AWS Summit 2026 breakout session on building frontier AI models with Amazon Nova Forge, focused on how organizations can go beyond off-the-shelf fine-tuning to create fully custom foundation models with their own data.

The talk walked through the phases of modern model development — pre-training, mid-training, and post-training — and showed how Nova Forge opens up each stage to customers through checkpoint access, proprietary data blending, and custom reinforcement learning with your own reward functions and environments.

We also covered how push-button recipes on SageMaker HyperPod reduce setup time from weeks to minutes, how Amazon’s curated training data can be combined with domain-specific corpora for measurable benchmark improvements, and how built-in responsible AI tooling lets teams implement customizable safety guardrails suited to their use case.