Infra · 28 May 2026 · 2 min read

Cloud providers are rebuilding the internet for AI agents

AWS and Cloudflare are redesigning core infrastructure for agentic workloads, forcing builders to rethink how they scale AI-native applications.

Pen-and-ink illustration: a vast, intricate neural network being woven. For the story "Cloud providers are rebuilding the internet for AI agents".
— Pen-and-ink illustration: a vast, intricate neural network being woven. For the story "Cloud providers are rebuilding the internet for AI agents". —

What happened

AWS has launched a new generation of its OpenSearch Serverless database. According to TechCrunch, the system is designed specifically for the bursty, unpredictable traffic patterns of AI agents. It decouples compute from storage, allowing it to scale instantly for agent tasks and then scale down to zero when idle.

The move comes as Cloudflare reports bot traffic already accounts for 31% of HTTP requests. The company predicts machine traffic will exceed human traffic by the first half of 2027.

How the room's reading it

The consensus among practitioners is that this isn't just an AWS move — it's a foundational shift across the cloud industry. Infra teams see similar patterns from Microsoft, which updated Azure to handle AI agent bursts, and from Cloudflare, which recently launched its own tools for agent scalability. Data platform builders note that Databricks and Snowflake are also repositioning themselves as core AI memory and retrieval systems.

The general sentiment reflects a growing realisation that infrastructure built for human clicks doesn't work for autonomous agents. For developers shipping these systems, this is a welcome change that could make productionising agents cheaper and simpler.

Sailfish's take

We think the "scale to zero" feature is the most important part of this story. It’s not just about saving money on idle compute — it changes the unit economics for entire classes of AI applications. Until now, deploying thousands of speculative, short-lived agents was prohibitively expensive because you paid for their infrastructure even when they were dormant. This changes that calculus entirely.

We're now looking at architectures where fleets of agents can be deployed for highly specific, ephemeral tasks without a constant cost overhead. If you've been putting off agent-based projects because of unpredictable costs, this is the week to revisit those plans.

Our take — your read?

Be the first to weigh in.

Sources
— END OF DISPATCH — Infra