Azure OpenAI

Azure OpenAI is now sold as "Azure OpenAI in Foundry Models": Microsoft folded the product into Azure AI Foundry, the old URL redirects there, and the access itself is unchanged — enterprise use of OpenAI's current frontier models inside Azure's compliance and security perimeter. Purchase maps to how much certainty you need: Standard, Global and Data Zone deployments meter tokens on consumption, Provisioned deployments bill throughput units (PTUs) for guaranteed capacity, and Batch covers offline jobs. The vendor pitch (page read 2026-09-20) is paperwork as product — a 99.9% uptime commitment, more than 100 compliance certifications with over 50 region-specific, and built-in content-safety and governance tooling.

The judgement: you are buying procurement, residency and predictability, not model novelty. Data Zone and paired-region deployments keep traffic and stored data inside a geography, and PTUs let an enterprise reserve capacity instead of hoping for rate-limit headroom. Costs stay token-metered like the source service, with enterprise-agreement discounts; the model catalog and regional availability rotate so fast that the docs matrix, not memory, is the reference. Against AWS Bedrock or Google Vertex the shape is similar and the tie-breaker is which cloud already owns your identity, network and invoices. On Azure with an EA this is the path of least friction; otherwise the compliance wrapper may not pay for itself.