July 1, 2026. Two announcements from Anthropic this week make the same point: the hard part of running AI in a business is no longer the model, it is governing who uses it and what it costs. Claude is now generally available in Microsoft Foundry on Azure, announced June 29, and Anthropic has introduced the Claude apps gateway for Amazon Bedrock, Google Cloud, and Foundry. Together they put Claude, with real controls, inside all three major clouds.
What shipped
- Claude in Microsoft Foundry reached general availability. Claude Opus 4.8, Haiku 4.5, and Sonnet 5 run in the Messages API, hosted on Azure, using the authentication, billing, and governance your teams already have, with prompt caching and extended thinking supported.
- The Claude apps gateway adds a self hosted control plane for Claude Code. It runs across Bedrock, Google Cloud, and Foundry behind corporate single sign on, with per group model access, per user cost tracking, spend caps, and OTLP telemetry.
- The net effect is that Claude is now a first class option in the cloud you already run, with the same access and cost controls in each one.
What it means for operators
The reason AI projects get canceled is rarely the model. It is ungoverned access and runaway cost, the exact failure pattern behind the 40 percent of agentic projects Gartner expects to be scrapped. These releases turn the guardrails into product. Single sign on means only the right people, and the right agents, can reach the model. Per user cost tracking and spend caps mean a runaway process cannot quietly run up a four figure bill overnight. Per group model access means you match an expensive reasoning model to the work that needs it and a cheaper one to the work that does not. You do not need to be an enterprise to adopt the pattern. A small team or an agency can put every AI workflow behind one control plane, cap spend, scope access by role, and keep telemetry, from the first day rather than after the first surprise invoice. That discipline is what separates automation that ships and keeps running from a pilot that gets pulled. If you want AI deployed with governance and cost control built in, we can set up your AI automation, or you can hire an AI engineer to put the right control plane around it.
Frequently Asked Questions
It means you can run Claude models, including Opus 4.8, Haiku 4.5, and Sonnet 5, inside your existing Azure environment using the authentication, billing, and governance you already have. For teams standardized on Microsoft, that removes a major barrier to adopting Claude in production.
It is a self hosted control plane for Claude Code that runs across Amazon Bedrock, Google Cloud, and Microsoft Foundry. It adds corporate single sign on, per group model access, per user cost tracking, spend caps, and telemetry, so an organization can manage who uses Claude Code and what it costs from one place.
No. The pattern of one control plane, scoped access, and spend caps works at any size. A small team or agency can apply the same discipline from day one, and it is far easier to start with guardrails than to add them after a cost or access problem appears.
Per user cost tracking shows exactly who and what is spending, spend caps stop a runaway process before it runs up a large bill, and per group model access lets you route expensive models only to the tasks that need them. Together they turn AI spend from a surprise into a managed line item.