Amazon Revamps Cloud Infrastructure for the Rise of AI Agents

Cloud infrastructure has long been designed around humans who search, click, scroll, and stream in a steady and predictable fashion. AI agents behave differently. They can unleash a swell of activity, spinning up multiple sub-agents that query hundreds of databases, search documents, and call APIs in seconds and then disappear as quickly as they arrived. 

In response to the evolving needs of AI, Amazon is overhauling a significant aspect of its cloud services. On Thursday, AWS unveiled its next generation of OpenSearch Serverless, a fully managed search and vector database tailored for agent-led activities. AWS claims this new system can swiftly scale up as agents initiate tasks and reduce to zero when not in use.

This development symbolizes a growing recognition within the tech sector that infrastructure initially intended for human interaction is insufficient in a landscape increasingly populated by AI agents.

Although AI agents currently represent a minor segment of online activity, their influence is rising rapidly. Cloudflare indicates that bots constituted 31% of total HTTP traffic over the past six months, with AI crawlers, search engines, and assistants accounting for around a quarter of all bot interactions during that time.

“Non-human traffic will surpass human traffic sometime in the first half of 2027,” stated Lai Yi Ohlsen, a senior product manager at Cloudflare, during an interview with ToolsMixAi.

At Google’s recent I/O developer conference, the company announced plans for users to begin delegating tasks to AI, which could involve researching purchases, booking travel, navigating the web, and interacting with applications. This trend isn’t limited to consumer-focused AI; businesses are increasingly integrating agents for internal processes and customer interactions, thus generating new types of machine-generated traffic.

Consequently, cloud service providers and infrastructure firms are grappling with how to adapt their systems for a new reality where agents continuously and autonomously retrieve data, use tools, and create machine-to-machine communication.

This is where AWS’s OpenSearch Serverless comes into play.

“The timing is clear. Agents are advancing from testing into practical use, generating traffic patterns that conventional infrastructure wasn’t designed to handle,” stated Tia White, general manager for Amazon OpenSearch Service, when speaking to ToolsMixAi. “Their activity surges unexpectedly, then drops off suddenly, necessitating a search solution that keeps pace without incurring costs during idle periods.”

The significant upgrade allows for a separation between compute and storage, enabling compute resources to scale up in seconds to meet sudden traffic increases and scale down to zero when not needed, allowing customers to incur no costs during dormant periods.

“Before, even in our previous Serverless iteration, at least one instance had to be active since storage and compute were linked,” White noted. “You couldn’t dynamically increase [compute] based on demand, meaning you always had idle compute reserved, regardless of actual use.”

This can be likened to consistently paying for a parking space even when not needed. With AWS’s improved Serverless, it’s akin to only paying for a spot when used.

Upon launch, OpenSearch Serverless will connect directly with AI development platforms like Vercel and Kiro, facilitating the deployment of production-ready search and vector backends for agents without the necessity of managing underlying infrastructure.

A similar shift is occurring across the cloud landscape. Databricks and Snowflake are positioning themselves as AI memory and retrieval systems for enterprise data. Microsoft has introduced enhancements to Azure to manage AI agent surges and provide memory sharing among agents. Similarly, Cloudflare recently revealed infrastructure intended to furnish agents with consistent environments and rapid scalability.

As companies increasingly adopt AI agents, there will be mounting pressure to redesign infrastructures for machine-generated workloads. This evolution could subsequently facilitate broader and more efficient deployment of agents at scale.