LLM INFERENCE MIGRATION
Move your LLM inference to Amazon Bedrock
Running on OpenAI, Azure OpenAI, Anthropic's API or OpenRouter? On Amazon Bedrock that inference lands on your AWS bill, where AWS credits can cover it. You keep your model wherever Bedrock serves it, and for eligible companies AWS funds the work.
Why move inference to AWS
Credits can cover it
Inference on Bedrock is AWS spend, so AWS credits can cover it, subject to each credit's terms. Tagged from the start, it also counts toward MAP.
The models you use, one bill
Claude, OpenAI's models, Amazon Nova, Llama, DeepSeek and more, behind one API on your AWS bill.
Your data stays in your account
Requests run in your AWS account and regions, under your IAM policies and logged in CloudTrail. Prompts and completions never train the models.
Control over latency and capacity
Regional and cross-region inference profiles, quotas sized for your load, and provisioned throughput for steady traffic.
How the migration runs
Already on AWS? It runs in your account, through new IAM roles tagged so the spend counts. Not on AWS yet? We set up the foundation first.
Assess
- Every model call mapped
- Cost and latency targets
- A golden dataset from your traffic
- Models benchmarked side by side
Prove it on your own traffic
Mobilize
- Prompts tuned per model
- Endpoints and SDKs switched
- Auth on AWS IAM
- MAP tags from the first request
Make your app speak Bedrock
Migrate & Modernize
- Traffic moved model by model
- Regression checks
- A rollback path
- Cost view and budget alerts
Run it on your AWS bill
Is it worth it for you?
When it pays off
You pay per token through an API, the spend is growing, or you want it on one bill with the rest of your stack.
When to wait
You are on a flat-rate subscription, where moving can cost more, or you rely on a model or feature Bedrock does not offer yet. We will tell you in our first conversation.
Funded by AWS, for eligible companies
The migration work
For eligible companies, AWS can fund our work, in part or in full, through its partner programs.
Your Bedrock usage
Usage that starts before a MAP engagement does not count toward it, so talk to us before you move more traffic.
Eligibility, amounts and timing are set by AWS program terms. We confirm them with your AWS account team before work starts.
Questions we hear
Do we need new AWS accounts for this?
Not if you are already on AWS: it runs in your account, through new tagged IAM roles. If not, we set up the foundation first.
Will our prompts work as they are?
Often with small changes, which we find by benchmarking before switching.
Do we have to rewrite our code?
Usually little. Bedrock offers OpenAI- and Anthropic-compatible APIs, so many apps change an endpoint and their auth.
Can some traffic stay where it is?
Yes. Traffic moves model by model, with a rollback path until you are confident.
We already started calling Bedrock. Is it too late?
No, but usage from before a MAP engagement does not count toward it, so talk to us before you move more.
Which providers can you migrate from?
OpenAI, Azure OpenAI, Anthropic's API, OpenRouter, and other OpenAI-compatible APIs.
Spending on LLM APIs? Let's look at the bill.
Send us what you call and what you pay, and we will show you the move to Bedrock.