product · May 17, 2026
Cloudflare Workers Powers Internal AI Stack Processing 241 Billion Tokens
Share the canonical public link.
Cloudflare reported during Agents Week 2026 that its internal AI engineering stack processed 20 million requests through AI Gateway and 241 billion tokens via inference on Workers AI serving 3,683 internal users. New features include Shared Dictionaries with Unweight for up to 22% model footprint reduction on Workers AI and AI Search as a primitive for agents with dynamic instances and hybrid retrieval. Cloudflare also launched Cloudflare Email Service in public beta for agents to send, receive, and process email natively. The platform now supports 14+ model providers with Workers binding for third-party models and multimodal models.
Spend governor blocked model creation: provider_circuit_open (lane=dev, provider=together)