Amazon Web Services (AWS) has unveiled a new service, Nova Forge, aimed at enabling enterprises to customize artificial intelligence (AI) models to better suit their specific business contexts. This announcement was made during the company’s re:Invent conference, where AWS CEO Adam Garman emphasized the need for businesses to integrate proprietary data directly into their AI systems.
Many companies that have adopted AI are now grappling with the limitations of existing models, which often lack the necessary business context due to their reliance on generic, publicly available data. Garman pointed out that traditional third-party models do not typically have access to proprietary data, making it impractical for businesses to build models from scratch or fine-tune existing ones.
To address these challenges, AWS is offering a solution that allows companies to blend their proprietary data with AWS-curated training data during the training phase. Garman posed a critical question: “What if you could integrate your data at the right time during the training of a frontier model and then create a proprietary model that was just for you?” The response is Nova Forge, designed to help businesses internalize their unique logic without the need for constant external referencing.
Revolutionizing AI Customization
Analysts have recognized the significance of Nova Forge in addressing the limitations of existing AI customization methods. John Tyagi, practice leader of AI stack at HyperFRAME Research, noted that while techniques like prompt engineering and supervised fine-tuning are valuable, they have inherent constraints. He stated, “Enterprises come up against context windows, latency, orchestration complexity. It’s a lot of work, and prone to error, to continuously ‘bolt on’ domain expertise.”
In contrast, the approach offered by Nova Forge aims to simplify the process. Jesse Walter, executive director of software research at ISG, highlighted that modifying a large language model (LLM) to incorporate relevant information can streamline the inference process, making it easier to manage and maintain.
The strategies of AWS and its competitor, Microsoft, differ significantly. According to HFS Research associate practice leader Ravi Tyagi, Microsoft seeks to own the AI experience while AWS focuses on providing the tools necessary for businesses to create their own intelligence. This reflects a broader strategy where AWS acknowledges its strengths in infrastructure, while Microsoft emphasizes a comprehensive, integrated ecosystem.
Cost-Effective AI Solutions
During his keynote, Garman also addressed the financial implications of developing custom AI models. He asserted that Nova Forge eliminates the prohibitive costs and engineering challenges associated with building and training an LLM from the ground up. By providing pre-trained models and various training checkpoints, AWS facilitates quicker custom model development without the need for extensive retraining, which Garman estimated could cost billions.
Businesses can choose to initiate their model building from different stages—early pre-training, mid-training, or post-training—allowing them to dictate the extent to which their domain expertise shapes the model. This flexibility is expected to attract a wide range of enterprises, especially those in sectors requiring high precision, such as healthcare and finance.
AWS plans to offer Nova Forge through a subscription model, although specific pricing details have not been disclosed. Interested enterprises are directed to an online dashboard to explore options. Currently, the service is available in the US East region in Northern Virginia, with plans for broader accessibility in the future.
The introduction of Nova Forge marks a significant step toward making AI more accessible and tailored for businesses, allowing them to harness the full potential of their proprietary data in an increasingly competitive landscape.
