Amazon Web Services (AWS) has announced the launch of Nova Forge, a service designed to help companies create personalized versions of AI models. Instead of building a model from the ground up, which is a process that can cost millions of dollars and require thousands of hours of expert labor, businesses can use Nova Forge to modify existing AWS models. These customized versions, which AWS refers to as “Novellas,” allow companies to integrate their own private data while benefiting from the core intelligence already built into Amazon’s frontier models.
The platform allows users to choose a starting model and then apply various levels of training, including pre-training, mid-training, or post-training. This flexibility means a business can fine-tune a model for a specific industry, such as legal, finance, or healthcare, without losing the general reasoning capabilities of the original system. By mixing proprietary company data with curated Amazon datasets, businesses can produce AI tools that are more accurate and relevant to their specific operations. Once a custom model is ready, it can be deployed through Amazon Bedrock, which provides the infrastructure needed to run AI applications at scale.
Table of Contents
Introducing the Nova 2 family
Alongside Nova Forge, AWS has revealed the next generation of its base technology, known as Amazon Nova 2. This new series includes four foundation models that are categorized into different tiers based on their complexity and intended use. The tiers are designed to handle everything from simple, everyday automation tasks to highly complex data analysis and creative work. One of the standout features of the Nova 2 family is a new speech-to-speech capability, which allows for more natural, human-like conversations between AI assistants and users.
The move to provide both the models and the tools to customize them shows AWS’s commitment to becoming a primary destination for corporate AI development. By offering a standardized way to build custom models, AWS is competing with other cloud giants like Google and Microsoft, who have similar offerings. AWS CEO Matt Garman noted that the goal of Nova Forge is to prevent models from “forgetting” their core reasoning skills when new data is added, which is a common technical challenge in AI training. This focus on reliability and precision is intended to give professional users more confidence in deploying AI across their business units.
Release and pricing information
AWS Nova Forge is currently available for enterprise customers, with several internal Amazon teams already utilizing the service to build custom features. The platform is offered at a starting price of $100,000 per year, which covers access to the training tools and the initial customization infrastructure. This price point is positioned as a significant saving compared to the costs of independent model development. Additional costs may apply based on the amount of data processed and the compute power required for specific training runs.
The Amazon Nova 2 foundation models are available to all AWS users through the Bedrock platform. Pricing for these models follows a usage-based structure, where companies pay for the number of tokens or data processed by the AI. This allows smaller firms to experiment with advanced AI features without a massive upfront investment. AWS plans to continue expanding the capabilities of Nova Forge throughout 2026, including more automated tools for data preparation and advanced monitoring features to help companies track the performance of their custom “Novellas” over time.
