AWS Podcast #749: re:Invent 2025 - Swami Sivasubramanian Keynote
29 snips
Dec 4, 2025 Catch up on the latest innovations as experts discuss AWS AI Factories and the power of Tranium 3 Ultra servers for generative AI. Discover the all-in-one Nova 2 Omni model that seamlessly handles text, images, video, and speech. Learn about SageMaker's new HyperPod features for efficient training and inference. Updates on Amazon Connect reveal enhanced agent tools and analytics. Finally, hear about S3's impressive 50 TB object support and the new EC2 instances designed for optimized performance.
AI Snips
Chapters
Transcript
Episode notes
Bring AWS AI Infrastructure On-Prem
- AWS AI Factories deliver enterprise-grade AI infrastructure inside customer data centers to accelerate model access and deployment. Les Hewbeth explains this avoids negotiating separate contracts and provides dedicated, isolated environments with Bedrock and SageMaker access.
Tranium 3 Boosts Large-Scale Training
- Tranium 3 is a 3nm AWS AI chip delivering much higher FP8 compute, memory, and bandwidth for generative AI training. Jillian Forde highlights scale to hundreds of thousands of chips via Ultra Servers and clusters for massive training jobs.
Fine-Tune Models With New Managed Tools
- Use reinforcement fine-tuning in Bedrock and serverless RL customization in SageMaker AI to produce more accurate, cost-effective models. The hosts explain these features make customization faster and cheaper than using base models alone.