AWS Bedrock: the honest hands-on guide for SMEs

What is AWS Bedrock?

AWS Bedrock is a fully managed service from Amazon Web Services that lets you use leading AI language models via API – without renting GPUs, hosting models or running infrastructure yourself. You send a prompt, get a response, and pay only for the tokens you use.

What makes it special: Bedrock bundles models from many providers under a single interface. Anthropic's Claude, Mistral, Meta's Llama, Amazon's own Nova models, Cohere and others run through the same API – you can switch models without rewriting your code.

For companies it's attractive mainly because Bedrock is part of the AWS world: same billing, same access management (IAM), same compliance certifications – and the option to keep processing within European data centres.

Why Bedrock for SMEs? The GDPR argument

The biggest hurdle for AI in the German Mittelstand is rarely the technology – it's data protection. 'Are we even allowed to send customer data to a US model?' is the question that stalls almost every project.

Bedrock answers it cleanly: choose the eu-central-1 (Frankfurt) region and the EU inference profiles (model IDs prefixed eu.), and the processing of your prompts stays within the EU. AWS does not use your inputs and outputs to train the models, and offers a GDPR data processing agreement (DPA).

So you get the same Claude model as via the direct API – but with European data residency and the contractual basis your data-protection team wants to see. That's exactly what makes Bedrock the first choice for regulated industries and cautious SMEs.

How processing stays in the EU

CurrentModels with EU data residency

These models can run on Bedrock with European data residency (Frankfurt or an EU inference profile). Prices are on-demand prices per 1M tokens in US dollars.

Note: for the Claude 4.5 models, AWS adds roughly 10% at the EU/regional endpoint – the surcharge for GDPR-compliant processing in the EU.

As of 25 June 2026. Prices and availability change – the AWS pricing page prevails. Official AWS pricing ↗

CurrentPrice range: cost per 1M tokens

What does AWS Bedrock cost?

By default Bedrock bills 'on-demand': you pay per processed token, separately for input and output. There's no base fee and no idle cost – send nothing, pay nothing.

For predictable, high loads AWS offers 'Provisioned Throughput' (reserved capacity at an hourly rate). For most SMEs that's unnecessary at first: start on-demand, measure your real usage, and optimise only then.

For a sense of scale: summarising two A4 pages with a cheap model like Claude Haiku costs a fraction of a cent. Costs only become noticeable at volume – and that's exactly when the model choice from the table above pays off.

CurrentCost calculator: what does it cost per month?

Step by step: your first Claude call on Bedrock

From zero to your first response – in about 15 minutes.

1. 1
   Pick AWS account & region

   Sign in to the AWS console and switch (top right) to the 'Europe (Frankfurt) eu-central-1' region. Everything else happens in this region.

2. 2
   Enable model access

   Open Bedrock → 'Model access' and request access to the models you want (e.g. Anthropic Claude). Approval usually takes only minutes.

3. 3
   Grant IAM permissions

   Create an IAM user or role with the bedrock:InvokeModel permission – a dedicated, narrowly scoped access rather than your root account.

4. 4
   Install the SDK

   Install the AWS SDK (Python: boto3, Node: @aws-sdk/client-bedrock-runtime). If you're used to the Anthropic SDK, you can use its Bedrock variant instead.

5. 5
   Send your first call

   Call the model via the EU inference profile (eu. prefix) – see the code example below. The response comes back as JSON.

import boto3

client = boto3.client("bedrock-runtime", region_name="eu-central-1")

response = client.converse(
    modelId="eu.anthropic.claude-haiku-4-5-20251001-v1:0",
    messages=[
        {"role": "user", "content": [
            {"text": "Erkläre in einem Satz, was AWS Bedrock ist."}
        ]}
    ],
    inferenceConfig={"maxTokens": 512, "temperature": 0.2},
)

print(response["output"]["message"]["content"][0]["text"])

With the eu. prefix you use the EU inference profile – processing stays in the EU.

Bedrock or the direct API? An honest take

Bedrock isn't always the best choice. The short decision aid:

Tip: because Bedrock uses the same API for many models, you can start small and switch later without rebuilding everything.

GDPR checklist for Bedrock

To keep your Bedrock use compliant, watch these points:

• ✓Set the region to eu-central-1 (Frankfurt) or an EU inference profile (eu.) – so processing stays in the EU.
• ✓Sign a data processing agreement (DPA) with AWS – part of the AWS contract terms.
• ✓No training use: AWS does not use your Bedrock inputs and outputs to train models.
• ✓Minimise data in the prompt – send only what the task needs; pseudonymise personal data where possible.
• ✓Configure logging deliberately – enable model invocation logging only when needed, and store it compliantly.
• ✓Add the AI processing to your record of processing activities (and a DPIA where required).

CurrentLatest Bedrock news

What's happened around AWS Bedrock recently:

• 1 November 2025Claude Opus 4.5 — new Bedrock flagship, EU inference profile availableSource: AWS ↗
• 1 October 2025Claude Haiku 4.5 on Bedrock — fast, cheap, with an EU profileSource: AWS ↗
• 29 September 2025Claude Sonnet 4.5 on Bedrock — EU inference profile for data residencySource: AWS ↗

Frequently asked questions about AWS Bedrock