The infrastructure layer for AI agents has been one of the messiest problems in enterprise AI deployment. Every team building agents has had to solve authentication, rate limiting, tool access governance, and observability from scratch. Cloudflare is making its move to own that layer.
Cloudflare today launched Agent Cloud, a new platform that gives developers and enterprises the infrastructure they need to build, deploy, and scale AI agents in production - with security controls, identity management, and network-level observability built in from the start.
What Agent Cloud Provides
The platform addresses the core operational problems that have blocked many enterprises from moving AI agents from successful pilots into production at scale. Agents need to authenticate to external tools and APIs. They need rate limiting to prevent runaway costs or loops. They need scoped permissions that limit what data and systems they can access. And they need observability so teams can understand what agents are doing, debug failures, and audit behavior for compliance purposes.
Cloudflare's network position - sitting between users, applications, and the broader internet for millions of organizations - gives it a structural advantage for this layer. Agent Cloud integrates with Cloudflare's existing Workers platform, meaning agents run on the same globally distributed edge network that already handles traffic for a significant portion of the internet. This matters for latency, reliability, and the ability to enforce security policies at the network layer rather than hoping application-level controls hold.
The launch includes support for the Model Context Protocol standard, which has become the default mechanism by which agents connect to external tools - and which Anthropic's Claude platform helped establish. Agents built on Agent Cloud can plug into MCP-compatible tool servers with Cloudflare handling authentication and access governance.
The Strategic Position
Cloudflare is making an explicit bet that the AI agent era creates the same kind of infrastructure opportunity that the cloud era created for AWS and Azure - but that this time, the network layer matters more than the compute layer for a different class of workloads. Agents are not primarily compute-intensive in the same way that model training is. They are coordination-intensive. They call external APIs, access databases, read and write files, and chain actions together across multiple systems. That traffic flows through networks, and Cloudflare owns a significant share of that network.
For enterprise architects evaluating AI agent infrastructure, Agent Cloud represents a credible alternative to building the governance and security layer in-house - which most teams have found more expensive and time-consuming than the agent logic itself. Cloudflare's global distribution and existing enterprise relationships give it a natural path to adoption among organizations already using its products for security and performance.
