The internet is being rebuilt for machines
Cloud infrastructure has traditionally been built to accommodate human behaviors such as searching, clicking, scrolling, and streaming, all of which tend to occur in steady and predictable patterns. However, AI agents operate in a fundamentally different manner. They can rapidly generate a surge of activity by launching multiple sub-agents that simultaneously query numerous databases, search through extensive documents, and invoke APIs within seconds, only to vanish almost as quickly. Recognizing this shift, Amazon is revamping a critical part of its cloud infrastructure. Recently, AWS introduced the next generation of OpenSearch Serverless, a fully managed search and vector database tailored specifically for agent-centric workloads. This system is designed to dynamically scale up on demand when agents initiate tasks and scale down completely when inactive, making resource usage highly efficient.
This launch highlights an increasing awareness within the technology sector that infrastructures built for a human-driven internet experience struggle to keep pace with the growing presence of autonomous agents in the digital landscape. While AI agents currently represent a smaller share of overall internet usage, the volume of machine-generated traffic is already substantial and is expected to expand rapidly. For example, recent data indicates that bots composed nearly a third of all HTTP traffic over a six-month span, with AI crawlers, search engines, and assistants representing roughly a quarter of those bot requests. Predictions suggest that non-human traffic will surpass human traffic by mid-2027. In line with this trend, major tech companies are enabling users to delegate a variety of tasks—ranging from researching purchases to managing travel arrangements and interacting with applications—to AI systems. Beyond consumer applications, enterprises are integrating AI agents both internally and externally, creating fresh types of machine-to-machine traffic.
Consequently, cloud service providers and infrastructure developers face the challenge of modifying systems originally engineered for human users to efficiently support agents that function autonomously and continuously. AWS’s new OpenSearch Serverless addresses this challenge by decoupling computing resources from storage, which allows computing capacity to be rapidly scaled up to handle sudden spikes in agent activity and completely scaled down to zero to avoid costs during idle periods. This contrasts with prior serverless models that required an always-on compute instance, effectively forcing users to pay for unused capacity. The change can be likened to moving from having to rent a dedicated parking spot at all times to using a metered space paid for only when occupied.
At its introduction, OpenSearch Serverless will seamlessly integrate with AI-focused development platforms such as Vercel and Kiro. This integration enables developers to deploy production-grade search and vector database backends for AI agents without needing to manage the underlying infrastructure. Similar shifts are occurring industry-wide. Companies like Databricks and Snowflake are positioning their services as key components for AI memory and retrieval within enterprise data systems. Microsoft has enhanced Azure to better accommodate AI agent traffic surges and facilitate shared memory among agents. Cloudflare, paralleling Amazon’s approach, recently launched infrastructure solutions designed to provide persistent environments and rapid scalability for AI agents. As the adoption of AI agents accelerates, the pressure to overhaul cloud infrastructure to efficiently handle machine-generated workloads intensifies, which is likely to drive down costs and simplify the deployment of agents at scale.
