AWS Bedrock or Vertex AI for Claude in the EU?

You’ve landed on wanting to use Claude. Good – that’s half the decision made. The other half is where the model runs: via AWS Bedrock or Google Vertex AI. For an EU company with data protection requirements, that’s not a detail – it’s what determines how simple your compliance documentation ends up being.

What’s the same

Start with the similarities, since they’re bigger than you’d think. On both platforms, the setup is fundamentally the same:

• The cloud provider is the data processor, and Anthropic is the subprocessor. So you have a Data Processing Agreement with the hyperscaler, not directly with the model company.
• EU storage is possible on both.
• You control the encryption keys yourself – with AWS KMS on Bedrock and Google CMEK on Vertex.

If the comparison ended here, the choice would be purely a matter of taste. But it doesn’t end here.

What differs: regional geography

The most important difference is about how “EU” is actually realized.

Bedrock combines two things: direct EU regions – including Stockholm – and EU cross-region profiles that keep traffic within the geography even when it’s distributed for capacity. So you can point to a specific member-state region.

Vertex’s EU path instead relies on a multi-region endpoint. And here’s a distinction auditors care about: multi-region isn’t the same as a single member-state region. If your requirement is strict single-region – that data must stay in one specific country – a multi-region solution isn’t enough to check that box without qualification.

The detail auditors ask about

Abuse detection – the process that catches disallowed use – also stores data somewhere. On Bedrock, that storage follows the destination region, according to the documentation. That means if your traffic goes to an EU region, that process does too.

It sounds like a technicality, but it’s exactly the kind of question that comes up in an audit: “where does the data from the abuse check end up?” Having a clear answer saves time, and it’s a point to verify against current documentation before the contract is signed.

A related pitfall is assuming the choice of model itself decides where data ends up. It doesn’t – the configuration does. On both platforms you have to actively point region and profile correctly, and verify that surrounding processes like logging and abuse detection follow along. A misconfigured EU setup that happens to route traffic outside the geography looks the same day to day but fails in an audit. Provability sits in the settings, not in the model’s name.

What actually decides it

Here’s the uncomfortable truth: the model list is rarely what should decide it. When both platforms give you Claude with EU storage and your own keys, the practical factors weigh heavier:

• Which cloud agreement do you already have? Adding a service where you already have contracts and routines is cheaper than opening a new vendor relationship.
• What does your IAM and access governance look like? The platform you already master produces fewer misconfigurations.
• Which audit tools do you use? Logging and traceability should fit into what you already audit.

In practice, the cloud provider you already have maturity in often wins. It’s not an exciting answer, but it’s what holds up in operations and audit.

One concrete way to break the deadlock: ask where your logs already end up and who can already read them. If you’ve built up permissions, alerts, and audit routines in one cloud, the other means rebuilding all of that in parallel – and every new place data passes through is another line in your compliance documentation. If your residency requirement is also strict single-region, that tilts toward Bedrock, which can point to a specific member-state region; if the requirement is softer, the practical factors weigh heavier. So let the need for provable geography and your existing maturity decide, not whichever platform happens to mention Claude first.

The region and model matrix also changes often at both providers, so treat this comparison as perishable and verify current documentation before deciding. Want help weighing your existing contracts and requirements against each other? Read about our AI services or get in touch with your current situation, and we’ll go through it together.