AWS doubles down on custom LLMs with features meant to simplify model creation

Right on the heels of announcing Nova Forge, a service for training custom Nova AI models, Amazon Web Services (AWS) has introduced additional tools to help enterprise customers build their own frontier models.

AWS announced new capabilities for Amazon Bedrock and Amazon SageMaker AI at its AWS re: Invent conference on Wednesday. These features are designed to make it easier for developers to build and fine-tune custom large language models (LLMs.

The cloud provider is rolling out serverless model customisation in SageMaker, allowing developers to start building a model without planning for compute resources or infrastructure, according to Ankur Mehrotra, general manager of AI platforms at AWS, in an interview with TechCrunch.

To access these serverless capabilities, developers can choose between a self-guided point-and-click workflow and an agent-led experience, in which they prompt SageMaker with natural language. The agent-led tool is launching in preview.

“If you’re a healthcare customer and you wanted a model to understand certain medical terminology better, you can simply point SageMaker AI to your labeled data, select the technique, and SageMaker will fine-tune the model,” Mehrotra said.

This capability supports customisation for Amazon’s Nova models as well as select open-source models whose weights are publicly available, including DeepSeek and Meta’s Llama.

AWS is also launching Reinforcement Fine-Tuning in Bedrock, allowing developers to pick a reward function or a preset workflow, after which Bedrock automatically runs a complete model customisation pipeline.

Frontier LLMs — the most advanced AI systems — and custom model development are clearly a significant theme for AWS at this year’s conference.

AWS previously announced Nova Forge during AWS CEO Matt Garman’s keynote on Tuesday. The service allows AWS to build custom Nova models for enterprise customers for $100,000 per year.

“Customers ask, ‘If my competitors have access to the same base model, how do I differentiate myself?’” Mehrotra explained. “The solution is being able to build customized models that reflect your brand, your data, and your use cases.”

So far, AWS has struggled to gain meaningful traction for its AI models. A Menlo Ventures survey from July showed that enterprises strongly prefer Anthropic, OpenAI, and Gemini over other options. However, AWS believes that deep customisation capabilities could become a long-term competitive advantage.

Sponsored: Watch AWS re: Invent 2025 live

Check out the latest reveals on everything from agentic AI and cloud infrastructure to security and much more from the flagship Amazon Web Services event in Las Vegas. This video is brought to you in partnership with AWS.