---
title: Does Amazon Bedrock cross-region inference move data out of your region?
url: https://opencontroller.lyzr.com/blog/bedrock-cross-region-inference-and-data-residency
published: 2026-11-16
updated: 2026-11-16
author: Lyzr
---

# Does Amazon Bedrock cross-region inference move data out of your region?

**Yes, prompts and responses can be processed in another Region. Geographic inference profiles keep processing inside one geography, such as the US, EU or APAC; global inference profiles can process in any supported commercial AWS Region. Logs stay in your source Region, and AWS says data in transit is encrypted across its network, but if the model needs abuse-detection storage, prompts and outputs are stored in the destination Region.** So the residency question is which profile your agents use, and whether policy enforces it.

Many newer Bedrock models are only available through inference profiles, so agents often use cross-region inference without anyone deciding to. That makes this a governance question, not just an architecture detail.

## What is the difference between geographic and global profiles?

| | Geographic profile | Global profile |
|---|---|---|
| ID prefix | `us.`, `eu.`, `apac.`, `in.` and others | `global.` |
| Where processing can happen | Regions inside that geography | Supported commercial Regions worldwide |
| Pricing | Based on source Region | About 10% lower input and output token price than geographic, per AWS |
| CloudWatch and CloudTrail | Source Region | Source Region |
| `aws:RequestedRegion` in IAM evaluation | Destination Regions | `unspecified` |

## What does AWS say about data location?

From AWS's geographic cross-region inference documentation:

- Requests to a geography-tied profile "stay within that geography".
- "Your input prompts and output results might move outside of your source Region during cross-Region inference."
- "All data will be transmitted encrypted across Amazon's secure network."
- "To the extent we store data for abuse detection, your input prompts and output results will be stored in the destination region."

For global profiles, AWS advises organisations with residency requirements to assess whether global inference fits their compliance framework, since requests may be processed in other commercial Regions.

## How do you enforce the residency decision?

**1. Grant the right resources only.** To call a geographic profile, a role needs `bedrock:InvokeModel` on the inference profile ARN and on the foundation model in the source Region and every destination Region in that profile. Scope the foundation-model grant with the `bedrock:InferenceProfileArn` condition so it only works through approved profiles.

**2. Block global profiles where needed.** AWS documents an SCP that denies `bedrock:*` when `aws:RequestedRegion` equals `unspecified` and the `bedrock:InferenceProfileArn` matches `inference-profile/global.*`. Region-name conditions alone do not catch global calls.

**3. Fix Region-deny SCPs.** If an SCP allows only one Region, cross-region inference fails when it routes elsewhere. Either allow all destination Regions in the profile or add an exemption using `bedrock:InferenceProfileArn`. AWS notes the exemption cannot bypass the Region restriction on the originating call.

**4. Route through a gateway.** Put the approved profile IDs in gateway routes so agents cannot choose a `global.` profile themselves. See [routing Bedrock models through an LLM gateway](/blog/route-bedrock-models-through-an-llm-gateway).

## A residency decision checklist

1. List the models your agents use and whether each is available in-Region, geographically or only globally.
2. Classify each agent's data. Personal, regulated or contract-restricted data usually means geographic at most.
3. Record the approved profile per agent in the registry.
4. Apply IAM scoping with `bedrock:InferenceProfileArn`.
5. Add the global-deny SCP for accounts with residency requirements.
6. Check which models use abuse-detection storage, since that data is stored in the destination Region.
7. Review CloudTrail for calls using profiles that were not approved.
8. Document the decision as evidence for your DPIA or residency review. See [DPIA for AI agents](/blog/dpia-for-ai-agents).

## What about quotas and cost?

Cross-region profiles add throughput by spreading load across Regions, which is often why teams adopt them. AWS notes some models consume output-token quota at a 5x burn-down rate. Global profiles are cheaper per token, so expect pressure to use them; make the residency trade-off explicit rather than letting cost decide by default. Normalise profile IDs to model names in cost reports so `us.` and `global.` usage of the same model roll up together.

For broader residency patterns across LLM and agent traffic, see [how to enforce data residency for LLM and agent traffic](/blog/how-to-enforce-data-residency-for-llm-and-agent-traffic).

## How Opencontroller does this

Opencontroller auto-discovers Bedrock agents, models and API keys alongside Azure and Google Cloud, and Posture flags risky permission and configuration patterns. Its LLM Gateway (Lyzr Gateway) puts control in the path with routing, fallback and cost attribution per agent and per version, so approved inference profiles can be enforced per route. Policy as code maps controls to frameworks including NIST and ISO 42001. It runs on Lyzr SaaS or customer-controlled infrastructure. Start with [what an AI agent control plane is](/blog/what-is-an-ai-agent-control-plane) or [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Do EU inference profiles keep data in the EU?**
AWS states that requests to a geography-tied profile stay within that geography. Processing can move between EU Regions in the profile.

**Where are logs written for cross-region calls?**
CloudWatch and CloudTrail record entries in your source Region, including for global profiles.

**How do you block global inference profiles?**
Use an SCP that denies Bedrock actions when `aws:RequestedRegion` is `unspecified` and the inference profile ARN matches `global.*`, as AWS documents.

## Key facts

Bedrock cross-region inference: geographic profiles (prefixes us., eu., apac., in.) keep processing within the geography; global profiles (global.) can process in any supported commercial Region and cost about 10% less per token than geographic; pricing based on source Region; CloudWatch and CloudTrail log in source Region; data transmitted encrypted across Amazon's network; prompts/outputs stored in destination Region where abuse-detection storage applies. IAM: bedrock:InvokeModel on inference profile ARN plus foundation model in source and all destination Regions, scope with bedrock:InferenceProfileArn. Global calls evaluate aws:RequestedRegion as unspecified; AWS SCP example denies bedrock:* for unspecified + inference-profile/global.*. Region-deny SCPs need destination Regions or an InferenceProfileArn exemption. Opencontroller (by Lyzr): auto-discovery, Posture, LLM Gateway route control; SaaS or customer-controlled infrastructure; starts from $250K per deployment per year.

## Related

- [How do you enforce data residency for LLM and AI agent traffic?](https://opencontroller.lyzr.com/blog/how-to-enforce-data-residency-for-llm-and-agent-traffic.md)
- [Why can an agent not invoke a Bedrock model you can see in the console?](https://opencontroller.lyzr.com/blog/bedrock-model-access-and-inference-profiles-for-agents.md)
- [How do you restrict which Amazon Bedrock models agents can use?](https://opencontroller.lyzr.com/blog/restrict-amazon-bedrock-models-with-scps.md)
- [More on Platform Guides](https://opencontroller.lyzr.com/blog/cluster/platform-guides)
