Cloudflare Expands Agent Cloud as AI Infrastructure Race Heats Up

Cloudflare bets big on agentic AI infrastructure

Cloudflare is expanding its Agent Cloud platform with a suite of new tools aimed at developers building and running AI agents. The company says agentic, multi-agent systems need different infrastructure than traditional cloud workloads, and existing options like VMs and containers are too expensive for the short-lived, high-volume execution patterns that AI agents demand.

The announcement includes Dynamic Workers (isolated runtimes for AI-generated code), Artifacts (Git-compatible storage for AI-produced files), and Sandboxes (persistent Linux environments with shell access). All of this runs on top of Cloudflare's Workers serverless platform.

What is actually new

The key addition is Dynamic Workers. These are lightweight, isolated execution environments designed for tasks that start, run, and stop frequently. Think of an AI agent that needs to call an API, transform some data, or run a quick code snippet. Spinning up a container for that is overkill. Dynamic Workers handle it at a fraction of the cost, without needing Kubernetes.

Cloudflare also acquired Replicate, giving developers access to a broader catalog of AI models, both proprietary and open-source, through a single interface. The Think framework adds session persistence, so agents can maintain state across multiple interactions and work on longer-running tasks.

Security angle

For WAF and security teams, the interesting question is how Cloudflare plans to secure thousands of autonomous agents running on its infrastructure. Each Dynamic Worker is an isolated runtime, which helps contain blast radius. But as AI agents gain the ability to write and deploy code autonomously, the attack surface grows in ways traditional WAFs were not designed to handle.

Cloudflare is positioning its existing security stack, including its WAF and DDoS protection, as part of the agent infrastructure. The pitch: build agents that are fast to deploy and protected by default.

WAFplanet take

This is Cloudflare doing what Cloudflare does best: taking an emerging developer trend and wrapping it in infrastructure you do not have to manage yourself. The agent cloud play makes strategic sense. Agentic AI workloads are bursty, ephemeral, and high-volume, which maps perfectly to Cloudflare's edge network and serverless model.

The timing is notable. This announcement came the same week Cloudflare was excluded from Anthropic's Project Glasswing cybersecurity coalition. Cloudflare appears to be responding by doubling down on the infrastructure side: if you cannot be part of the AI security inner circle, become the platform everyone builds on. Whether that strategy works depends on how quickly the WAF and security market shifts toward agent-driven architectures. For now, it is a strong positioning move from a company that needs to show it is not falling behind in the AI arms race.