Cloudflare Launches Agentic Cloud Infrastructure to Power AI Agents at Scale

Breaking News: Cloudflare Announces Agentic Cloud Platform

San Francisco, CA — Cloudflare today unveiled a new cloud infrastructure built specifically for AI agents, marking a major shift from traditional multi-tenant computing. The announcement, which concluded the company's first Agents Week, aims to solve scalability challenges as agents become a primary workload.

“The old cloud model was designed for one app serving many users, but agents demand compute capacity for tens of millions of simultaneous sessions,” said Dane Knecht, CTO of Cloudflare. “We are creating Cloud 2.0 — the agentic cloud.”

Compute: From Full Operating Systems to Lightweight Isolates

Agents require flexible computing environments. Some need full OS access to install packages and run terminal commands, while others require millisecond startup times and massive scalability.

Cloudflare introduced Sandboxes — persistent, isolated environments with shell, filesystem, and background processes. These are designed for AI agents to run reliably without interfering with each other. The company also launched Artifacts, a Git-compatible versioned storage that lets agents, developers, and automations create tens of millions of repositories.

Expert Reactions

“Agents have swiftly changed how people work in the past year,” said Rita Kozlov, VP of Product at Cloudflare. “Coding agents ship faster, support agents resolve tickets end-to-end, and research agents validate hypotheses across hundreds of sources in minutes.”

Kozlov added: “The potential scale is staggering. If even a fraction of knowledge workers run a few agents in parallel, we need infrastructure that can handle that.”

Background: The Rise of the Agent Era

Over the past year, AI agents have transitioned from experimental projects to production tools. Developers now run multiple agents in parallel around the clock, forcing a rethinking of cloud architecture.

Cloudflare’s Workers platform, launched eight years ago, inadvertently laid the foundation. The containerless, serverless model proved ready-made for agent workloads. Now the company has added purpose-built primitives for security, identity, and tool integration.

What This Means: Redefining Cloud Computing

This announcement signals that traditional public clouds may need to adapt to agent-centric workloads. Cloudflare’s approach offers an alternative to legacy models, potentially reducing costs and latency for agent-heavy applications.

Developers can now deploy agents with built-in security and identity, and scale from prototype to production seamlessly. The web itself may need to adapt as agents drive a growing share of internet traffic.

Key Launches from Agents Week

• Artifacts: Git-compatible versioned storage for agents, allowing forking from any remote and handing off URLs.
• Sandboxes: Real computers with shell, filesystem, and background processes for persistent agent environments.
• Agent Toolbox: Models, tools, and context needed for agents to do real work.

For more details, see the full compute section above.

This is a developing story. More information will be released as it becomes available.