Introducing AWS Nova Forge and Nova 2: Revolutionizing Custom AI Model Development
Unlocking Tailored AI with Nova Forge
Amazon Web Services (AWS) has launched Nova Forge, an innovative platform designed to empower businesses to tailor AI models precisely to their unique requirements. Announced at AWS re:Invent 2025, Nova Forge enables users to start with a base model and checkpoint, then seamlessly integrate their proprietary data alongside Amazon’s curated datasets. This hybrid approach culminates in deployment through AWS Bedrock, streamlining the creation of cutting-edge AI solutions.
Priced at $100,000 annually, Nova Forge offers companies the ability to develop what AWS terms “Novellas”-customized iterations of the AWS Nova foundational models. These can be fine-tuned at various stages: pre-training, mid-training, or post-training. This flexibility dramatically reduces costs and time compared to building models from the ground up, which can otherwise require millions of dollars and thousands of labor hours.
Addressing the Challenges of Model Customization
Matt Garman, AWS CEO, highlighted a key challenge in AI model training during his keynote: “When models undergo extensive customization, especially in post-training phases, they risk losing some of the core reasoning capabilities initially learned.” He likened this phenomenon to human language acquisition-while learning a language early in life is relatively effortless, acquiring a new language later becomes significantly more difficult. Similarly, AI models can struggle to retain foundational knowledge when heavily adapted later in their training lifecycle.
Nova 2: Expanding the Frontier of Foundation Models
Alongside Nova Forge, AWS unveiled Amazon Nova 2, a suite of four new foundational AI models designed to cater to a broad spectrum of workloads. These models are offered in three scalable tiers, ranging from everyday applications to highly complex tasks. Notably, Nova 2 introduces advanced speech-to-speech capabilities, enabling more natural, human-like conversational AI experiences.
Since its initial debut in 2024, the Nova platform has been adopted internally within Amazon, demonstrating its robustness and versatility. The enhancements in Nova 2 further solidify AWS’s commitment to providing scalable, customizable AI tools that meet the evolving demands of enterprises worldwide.
Why Custom AI Models Matter Today
In 2025, the demand for AI models tailored to specific business needs has surged, with Gartner reporting that over 70% of enterprises plan to invest in customized AI solutions within the next two years. Nova Forge’s approach-leveraging pre-existing models and enriching them with proprietary data-offers a cost-effective and efficient pathway to innovation.
For example, a financial services firm could use Nova Forge to fine-tune a model on their unique transaction data, enhancing fraud detection accuracy without the prohibitive costs of developing an AI system from scratch. Similarly, healthcare providers might customize speech-to-speech models to better understand diverse patient accents and dialects, improving telemedicine interactions.
Getting Started with AWS Nova Forge and Nova 2
Businesses interested in harnessing the power of Nova Forge can subscribe to the service and begin by selecting a starter model that aligns with their objectives. From there, they can upload their datasets, combine them with Amazon’s extensive data resources, and deploy the customized AI via Bedrock’s scalable infrastructure.
As AI continues to transform industries, AWS’s Nova Forge and Nova 2 represent significant strides toward democratizing access to sophisticated, adaptable AI technologies.
