Serverless integrates natively with Vercel and Kiro, letting developers build production-ready search backends without managing infrastructure. Meanwhile, Databricks, Snowflake, and Microsoft are rolling out similar agent-aware updates. The more agents companies deploy, the more pressure mounts to rebuild the cloud for machines—which, in turn, makes agents cheaper and easier to scale.