Cloudflare Merges AI Gateway and Workers AI Into Single Platform

Cloudflare is consolidating two previously distinct AI products—AI Gateway and Workers AI—into a single unified platform that promises to simplify how developers access and manage AI models across multiple providers.

The company announced the convergence will allow developers to connect to any model provider, including Cloudflare's own Workers AI inference service, while managing observability, billing, security, and logging from one control plane. According to details first reported on the Cloudflare blog, the architectural changes represent a fundamental shift toward what the company calls "model-first routing."

Automatic observability without setup

One immediate benefit: developers no longer need to explicitly create an AI Gateway before gaining visibility into inference traffic. By passing "default" as the gateway ID in API calls, the system automatically creates a gateway on the first authenticated request and begins logging full request and response payloads, tracking token counts per model, and attributing costs.

The unified binding works identically whether calling Workers AI models or external providers. Developers add a single gateway parameter to their existing Workers AI calls and immediately inherit request logging, latency breakdowns, token usage tracking, and error rates—capabilities that previously required separate dashboard configuration.

Unified billing across all providers

Cloudflare is now enabling AI Gateway credits to work with Workers AI, a capability that was previously limited to external providers like OpenAI and Anthropic. Developers can load a single wallet and spend credits across any supported provider without managing separate billing relationships.

As an incentive to adopt the unified billing path, Cloudflare is offering elevated rate limits on Workers AI models for users who route through AI Gateway. The company notes that specific rate limit details are available in developer documentation.

Model-first routing coming soon

The more significant architectural shift involves moving from provider-first to model-first routing. Instead of specifying which provider hosts a model, developers will soon be able to request a model by name—such as Kimi K2.7 Code—and let the control plane handle provider selection, failover, and load balancing.

If Workers AI has capacity, requests route to Cloudflare's managed infrastructure. If Workers AI reaches capacity, the gateway transparently shifts traffic to another vetted provider hosting the same model weights. The system is designed to maintain output quality and respect requirements like Zero Data Retention while improving resilience without application-level retry logic.

Cloudflare is also building intelligent routing that goes beyond simple failover. A classifier running on Workers AI will analyze prompts to predict task type, complexity, and context requirements, then map requests to optimal models from a curated pool. This zero-configuration path aims to improve economics and performance without requiring developers to maintain custom routing logic. The company is currently piloting this capability internally.

Why it matters

Consolidating AI infrastructure under a single control plane addresses a persistent pain point for enterprises building AI applications: managing multiple provider relationships, handling failover logic, and reconciling separate billing systems. By abstracting provider complexity and enabling automatic model routing based on availability and task requirements, Cloudflare is positioning its platform as infrastructure that adapts to application needs rather than forcing applications to adapt to infrastructure constraints. For organizations concerned about vendor lock-in or provider outages, model-first routing with automatic failover could significantly reduce operational overhead.

Developers already using Workers AI can begin routing calls through the default gateway immediately to gain observability features. Those using AI Gateway can add Workers AI models and access unified billing across all providers. Details were first reported on the Cloudflare blog.