As AI agents move from the experimental stage to production environments, cloud platforms are beginning to rewrite the underlying systems that were originally designed for human operation. TechCrunch reports that AWS has launched a new generation of OpenSearch Serverless, which focuses on handling the bursty search, query, and API call demands brought about by AI agents.
Compute and storage split
The core of this AWS update is the separation of compute and storage resources. When a proxy task suddenly starts, the system can scale up within seconds; after the task ends, compute resources can drop back to zero, avoiding the retention of idle computing power for extended periods.
According to AWS, even with a serverless architecture, the older version required at least one running instance because compute and storage were previously tied together. The new version, with its on-demand start/stop feature, eliminates the need for customers to pay for compute resources when the agent is idle.
Machine traffic continues to increase
Behind this adjustment is a changing structure of internet traffic. Cloudflare data shows that in the past six months, bot traffic accounted for 31% of all HTTP traffic; among them, AI crawlers, search engines, and smart assistants accounted for about a quarter of all bot requests.
Lai Yi Ohlsen, product lead at Cloudflare, told TechCrunch that non-human traffic could surpass human traffic in the first half of 2027. While AI agents currently account for a limited share, machine-generated traffic has already reached a considerable scale.
Cloud vendors follow suit
Similar changes are not limited to AWS. At its I/O developer conference, Google stated that users will be able to delegate tasks such as shopping research, travel booking, web browsing, and app interaction to AI systems. Enterprise-wide and customer-facing agent deployments are also driving up machine-to-machine traffic.
Databricks, Snowflake, Microsoft, and Cloudflare have all recently shifted their product focus, emphasizing AI retrieval, agent memory, and bursty load handling. AWS also stated that the new version of OpenSearch Serverless natively integrates with Vercel and Kiro, allowing developers to directly deploy the search and vector database backends required by their agents.
