Menu
Products

SageMaker AI

4 min readintermediateUpdated 28 Sept 2026
1 · In one line

SageMaker AI, from AWS, lets you build, train and run machine learning models without managing your own servers.

1 · What it is

SageMaker AI, from AWS, lets you build, train and run machine learning models without managing your own servers.

A little history helps with the name. On 3 December 2024, the service then called Amazon SageMaker got its new name, SageMaker AI. Today the plain name Amazon SageMaker covers data, analytics and AI in one platform. The old technical names that programmers type, such as sagemaker, stayed the same.

You can use its built-in methods for training, or bring your own code and tools. It packs both training and serving into Docker containers.

Once a model is trained, it answers through an endpoint. Real-time endpoints reply quickly to each request. Serverless endpoints look after the machines and grow or shrink for you. Asynchronous endpoints are meant for batches of requests.

You do not always start from zero. SageMaker JumpStart offers publicly available foundation models and some proprietary ones from other companies.

2 · Why it exists

Getting a model from data to real users raises three problems.

Training needs machinesTeaching a model on a large dataset means setting up and looking after powerful computers.
Serving is separateA trained model cannot answer apps until it is deployed somewhere that runs it.
Skills varySome people need a no-code tool for quick tests. Others need code for more control.
3 · How it works

Follow one model from training data to a live endpoint.

How a model moves through Amazon SageMaker AI Training data and your code or a built-in algorithm feed a training job. The highlighted training job runs in a container while SageMaker AI sets up the machines. The finished model is saved to Amazon S3. It is deployed to an endpoint. The endpoint returns predictions to your app. Deleting the endpoint stops its charges. FROM TRAINING DATA TO A LIVE ENDPOINT Data + code Training job Model in S3 Endpoint Your app Delete when done your data, plus a built-in or own code runs in a container; AWS sets up machines model files copied to your S3 location real-time, serverless or async gets predictions from the endpoint deleting the endpoint stops its charges You can also start from a JumpStart foundation model instead of training from scratch.
You can also start from a ready-made foundation model in JumpStart instead of training from scratch.
  1. 1 · bringYou bring your data and choose a built-in algorithm or your own training code.
  2. 2 · trainA training job runs your work in a container while SageMaker AI sets up the machines.
  3. 3 · saveThe finished model files are copied to an Amazon S3 storage location you chose.
  4. 4 · deployYou deploy the model to an endpoint, which apps can call to get predictions.
  5. 5 · cleanWhen you no longer need the endpoint, you delete it to stop the charges.

SageMaker AI handles the machines; you still choose the data and the model.

4 · Where it's used
WhoWhat they askWhat it works with
Bank data team“Can we train a fraud model on last year's transactions without buying servers?”A training job with a built-in algorithm
Startup developer“Can I try an open language model and fine-tune it on our support tickets?”A foundation model from SageMaker JumpStart
Shop analyst“Can I build a sales forecast without writing code?”SageMaker Canvas
Photo app team“Can our app get a label for each uploaded picture within moments?”A real-time endpoint
5 · What it solves, and what it doesn't
solves
  • It sets up and manages the computers that training needs.
  • It offers a low-code way to try ideas quickly.
  • It lets you train with your own code for more control.
  • It supports training models at scale.
  • It deploys models to managed endpoints that can scale automatically.
  • It offers a catalogue of foundation models you can fine-tune, evaluate and deploy.
doesn't solve
  • Running at scale still takes knowledge of AWS and distributed training.
  • An endpoint keeps costing money until you delete it.
  • The simplest no-code option gives you little room to customise the model.
6 · Go deeper
Amazon BedrockFine-tuningInferenceFoundation modelsVertex AI

Sources used

This explainer is written in original language. The links below support its factual claims.

  1. docsWhat is Amazon SageMaker AI?, Amazon Web Services · read 28 Sept 2026
  2. docsTrain a Model with Amazon SageMaker, Amazon Web Services · read 28 Sept 2026
  3. docsHow Amazon SageMaker AI Processes Training Output, Amazon Web Services · read 28 Sept 2026
  4. docsDeploy models for inference, Amazon Web Services · read 28 Sept 2026
  5. docsReal-time inference, Amazon Web Services · read 28 Sept 2026
  6. docsAmazon SageMaker JumpStart Foundation Models, Amazon Web Services · read 28 Sept 2026
  7. docsDelete Endpoints and Resources, Amazon Web Services · read 28 Sept 2026
Suggest a correction