Claude on Google Cloud Anthropic-Managed: What’s Available and How to Choose

Google Cloud has made an Anthropic-managed deployment option for Claude generally available. It gives Google Cloud customers access to Anthropic’s Claude platform features while using Google Cloud for identity integration, API access, audit logging, and billing. The option is designed for teams that want early access to Claude platform capabilities without managing a separate cloud billing relationship for model use.

The main decision is not simply where the service is sold. It is who operates the model-serving platform, which company processes data for which part of the service, and which governance controls your organization requires. Google-managed Claude remains available for teams that need Google Cloud to operate inference and provide its associated security and residency controls.

What the Anthropic-managed option includes

The service brings Claude’s developer platform to Google Cloud infrastructure while Anthropic operates the Claude platform. Google Cloud provides the integration layer: customers can use Google Cloud Identity and Access Management (IAM), Cloud Audit Logs, and Google Cloud Marketplace billing. Anthropic operates the Claude API and related platform features under Anthropic’s commercial terms.

The launch makes Claude platform capabilities available on Google Cloud, including Claude Managed Agents (beta), skills, and code execution. Managed Agents provides a configurable agent harness for managed, longer-running or asynchronous work. Skills are reusable resources that can give an agent task-specific context and workflows. Code execution lets Claude write and run code in a sandboxed environment for tasks such as analyzing data or generating files.

These features can reduce the amount of agent infrastructure a team needs to build and operate itself. They do not remove the need to design access controls, test agent behavior, review outputs, and decide which data or tools an agent may use. Managed Agents is identified as a beta feature, so teams should validate its current behavior and availability against their production requirements.

How the two Google Cloud deployment choices differ

Google Cloud now offers two distinct ways to access Claude. Both options provide access to Claude models on Google Cloud, and Google’s announcement describes day-zero model availability for both. The difference is the operating model, data-processing arrangement, feature rollout, and set of cloud controls.

Both routes can make sense within one cloud strategy. An organization might select Anthropic-managed Claude for applications that need new Claude platform features quickly, while using Google-managed Claude for workloads whose requirements depend on Google-operated inference or Google Cloud’s specific security boundary. The appropriate design depends on each workload’s contractual, regulatory, and technical requirements—not just on which interface is more convenient.

Understand the data and residency boundaries

“On Google Cloud” does not mean that all processing and storage necessarily remain inside Google Cloud’s security boundary. Google’s documentation says data storage and inference for the Anthropic-managed service occur on Anthropic-managed infrastructure, which may route outside Google Cloud. It also describes Google Cloud and Anthropic as separate, independent data processors for their respective parts of the service.

For Anthropic-managed Claude, Anthropic’s Commercial Terms of Service, Data Processing Addendum, and Usage Policy govern use of the Claude API and Console. Google Cloud’s and Google Cloud Marketplace’s terms govern the Google Cloud integration layer, including IAM authentication, API routing, Cloud Audit Logs, and Marketplace billing. Before sending sensitive information, teams should review the applicable terms and confirm that this division of responsibilities fits their obligations.

The documentation describes an inference_geo request parameter for pinning inference to a specific geography. That option should not be treated as interchangeable with Google-managed regional data residency or as proof that every part of storage, routing, or service operation stays in a particular location. Confirm the exact geographic requirements and current service behavior with the provider documentation and your compliance team.

The Anthropic-managed service follows the same data-retention policy as Anthropic’s first-party Claude API. Zero Data Retention may be available on request; it is not something to assume is automatically enabled. Google’s setup guide also says a new, separate Anthropic organization is created for the Google Cloud project. Existing first-party organization API keys, workspaces, HIPAA controls, ZDR settings, and other settings do not carry over automatically.

Practical steps before adopting it

1. Classify the workload. Document the kinds of prompts, files, and outputs the application will handle, along with applicable residency, retention, and regulatory requirements.
2. Select the operating model. Choose Anthropic-managed if direct access to Claude platform features and betas is important and Anthropic’s processing terms work for the workload. Choose Google-managed if the requirements call for Google-operated inference, Google Cloud’s security boundary, or its listed governance integrations.
3. Review the contract and processor arrangement. Check the relevant Google Cloud and Marketplace terms as well as Anthropic’s commercial terms, data-processing terms, and usage policy. Clarify which provider handles each service layer.
4. Plan identity and permissions. The Anthropic-managed route uses Google Cloud IAM rather than an Anthropic API key. The documentation lists separate roles for administration and API use: roles/claude.admin for the Claude Console and roles/claude.user for API calls. Apply least privilege and test access with the identities your application will use.
5. Set up the service and organization. In Google Cloud Console, select the Anthropic-managed option on the Claude models page, enable the Claude API if needed, and purchase the product through Marketplace. The setup flow provisions a separate Anthropic organization and requires an administrator to complete organization details and accept Anthropic’s terms.
6. Test your application path. The API uses the Messages API pattern with the claude.googleapis.com endpoint and Google Cloud authentication. Check the current documentation for workspace-scoped endpoint details, supported models, feature headers, and SDK instructions before migrating existing code.
7. Check cost and capacity assumptions. Google Cloud describes standard pricing and eligible commitment drawdown, with Marketplace billing. Anthropic manages rate limits and quotas for the Anthropic-managed option; plan to review quota requirements and monitor usage rather than assume Google Cloud’s usual Agent Platform quota behavior applies.

For an initial evaluation, test representative prompts and tool workflows, including error handling, latency, access permissions, and retention requirements. If you are comparing the two deployment options, run the same workload through each and record both the developer experience and the operational controls you can verify. Avoid treating feature parity, quotas, or governance as identical just because both choices expose Claude through Google Cloud.

Limitations and trade-offs to keep in mind

• Different processor and terms: Anthropic operates the Claude service and processes data for that portion; Google Cloud handles its integration layer. Review both providers’ terms.
• Not the same residency boundary: Anthropic-managed storage and inference may occur outside Google Cloud. Geography controls should be checked against the workload’s specific obligations.
• Separate Anthropic organization: Existing Anthropic account settings, keys, and controls do not automatically transfer to the organization created through Google Cloud.
• Beta status: Managed Agents is beta. Assess its feature maturity and suitability for the intended use before depending on it for critical processes.
• Different quotas and integration surface: Anthropic manages rate limits and quotas for its managed option. Teams may also need to adapt API calls and operational procedures when moving between deployment types.
• Terms and pricing should be confirmed: Marketplace billing and eligible commitment drawdown can simplify purchasing, but customers should check current commercial terms, pricing, and commitment eligibility for their account.

Frequently asked questions

Is Claude on Google Cloud Anthropic-managed the same as Google-managed Claude?

No. Both provide access to Claude models through Google Cloud, but Anthropic operates the Anthropic-managed Claude platform, while Google Cloud operates the Google-managed deployment. Their data-processing terms, feature rollout, quota management, and available governance controls differ.

Does Anthropic-managed Claude run entirely inside Google Cloud?

No such guarantee should be inferred from the service name. Google Cloud documentation says Anthropic-managed data storage and inference occur on Anthropic-managed infrastructure, which may route outside Google Cloud. Use the documented geography options and verify requirements with the providers if residency is important.

Can I use Managed Agents, skills, and code execution?

These are among the Claude platform features available with the Anthropic-managed option. Managed Agents is labeled beta. Review the current feature documentation and test the capabilities you need before relying on them in production.

Do I need a separate Anthropic account?

The setup provisions a separate Anthropic organization for the Google Cloud project. Existing first-party Anthropic organization settings and credentials do not automatically carry over. The setup also requires completing the organization details and accepting Anthropic’s applicable terms.

How does authentication work?

The Anthropic-managed Google Cloud service uses Google Cloud IAM and OAuth-based Google Cloud authentication rather than an x-api-key header. The documentation identifies roles/claude.user for API access and roles/claude.admin for Console access.

Which option should a regulated organization choose?

There is no universal answer. Google’s documentation recommends the Google-managed option when a workload needs Google-operated inference, Google Cloud to be the sole data processor, Google Cloud’s security boundary, or specified compliance and residency controls. Organizations should validate the exact requirements and current eligibility with their cloud and compliance teams.

Bottom line

The Anthropic-managed Claude option expands choice for Google Cloud customers who want Claude platform features, including Managed Agents beta, skills, and code execution, alongside Google Cloud IAM and Marketplace billing. The trade-off is a distinct operating and data-processing model: Anthropic manages the Claude platform, while Google Cloud supplies the integration layer. Teams that need Google-operated inference and Google Cloud’s governance and residency controls should compare it carefully with the Google-managed deployment before choosing.

For setup and current technical details, consult the Google Cloud documentation for Anthropic-managed Claude and the Google Cloud availability announcement.