AWS Launches Next-Gen OpenSearch Serverless Built for AI Agent Traffic Bursts
Summary
- • AWS launched a redesigned OpenSearch Serverless that scales compute to zero when AI agents are idle
- • Bots already account for 31% of HTTP traffic; non-human traffic projected to exceed human by H1 2027
- • New system decouples compute from storage, enabling instant scale-up during agent activity spikes
- • Vercel and Kiro integrations ship at launch for production-ready agentic search and vector backends
Details
AWS launches next-gen OpenSearch Serverless with true scale-to-zero compute for agentic workloads
Unlike the prior Serverless version, which required at least one instance running at all times due to coupled compute and storage, the new architecture fully decouples the two. Compute scales up in seconds during agent activity and back down to zero when idle — customers pay nothing during quiet periods.
Decoupling compute from storage enables instant burst scaling and zero-cost idle states
The prior constraint was structural: storage and compute were coupled, meaning compute could not scale independently. AWS GM Tia White described the old model as always paying for a parking space; the new model is metered, paying only for active use.
OpenSearch Serverless launches with native integrations for Vercel and Kiro
Developers using Vercel or Kiro can deploy production-ready search and vector backends for agents without managing underlying infrastructure — lowering the barrier for teams building agentic applications.
Cloudflare: bots at 31% of all HTTP traffic; AI agents account for roughly 25% of bot requests
Measured over the past six months. Cloudflare senior PM Li Yi Ohlsen projects non-human traffic will exceed human traffic sometime in H1 2027 — underscoring that the infrastructure challenge is not hypothetical.
AI agents create fundamentally different traffic patterns — unpredictable spikes followed by complete idle periods
Agents spin up multiple sub-agents that query hundreds of databases, call APIs, and retrieve documents in seconds, then disappear entirely. AWS GM Tia White noted these systems 'spike without warning, go idle without notice' — a pattern previous infrastructure was not architected to absorb efficiently.
Microsoft, Cloudflare, Databricks, and Snowflake all simultaneously repositioning for agentic workloads
Microsoft updated Azure to handle agent traffic bursts and enable shared memory between agents. Cloudflare launched persistent agent environments with instant scaling last month. Databricks and Snowflake are repositioning as AI memory and retrieval systems for enterprise data. The convergence signals industry-wide recognition that existing cloud infrastructure is not fit for agentic traffic at scale.
Enterprise agent deployments are a primary driver of new machine-generated traffic — not just consumer AI
Beyond consumer-facing agents, enterprises are deploying agents internally and for customers — creating machine-to-machine traffic that operates entirely behind the scenes and at volumes that stress existing infrastructure.
Product Launch = new product or feature release, Tech Info = technical architecture detail, Industry Update = competitive or market-wide development, Context = background framing, Insight = analysis or forward-looking observation
What This Means
AI practitioners building production agentic systems now have a cloud-native search and vector database option that eliminates the cost of idle compute — a meaningful change for workloads with unpredictable traffic patterns. The architectural shift from coupled to decoupled compute-storage is the pattern to watch: it will propagate across other managed services as agentic workloads become the default design target. Teams evaluating infrastructure stacks for agents should treat true scale-to-zero and burst elasticity as baseline requirements, not differentiators, given how quickly competitors are converging on the same capability.