What is Amazon Nova Forge?
Amazon Nova Forge is a service that, according to AWS, lets organisations build their own frontier models using Amazon Nova. AWS describes it as a “first-of-its-kind service” offering what it presents as the easiest and most cost-effective way to build a frontier model of your own.
At its centre is the concept of open training models: you get access to a variety of early model checkpoints and can blend proprietary data with Amazon-curated data sets at every stage of training. The aim is to maximise learning from proprietary data while minimising the risk of forgetting foundational skills such as reasoning.
Core Features
- Checkpoint access: Checkpoints across all phases of model development, including the use of new Nova models before they are widely available
- Data blending: Combining proprietary data with Amazon Nova-curated training data
- Reinforcement learning: Performing reinforcement learning with reward functions in your own environment
- Push-button recipes: Recipes optimised for Nova, available through visual workflows or a command line interface
- Responsible AI Toolkit: Aligning models to Amazon Nova’s responsible AI guidelines during training, plus runtime controls to moderate model responses during inference
- SageMaker integration: Subscription management through the SageMaker AI console; HyperPod users need a Restricted Instance Group in the cluster
Typical Use Cases
Domain-specific models: Organisations with substantial proprietary data train models tailored to their domain.
Early use of new Nova models: Checkpoint access allows work with Nova models before they become widely available.
Reinforcement learning against your own objectives: Reward functions in your environment align model behaviour with measurable goals.
Custom models with responsible AI guardrails: The Responsible AI Toolkit applies during training and at runtime.
Benefits
- Build on a pre-trained Nova foundation instead of starting entirely from scratch
- Proprietary data can be incorporated at every training stage
- Recipes optimised for Nova instead of self-built training pipelines
- Responsible AI guardrails are part of the training and inference process
- Runs inside AWS with connections to SageMaker AI and Amazon Bedrock
Integration with innFactory
As an AWS Reseller, innFactory supports you with Amazon Nova Forge: verifying the prerequisites and requesting the subscription, preparing your training data, planning training and evaluation phases, and assessing Region availability against your data protection and sovereignty requirements. Because no EU Region is currently documented for Nova Forge, we clarify this point before a project starts.
Typical Use Cases
Frequently Asked Questions
What is Amazon Nova Forge?
According to AWS, Amazon Nova Forge is a service that lets organisations build their own frontier models using Nova. It introduces the concept of open training models: you get access to early model checkpoints and the ability to blend proprietary data with Amazon-curated data sets at every stage of model training.
Which capabilities does Nova Forge provide?
The documentation lists access to checkpoints across all phases of model development and the ability to leverage new Nova models before they are widely available, blending proprietary data with Amazon Nova-curated training data, reinforcement learning with reward functions in your environment, push-button recipes optimised for Nova through visual workflows or a command line interface, and the Responsible AI Toolkit to align models to Amazon Nova's responsible AI guidelines during training plus runtime controls to moderate model responses during inference.
How do I get access?
Access runs through a subscription. Per the documentation you add the tag forge-subscription with value true to your console IAM role, then open SageMaker AI Console > Model training and customization and click Nova Forge. That page carries service details, pricing information, and the option to request and manage the subscription.
Which permissions are required?
Per the documentation the role needs permission to call the ListAttachedRolePolicy API, and the response must include either the AdministratorAccess or the AmazonSageMakerFullAccess policy. The sign-in role must also be able to call ListRoleTags, and its tags must include forge-subscription.
In which Regions is Nova Forge available?
The documentation names US East (N. Virginia) with no limitations and US West (Oregon). In US West (Oregon), Amazon Bedrock inference is not available; AWS recommends deploying via SageMaker inference there or copying the model to US East (N. Virginia) using Amazon Bedrock model copy.
Is SageMaker HyperPod supported?
Yes. For SageMaker HyperPod users the documentation names one additional step: adding a Restricted Instance Group (RIG) to the HyperPod cluster.
Note: All product information on this page has been compiled with care, but is provided without guarantee and may be outdated or incomplete. Cloud services evolve rapidly — features, pricing, SLAs, and availability change frequently. Authoritative and up-to-date information can only be found on the official product page of AWS (official documentation). This page does not represent an offer by AWS.