Something fundamental just shifted in the cloud — and most businesses haven’t noticed yet.

TechCrunch reported yesterday that AWS, Cloudflare, Microsoft, and Google Cloud are all redesigning their core infrastructure not for the humans who built the web, but for the AI agents now running on top of it. The catalyst: non-human traffic is already massive and growing faster than anyone expected.

According to Cloudflare’s own data, bots now account for 31% of all HTTP traffic across the web. AI crawlers, agents, and assistants represent roughly a quarter of that bot traffic. Put simply, around one in twelve web requests today is generated by an AI agent rather than a person. By Cloudflare’s projections, machine-generated traffic will exceed human traffic entirely sometime in the first half of 2027 — less than 12 months away.

The Problem With Human-Designed Infrastructure

Cloud infrastructure was built around predictable human behaviour: people search, click, scroll, and stream at a steady pace. AI agents behave nothing like this. A single agent task can spin up dozens of sub-agents simultaneously, hitting hundreds of APIs and databases in seconds, then disappear completely once the job is done.

Traditional cloud architecture doesn’t scale well for this spike-and-idle pattern. It was designed for sustained, predictable load — not for burst-then-zero demand.

AWS addressed this directly this week with the launch of its next generation of OpenSearch Serverless, a fully managed search and vector database built specifically for agentic workloads. The system is designed to scale instantly when agents trigger tasks and return to zero when they go idle. You’re not paying for capacity that sits unused between agent runs.

Microsoft and Google Cloud are making similar infrastructure moves, with both companies announcing architectural changes at their recent enterprise events to support the machine-traffic patterns that agentic deployments create.

Why This Matters More Than It Sounds

This is not a story about cloud vendors upgrading their products. It is a story about the entire internet reorienting around a new type of user — one that doesn’t browse, doesn’t sleep, and generates activity in patterns that human-centric infrastructure was never built to handle.

For businesses deploying AI agents, this infrastructure shift is actually good news. It means the reliability and cost profile of running agents at scale is about to improve significantly. The early days of agentic AI saw real friction: unpredictable latency, cold-start penalties, and infrastructure costs that didn’t match the intermittent nature of agent workloads. The platforms are catching up.

For businesses that haven’t yet deployed agents, this data point matters for a different reason. When Cloudflare reports that 31% of web traffic is now non-human, the implication is that your competitors’ AI agents are already operating across the web — researching, analysing, taking actions — while you’re still evaluating options.

What This Means for Business

Agent costs will fall. Infrastructure redesigned for agentic workloads means more efficient pricing. The spike-and-scale model AWS is building reduces wasted compute, which eventually flows through to lower running costs for businesses using agent platforms.

Reliability will improve. Current agentic deployments sometimes hit limits when multiple agents run simultaneously. Infrastructure built for machine-first traffic handles concurrency better. Multi-agent workflows — where several specialised agents handle different parts of a task in parallel — become more viable.

The window to start is now. The infrastructure is maturing at exactly the moment enterprise AI adoption is accelerating. Businesses that establish agent workflows today will be running on progressively better infrastructure as platforms like AWS, Cloudflare, and Google Cloud complete this rebuild — rather than starting from scratch when the technology is further along.

Data and context management become critical. AWS’s focus on vector databases for agentic workloads points to where the bottleneck is shifting: not compute, but retrieval. Agents are only as useful as the data they can access quickly and accurately. Businesses with clean, well-structured data will get far more value from agents than those with fragmented information systems.

The internet was built for humans. For about 30 years, that assumption held. Somewhere in 2027, it will officially no longer be true. The question for any business leader is simple: are your operations designed for the world that’s arriving, or the one that’s leaving?

---

The practical next step is the free Working With Claude field guide. Thirty-two pages covering the ecosystem, Claude Code, and how to govern a rollout properly. Get your copy.