A new way to run Claude on Google Cloud has shown up in Anthropic’s official documentation before any announcement. The pages describe “Claude Models on Google Cloud (Anthropic managed),” a version of the Claude API that Anthropic itself operates, reached through your Google Cloud project and billed through Google Cloud Marketplace. Neither Anthropic nor Google has announced the service yet, so treat the details below as what the docs show today, not a confirmed launch.
What the New Claude on Google Cloud Option Is
Until now, most Google Cloud customers reached Claude through Google’s own model platform, which the docs now call Agent Platform (formerly Vertex AI). The new pages split that into two offerings. In the Google-managed version, Google runs the Claude models as part of Agent Platform.
In the Anthropic-managed version, Anthropic runs the Claude API, while Google Cloud handles access and billing. Requests go through a Google gateway at claude.googleapis.com, which checks your IAM permissions and then forwards the call to Anthropic. The docs say Anthropic validates each request from there.
The setup flow even creates a brand-new Anthropic organization tied to your Google Cloud project. It stays separate from any organization your company already has, including a Claude Enterprise one, and an existing organization cannot be converted.
How Anthropic-Managed Claude Differs on Google Cloud
The documentation spells out several practical differences from calling the regular Claude API directly. Here is a quick comparison based only on what the pages state.
| Area | Anthropic-managed on Google Cloud | Standard Claude API |
|---|---|---|
| Sign-in | Google Cloud credentials and IAM roles only | Anthropic API keys |
| Billing | Google Cloud Marketplace, on your Google bill | Invoiced by Anthropic |
| Usage unit | Claude Consumption Units, each worth $0.01 | Billed directly in dollars |
| Model IDs | Same IDs, such as claude-opus-5-5 | Same IDs |
| Logging | Google Cloud Audit Logs for API activity | Claude Console usage tools |
| Rate limits | Managed by Anthropic | Managed by Anthropic |
The docs say usage counts toward a Google Cloud spend commitment the way other Marketplace purchases do. Anthropic does not send a separate invoice. Customer-managed encryption keys work with Google Cloud KMS keys, and zero data retention is available on request.
Features That Come Along, and Some That Don’t
According to the docs, Claude Managed Agents works through the new offering, including sessions and scheduled deployments. That fits Anthropic’s recent push around self-hosted managed agents for enterprise teams. A few pieces are listed as unsupported for now:
- API keys, since sign-in uses Google Cloud credentials only
- MCP tunnels
- Memory stores on self-hosted environments
- The pre-built self-hosted sandbox workers in Anthropic’s CLI and SDKs
Anthropic is listed as the data processor, and the docs note that some supporting services, such as trust and safety systems, process data outside Google Cloud. Inference location defaults to global, with a per-request option for US-only processing.
Claude Code Gets Its Own Setup Page
A matching page in the Claude Code docs explains how to point the coding tool at the new service. You turn it on with a new environment setting and need Claude Code version 2.1.307 or later. Each person or service account also needs the claude.user IAM role on the project, plus a workspace ID from the Claude Console.
Several Claude Code features are listed as unavailable on this route, including fast mode, Remote Control, cloud sessions and claude.ai MCP connectors. Server-managed settings are not supported either, so teams have to enforce settings on each machine instead. If you already tune Claude Code locally, guides like managing permission prompts still apply.
What This Means for You
For most people using the Claude app, nothing changes. This is aimed at companies that run on Google Cloud and want Anthropic’s own Claude API without a separate vendor contract. Here is what to keep in mind:
- If your company buys through Google Cloud, this could let Claude spending land on the same bill and count toward existing commitments.
- You would get the Anthropic-run API, including newer features such as managed agents, while keeping Google IAM and audit logs.
- Moving over is a cutover, not a switch: API keys, workspaces, files and agent resources stay in your old organization.
- Nothing is official yet, so wait for an announcement before planning budgets or migrations around it.
The move also mirrors how cloud providers are mixing rival AI models into their platforms, as seen when OpenAI models reached Google Cloud. For more on the latest Claude models you could run this way, see our coverage of Claude Fable 5.2.
What We Still Don’t Know
The pages do not give a launch date, per-model prices in Claude Consumption Units, or a list of regions. They also do not say whether the offering is open to everyone or limited to selected customers at first. We will update our AI news coverage once Anthropic or Google confirms the service.
Sources
Primary source: Anthropic’s Claude Platform docs for the Anthropic-managed Google Cloud offering. Additional details from Anthropic’s Google Cloud offerings overview and the Claude Code setup page.
Conclusion
Anthropic’s own documentation now describes a Claude API that it runs inside customers’ Google Cloud projects, with Google handling sign-in and billing. The pages cover setup, IAM roles, billing in Claude Consumption Units and supported features in detail, which suggests the service is close, but neither company has announced it. Until they do, there is no confirmed date or pricing. If your team lives on Google Cloud, it is worth watching for the official launch.
