Claude on Google Cloud Adds Anthropic-Run Option

Businesses buying Claude through Google Cloud now have to decide who handles their data. Google and Anthropic this week made generally available a version of Claude on Google Cloud that Anthropic operates itself, alongside the existing version that Google runs. Both offer the same models and, Google says, “the same pricing, and the same Google Cloud billing”.

The difference is in what sits underneath. With the new Anthropic-managed option, Anthropic acts as the data processor under its own commercial terms, and customers get new Claude features the day they launch. With the Google-managed option, Google remains the processor and data stays inside Google’s security boundary, but new features arrive only once Google has integrated them.

Anthropic’s version brings every Claude feature on day one

In its announcement, dated 22 October, Anthropic says Google Cloud customers using its managed option receive “the full set of Claude API features, with new features and betas shipping the same day they go live on the native Claude Platform.” That includes Claude Managed Agents and the MCP connector (both in beta), code execution, computer use, agent skills, web search, prompt caching and batch processing, as well as access to the Claude Console. Anthropic has just handed its Agent Skills format to an industry foundation.

Access is still controlled through Google Cloud identity and access management (IAM), and usage is billed on a single Google Cloud invoice. Spending also draws down existing Google Cloud commitments, a selling point Google also used earlier this week when it brought OpenAI’s GPT-6 models to Google Cloud. Claude Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5 are available on both options.

Anthropic is also offering a faster mode for Claude Opus 5.5 on its managed option, which it says delivers “up to 2.5x faster output token generation” at $8 per million input tokens and $40 per million output tokens. Customers with existing discounts are told to contact their account executive first, because discounts “cannot be applied retroactively”.

The data may leave Google’s infrastructure

Google’s documentation for the new option is more direct about the trade-off than either announcement. It says Google Cloud and Anthropic “act as separate, independent data processors for their respective parts of the service”, and that “Data storage and inference occur on Anthropic-managed infrastructure (which may route outside Google Cloud)”. Customers can set a per-request parameter to pin inference to a geography.

	Anthropic-managed	Google-managed
Data processor	Anthropic	Google Cloud
New features and betas	Same day as Claude Platform	Once Google integrates them
Data residency	US and Global endpoints	US, EU and APAC
Google controls (VPC, CMEK, Model Armor)	Not listed	Yes

Source: Google Cloud blog and documentation; Anthropic announcement, 22 October 2026.

Google recommends its own managed version for anyone who needs Google as the sole processor, data that stays within Google’s security boundary, or coverage under Google compliance programmes including FedRAMP High and HIPAA eligibility. The documentation also warns that activating the Anthropic option creates a new Anthropic organisation, so settings such as HIPAA controls and zero data retention from an existing Anthropic account “don’t carry over”.

Feature speed now comes with a compliance choice

For developers, the appeal is obvious. Agent features change quickly, and waiting for a cloud provider to integrate each beta has been a real cost for teams building on Claude through Google. Combined with Google’s work on running agents in isolated environments, such as its Agent Compute Engine, and Anthropic’s own Claude for Google Workspace add-on, the two companies are making it easier for Google customers to build with Claude without leaving familiar billing and identity tools.

The risk is that a choice which looks like a deployment setting is really a data-protection decision. A team that switches to get a new beta could move personal or regulated data to a different processor, under different terms, possibly outside Google’s infrastructure, without the compliance team noticing. UK and European organisations have particular reason to check: Google lists EU data residency only for its own managed option, and neither company says which countries the “Global” endpoint covers. The sensible approach is to treat the Anthropic-managed option as a new supplier, with its own data processing agreement and risk assessment.

The next things to watch are whether Anthropic adds regional endpoints beyond the US and Global, and how quickly the Google-managed version catches up with the betas.