Amazon Web Services launched Amazon Nova Forge, a service that gives enterprises the ability to train their own foundational models.
With Amazon Nova Forge, AWS is looking to deliver real enterprise outcomes. The reality has been that off-the-shelf foundational models are inaccurate in many enterprise use cases. When enterprises build on top of open source models, results can degrade as more data is added.
Speaking at AWS re:Invent 2025, AWS CEO Matt Garman said the company is focused on model selection as well as building out Nova. Enterprises will need multiple models as a start kit to customize with their own enterprise data.
“Your data is unique. It's what differentiates you from the competition," said Garman. "If your models have more specific knowledge about you and your data and your processes, you can do a lot more."
More from re:Invent 2025
- AWS adds AI agent policy, evaluation tools to Amazon Bedrock AgentCore
- AWS launches Amazon Nova Forge, Nova 2 Omni
- AWS launches AI factory service, Trainium 4 with Trainium 4 on deck
- AWS Transform aims for custom code, enterprise tech debt
- AWS, Google Cloud engineer interconnect between their clouds
- AWS Marketplace adds solutions-based offers, Agent Mode
The goal with Amazon Nova Forge is to get better results with proprietary data in a seamless way. Garman said Nova Forge is aimed at enabling enterprises to add their unique expertise to a foundation model. Garman added that today’s techniques can only go so far in adding expertise to models.
Amazon Nova Forge gives enterprises the ability to do the following:
- Select a starting model and checkpoint.
- Mix your data with Amazon curated datasets.
- Leverage multiple checkpoints to ensure performance.
- Deploy these custom models on Amazon Bedrock for AI agent deployments.
Garman said:
“With Nova Forge, you get exclusive access to a variety of Nova training checkpoints, and then you get the ability for you to blend in your own proprietary data together with an Amazon curated training data set at every stage of the model training, this allows you to produce a model that deeply understands your information, all without forgetting the core information the thing has been trained on. We call these resulting models novellas, and then we allow you to easily upload your novella and run it in Bedrock.”
Nova Forge is being used by Reddit as well as Sony. Sony said it was putting Nova Forge in the middle of its agentic AI architecture, which is built on AgentCore.
Nova Forge is part of a broader Amazon Nova rollout. The company, which first introduced Nova models a year ago and has updated since, launched Amazon Nova 2 including the following versions:
- Nova 2 Lite, a fast cost effective reasoning model.
- Nova 2 Pro, AWS most intelligent reasoning model.
- Nova 2 Sonic, a speech-to-speech foundational model for conversational AI that's embedded into Amazon Connect.
- Nova 2 Omni, a multimodal reasoning and image generation model.
“Over the last year, we've actually extended Nova family to support more use cases and deliver more possibilities for you that deliver real value,” said Garman, who said AWS will continue to add Nova models.
Amazon Bedrock is adding new models from Google, Nvidia, Mistral including Mistral Large 3 and Ministral 3, Alibaba's Qwen and Amazon.
“We think model choice is so critical. We've never believed that there was going to be one model to rule them all, but rather that there will be a ton of great models out there, and it's why we've continued to rapidly build upon an already wide selection of models,” said Garman. “We have open rates models and proprietary models, general purpose, specialized ones. We have really large ones and small models, and we've nearly doubled the number of models that we offer in bedrock over the last year.”
