Azure AI Gateway
Intro
Azure AI Gateway is the fully managed gateway Microsoft declared generally available at Ignite on 17 November 2026. The Tech Community post Azure AI Gateway is now generally available describes a single, controlled path from AI apps and agents to models and tools across providers, so platform teams can manage access, policies, costs, and usage in one place without operating gateway infrastructure themselves.
Microsoft says it was previously in public preview as the AI Gateway tier of Azure API Management and is now a standalone product whose roadmap is focused on governing access to AI models and tools. It can be used on its own, or together with Microsoft Foundry, where you build and run agents, and Microsoft Agent 365, where you manage them. Apps and agents call the gateway instead of models or MCP tools directly; the gateway authenticates with Microsoft Entra ID or a gateway access key, then enforces policy, routes the request, and records telemetry.
This entry records the name Azure AI Gateway as it became a standalone, generally available product. It is not Vercel AI Gateway, and it should not be merged with similarly named products such as TrueFoundry Agent Gateway or Walmart Agent Gateway.
Advertisement
Explain
The stated problem is that enterprise AI is no longer one app calling one model. Organizations use models from several providers, let agents act through enterprise tools, and spread AI across teams. If credentials, routing, safety, and limits stay inside each app, a platform team cannot see who is using what, what it costs, or which policies apply. Azure AI Gateway moves those controls out of individual apps and into one layer.
Public materials describe calling models from Microsoft Foundry, Azure OpenAI, OpenAI, Anthropic, Amazon Bedrock, Google Vertex AI, or any OpenAI-compatible endpoint by an alias you define, so you can switch models without changing app code; routing and failover by priority, weight, or conditions; gathering MCP servers, OpenAPI specs, connectors, and API Management APIs as tools; and using an OpenAI-compatible API or the Anthropic Messages API. A Model Router in preview can choose by cost or quality. Region coverage, the free allowance, and paid billing that begins on 1 January 2027 should be taken from the announcement as published.
Azure AI Gateway is not a model. It is a governance layer in front of models and MCP tools. Spinning it out from a preview tier of API Management also does not make it a synonym for every Azure service with “gateway” in the name.
Usage
Start with the GA announcement and the product site ai.gateway.azure.com, then follow Microsoft Learn’s Azure AI Gateway docs to create a gateway and configure aliases, policies, and billing. The intended use is for apps and agents to call only the gateway, while a platform team centrally manages approved models and MCP tools, identity, quotas, and observability.
Evaluate whether it covers the providers and MCP versions you need, whether failover is testable, whether policy can cap cost per app or team, and what you pay after the free allowance. The 17 regions, 10,000 free requests per gateway each month, and the paid-billing start date follow official documentation at the time you enable it.
If you only put a key on a single Azure OpenAI resource, or you are looking for Vercel’s AI Gateway, that is not this standalone Azure product name.