Claude on Google Cloud now has two operators. Choose by who processes your data, not by the model
Google Cloud made an Anthropic-managed Claude option GA beside the Google-managed one. Same models, prices and bill; a different data processor, terms, controls and feature timing.
Listen to this article · 4 min
AI-generated narration of the full article.
If two teams in your company both “use Claude on Google Cloud” from next month, are they under the same contract? Not necessarily, and the answer will not be visible in the bill.
On October 23, Google Cloud made an Anthropic-managed deployment option for Claude generally available, alongside the Google-managed option customers already had through Model Garden. Google’s own summary is precise: both options deliver the same models on day zero, at the same pricing, on the same Google Cloud billing. “The difference comes down to who acts as your data processor and whether you need access to the new Claude platform features when they launch.”
Same models, two operating models
What is shared and what differs
Shared by both
- Same Claude models on day zero
- Same pricing
- Google Cloud billing and eligible commitments
- Google Cloud infrastructure
Anthropic-managed
- Anthropic is the data processor
- Anthropic's Commercial Terms
- Claude platform features on day zero, Managed Agents in beta
Speed of features
Google-managed
- Google Cloud is the data processor
- Existing Google Cloud agreements
- VPC Service Controls, CMEK, residency in U.S., EU, APAC
Depth of controls
In the Anthropic-managed option, Anthropic operates the serving stack on Google Cloud infrastructure, as the data processor under Anthropic’s commercial terms. Customers authenticate with Google Cloud IAM, keep Cloud Audit Logging and draw down eligible Google Cloud commitments. What they gain is timing: Claude platform features arrive on day zero, including Claude Managed Agents (still in beta), skills and code execution.
In the Google-managed option, Google Cloud is the data processor under the customer’s existing agreements. Google positions it for “strict governance, compliance, and sovereignty requirements”: VPC Service Controls, customer-managed encryption keys, data residency in the U.S., EU and APAC, and native integrations with BigQuery and Model Armor. New Claude platform features “roll out over time” as Google integrates them.
The decision is a data-processing decision first
We have described the same pattern for Claude on Azure and AWS and for OpenAI’s models on Google Cloud: the model is constant, and the boundary around it is what you are choosing. This announcement makes the choice unusually explicit, because it happens inside one cloud and one bill. A procurement team cannot rely on “we already have a Google Cloud contract” to settle it.
Four questions decide most workloads:
- Which contract must govern this data? If your data protection agreement, security addendum and audit rights were negotiated with Google Cloud, the Google-managed option keeps you inside them. The Anthropic-managed option adds Anthropic as processor under its own terms, which legal and privacy teams need to review like any new processor.
- Does the workload depend on perimeter and key controls? VPC Service Controls and CMEK are described for the Google-managed option. Google’s post does not list them for the Anthropic-managed one, so confirm with both vendors before assuming parity.
- Does it need a beta feature now? Managed Agents, skills and code execution arrive first on the Anthropic-managed side. For a prototype that is an advantage; for a regulated workflow, a beta feature is rarely the right foundation anyway.
- Where must processing happen? Residency in the U.S., EU and APAC is listed for the Google-managed option. If your obligations name a specific region, check it explicitly for each option.
Write the route down, per workload
The practical risk is not choosing wrong once. It is choosing differently by accident: one team picks the option with the newest feature, another the one with the controls, and nobody records that the company now has two processors for the same model. We argued that every route to a model is a supplier and should be recorded per workload. Here that record needs one more field: who operates the stack.
A reasonable default for most enterprises is the Google-managed option for production workloads touching regulated or customer data, and the Anthropic-managed option for teams that need platform features early and whose data the Anthropic terms already cover. The useful part is that the choice no longer forces a change of cloud, billing or identity system. The part that still requires work is the contract.