Azure's New AI Gateway Governs Everything, Promises Nothing (Yet)

Microsoft has released a dedicated AI Gateway tier of Azure API Management in public preview, built as a separate experience organized around models, MCP servers, and tools rather than APIs. It fronts multiple providers—Foundry-hosted models, AWS Bedrock, Google Vertex AI, and direct OpenAI—using card-based policies instead of XML, with OpenTelemetry token metrics and per-backend auth options. Reactions are broadly positive on consolidation and centralized cost governance, but architects raise concerns about gateway-scoped runtime keys with large blast radius, unclear migration from existing AI gateway capabilities, and whether governing agent output lifecycle belongs at the gateway or the orchestration layer. The preview is free with no SLA, unpublished quotas, and pricing TBD.