AWS redesigns cloud infrastructure to support AI agent traffic

On Thursday, AWS launched its next generation of OpenSearch Serverless, a managed search and vector database service. A vector database is a system for storing and retrieving information at scale, often used for artificial intelligence. This launch represents a redesign of a core piece of AWS’s cloud infrastructure to support AI agents. Traditional cloud infrastructure was built for a human-driven internet, where users search, click, and scroll in predictable patterns. In contrast, AI agents generate highly unpredictable traffic, spinning up multiple sub-agents to query databases and call APIs in seconds before disappearing.

This infrastructure shift is driven by a rise in machine-generated traffic. According to the web infrastructure and security company Cloudflare, bots accounted for 31% of overall HTTP traffic over the last six months. This non-human traffic is projected to grow. Lai Yi Ohlsen, senior product manager at Cloudflare, stated: “Non-human traffic will exceed human traffic sometime in the first half of 2027.” This shift is also visible in consumer tech; at the Google I/O developer conference last week, Google said users will be able to start delegating tasks like researching purchases and booking travel to AI systems.

To handle these sudden bursts of activity, the new AWS service decouples compute—the processing power in cloud infrastructure—from storage. Previously, storage and compute were coupled, meaning customers had to keep at least one instance running and pay for idle compute. Tia White, general manager for Amazon OpenSearch Service, compared the previous model to paying for a parking space even when it is empty. The new architecture allows the system to scale up instantly during agent bursts and scale down to $0 when agents are idle. At launch, OpenSearch Serverless integrates with platforms Vercel and Kiro, allowing developers to deploy search and vector backends without managing the underlying infrastructure.

AWS is not alone in adapting to agentic workloads. Other cloud and data infrastructure providers are updating their systems to support AI agent memory and burst workloads. Databricks and Snowflake are repositioning themselves as AI memory and retrieval systems for enterprise data. Microsoft has updated Azure to handle AI agent bursts and share memory between agents, while Cloudflare recently introduced infrastructure to provide agents with persistent environments and instant scalability.

Why it matters

AWS is redesigning its cloud infrastructure to accommodate AI agents, which create unpredictable traffic patterns that traditional systems, built for human users, cannot handle efficiently. This shift reflects a broader industry trend where cloud providers are adapting to a world increasingly populated by autonomous machine-generated traffic.