Cloudflare's AI stack stats

Cloudflare says the artificial intelligence stack it built for its own employees now handles production-scale traffic, not internal demos. (blog.cloudflare.com) In a post published April 20, Cloudflare said its internal setup routed 20.18 million monthly requests through AI Gateway, processed 241.37 billion tokens there, and ran 51.83 billion tokens on Workers AI. (blog.cloudflare.com) The company said 3,683 employees actively use its AI coding tools, equal to 60% of the company and 93% of research-and-development staff, with 295 teams using agentic AI tools and coding assistants over the last 30 days. (blog.cloudflare.com) Large language models are systems that predict the next chunk of text, and companies usually wrap them with routing, identity checks, logging, and cost controls before employees can use them at work. Cloudflare said it assembled those layers with its own products, including Access for authentication, AI Gateway for model routing, and Workers AI for inference. (blog.cloudflare.com) The post lands during Cloudflare’s April 2026 “Agents Week,” a product push focused on tools for long-running software agents that can call models, store state, and execute tasks across applications. Cloudflare tied the internal stack to that launch cycle by saying the system runs on the same platform it is selling to customers. (blog.cloudflare.com 1) (blog.cloudflare.com 2) Cloudflare said adoption changed its software workflow over the last 11 months, with a four-week rolling average of merge requests rising from about 5,600 a week to more than 8,700, and one week in late March reaching 10,952. The company said it had not previously seen a quarter-to-quarter increase of that size. (blog.cloudflare.com) The company described the stack as more than a chatbot layer. It includes a portal for Model Context Protocol servers, a continuous-integration code reviewer, sandboxed environments for agent-generated code, and durable sessions for long-running agents. (blog.cloudflare.com 1) (blog.cloudflare.com 2) Cloudflare has been expanding the commercial pieces underneath that stack at the same time. On April 16, it said AI Gateway was becoming a unified inference layer with access to models from more than 14 providers, alongside tighter integration with Workers AI. (blog.cloudflare.com) The company’s pitch is that AI inside a business now looks like network plumbing: authenticated, metered, logged, and built into code review and deployment systems. Cloudflare used its own numbers to argue that the operational work around models is becoming as important as the models themselves. (blog.cloudflare.com)