Amazon Bedrock

advancedwidespread

Amazon's cloud service through which companies can rent and integrate large AI models from multiple manufacturers into their own applications without their own servers.

In plain terms

Amazon Bedrock is like a shopping mall of AI models. Under one roof, you find offerings from multiple manufacturers (for example, Anthropic, Meta, Mistral, and Amazon's own models), and you pay at a single checkout, on a single bill.

It is mainly designed for companies. If a business wants to build a customer service chatbot, document search tool, or AI agent, it does not need to buy its own hardware or operate a model. It rents the model on Bedrock and pays only for what it uses.

Unlike chat applications like ChatGPT, this is not a ready-made product you open and chat with. It is a toolkit from which developers assemble services. You mostly only encounter the final result and do not know that Bedrock is working in the background.

How it works

The company activates Bedrock in Amazon's cloud (AWS) and selects from the offerings which model or models it wants to use.
The company's application sends questions or tasks to the model via the internet through a standard programming interface (API) and receives the response at the same point.
The company can attach its own documents and databases, so the model can answer from the company's internal knowledge, not just its general knowledge.
You can set up guardrails that filter out prohibited topics or sensitive data, and build agents that solve tasks in multiple steps using various tools.
Amazon mostly bills based on the volume of processed text (tokens), but the company can also purchase pre-booked, fixed-fee capacity. Amazon handles the operation, expansion, and maintenance of the machines.

An everyday example

At 10 p.m., you type into the chat window on your insurer's website and within seconds get a sensible answer about the status of your claim. Quite possibly, the agent replying is backed by a language model rented through Bedrock, which assembles the answer from the insurer's own policies and your case data.

Who is affected

developerYou can test and integrate models from multiple vendors through a single interface, without signing a separate contract with each or running your own server. → Code

entrepreneur, executiveIt is up to them to decide which cloud provider the company's AI solution will be built on, a choice that determines costs and dependency for years.

lawyer, citizenFrom a data protection perspective, it matters on a server in which country the company processes your data, and under what contractual terms. → Law

everyoneBehind many everyday customer service chatbots, search engines, and apps lies such a rented model, even if this is not stated anywhere.

A common misconception

MythAmazon Bedrock is an AI model, Amazon's response to ChatGPT.

In realityBedrock is not a model but a platform on which models from many manufacturers are available. Amazon's own models (such as the Nova family) make up only part of the offering.

Limits and risks

Anyone building everything on Bedrock and Amazon's cloud will find it difficult and expensive to migrate to another provider later.
Usage-based fees can rise quickly with high traffic, so costs must be monitored continuously.
Not every model and feature is available in every region, which, given European data-handling requirements, may narrow the choice.
The platform does not eliminate model errors: a model running on Bedrock can also make mistakes or hallucinate, so its answers must be verified.

Similar, but not the same

Amazon SageMaker – SageMaker (its machine learning part has run as SageMaker AI since the end of 2024) is mainly a workshop for building, training, and operating custom models, while Bedrock rents out ready-made large models.
ChatGPT – ChatGPT is a ready-made chat application for end users; Bedrock is a backend service for developers and companies.
Microsoft Foundry (formerly Azure AI Foundry) – This is Microsoft's similar-purpose model platform in its own cloud, with a different model offering and surrounding services.

Related concepts

Broader

Cloud service, AI infrastructure, Model as a service

Narrower

Amazon Bedrock AgentCore, Guardrail, Knowledge base

On the edge

Foundation model, API, Large language model, AI-agent, Google Vertex AI

A brief history

Amazon announced Bedrock in April 2023 and made it generally available in September 2023. Since then, it has expanded with agent-building tools, guardrails, and an increasing number of manufacturer models.

Where things stand

At the end of 2026, Bedrock is among the leading model platforms of the major cloud providers, and many companies build their AI applications on it. The focus of development has shifted to operating agents.

Status: 2026 November

In the news

Appears in 5 Agent Age stories; the curve shows weekly occurrence in the news magazine.

OpenAI's Agents API offers Codex harness as a service with computer use; data stored only in USSep 30, 2026 · Code
Amazon Bedrock AgentCore flaws could have exposed AWS credentialsSep 29, 2026 · Code
Fully automated excavators deployed on construction sites; Skild AI unveils S1 robot foundation modelAug 31, 2026 · Ipar
AWS AgentCore payments generally available; OpenAI and AWS publish x402 guideAug 17, 2026 · Money
OpenAI Codex arrives on Amazon Bedrock and gets a security API connectionAug 4, 2026 · Code

Also known as

Amazon Bedrock · AWS Bedrock · Bedrock · Amazon Bedrock AgentCore · Bedrock Agents · Bedrock agent · az Amazon AI-modellplatformja

Sources

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.