Secure AI for Business

Azure OpenAI and Amazon Bedrockprivate models in your own cloud

The strongest control over where data is processed and how long it is kept, for companies with engineers to build around the model.

Contents
  1. Azure OpenAI
  2. Amazon Bedrock
  3. Strengths and limits
  4. Pricing
  5. What to ask before building
  6. Sources
TypePrivate model deployment in your cloud
PricingPublished
Websitelearn.microsoft.com

When a regulator or a large customer asks where prompts are processed and who can see them, the two big clouds give the most precise answers, because the model runs inside the company's own cloud account.

Best for: Companies with their own engineers that want to build a private AI application on frontier models inside their own cloud account, in a region they choose.

Not the right fit for: Companies without engineers: these are building blocks, not an assistant staff can open on Monday.

Azure OpenAI

Microsoft says prompts, completions, embeddings and training data are not available to other customers or to OpenAI and other model providers, and are not used to train generative AI foundation models without the customer's permission. Prompts and responses are processed within the customer's chosen geography, unless a Global or Data Zone deployment is chosen. Stored data is encrypted at rest with AES-256, with the option of a customer-managed key. For abuse monitoring, flagged prompts and completions can be stored for human review by authorised Microsoft employees, located in the European Economic Area for EEA deployments⁠1.

Amazon Bedrock

AWS says model providers have no access to Bedrock's model deployment accounts, and therefore no access to customer prompts and completions⁠2. Retention is set per region by mode: in zero-retention mode no request or response data is written to durable storage, and a company can enforce that mode across its accounts with service control policies. Some models require AWS human review as a condition of access, in which case prompts and completions are kept within AWS for up to 30 days and are not shared with the model provider⁠3.

Strengths and limits

Strengths

  • Model providers cannot see prompts and completions: Microsoft and AWS both state this for their hosted models.
  • Location control: Azure processes prompts within the chosen geography for regional deployments; Bedrock retention settings are per region.
  • Bedrock’s zero data retention mode writes no request or response data to durable storage and can be enforced across an organisation with policies.
  • Azure stores feature data encrypted at rest with AES-256, with the option of a customer-managed key.

Limits

  • Abuse monitoring can mean retention: on Azure, flagged prompts can be stored for human review; on Bedrock, some models require AWS review and retain data for up to 30 days.
  • Global and Data Zone deployments on Azure may process prompts outside one region.
  • Everything around the model, from permissions to approvals, is what the company’s own engineers build.

Choose another option if: You have no engineering team (choose a business plan or an owned platform), or you want approvals per action designed with you rather than built from scratch (OpenLazarus).

Pricing

Pay per input and output token, with provisioned capacity and batch options; rates are listed per model and region on each vendor’s pricing page.

What to ask before building

  • Which deployment type keeps processing in our region, and which models are available there?
  • Do the models we need require retention or human review?
  • Who will build and run permissions, logging and approvals around the model?
  • What will a month of our expected usage cost at today's per-token rates?

For these clouds beside the other options, see the comparison table.

Sources

  1. Data, privacy, and security for Foundry Models sold by Azure. Microsoft Learn, 18 May 2026. Accessed 19 Nov 2026.
  2. Data protection in Amazon Bedrock. AWS Documentation. Accessed 19 Nov 2026.
  3. Data retention in Amazon Bedrock. AWS Documentation. Accessed 19 Nov 2026.
  4. Azure OpenAI Service pricing. Microsoft Azure. Accessed 19 Nov 2026.
  5. Amazon Bedrock pricing. AWS. Accessed 19 Nov 2026.