AWS re:Invent 2025 - Tapping into the Power of Agentic AI: Driving Mission Success with NVIDIA & AWS

Summary of AWS re:Invent 2025 - Tapping into the Power of Agentic AI: Driving Mission Success with NVIDIA & AWS

Introduction to Agentic AI and the NVIDIA-AWS Partnership

• The presentation discusses the partnership between NVIDIA and AWS in the realm of generative AI and agentic AI design.
• The adoption of large language models (LLMs) and generative AI started gaining momentum in 2023, as people became more comfortable using human language to interact with highly intelligent scientific models.
• The introduction of Retrieval-Augmented Generation (RAG) and parameter-efficient fine-tuning techniques paved the way for the development of AI agents.
• Agents are described as an ecosystem of highly specialized, intelligent models orchestrated by a central model, rather than a single monolithic entity.

Key Concepts and Terminology

• NEMO (Neural Modules): NVIDIA's terminology for the modularization of the machine learning pipeline process, which allows for optimization of individual components.
• NEMO Agent Toolkit: NVIDIA's toolkit for designing and implementing agentic AI systems using NEMO modules.
• NVIDIA Inference Microservices: Optimized containers of pre-trained models (e.g., Hugging Face models) with NVIDIA driver packages and acceleration libraries.
• Blueprints: NVIDIA's term for Helm charts and the implementation of multiple containers (e.g., NEMO modules) as a cohesive solution.

Challenges in Scaling Agentic AI Systems

• As agentic AI systems are scaled, various challenges arise, including:
  • Increased complexity due to the exponential growth of tokens and reasoning processes
  • Difficulties in maintaining reliable performance, data governance, security, and authentication at scale
  • The "chasm" problem, where issues become amplified as the system is scaled, leading to potential failures.

NVIDIA's Approach to Addressing Agentic AI Challenges

• NVIDIA provides a modular, framework-agnostic approach to agentic AI design, leveraging the NEMO toolkit and blueprints.
• The NEMO toolkit includes components for data curation, customization, safety, security, evaluation, and ecosystem connectors, aiming to simplify the development and deployment of agentic AI systems.
• NVIDIA emphasizes the importance of a "best developer experience" and provides free access to their software frameworks and SDKs to enable easy integration into production pipelines.

Technical Metrics and Benefits

• NEMO-based agentic AI systems can provide:
  • 57% fewer lines of code, reducing development time and expertise required
  • Higher throughput and faster response times, enabling more efficient processing
  • 16x speed-up in data processing using GPU-accelerated NEMO Curator
  • Simplified customization and fine-tuning of models on proprietary data

Deployment and Integration with AWS

• NVIDIA's agentic AI solutions can be deployed across various AWS compute services, with a focus on Amazon EKS (Elastic Kubernetes Service) for scalable and managed Kubernetes deployments.
• Integration with AWS services, such as SageMaker and the AWS Marketplace, allows for seamless development, deployment, and access to NVIDIA's offerings.
• The presentation highlights the availability of NVIDIA's services and solutions on the AWS Marketplace, including both commercial and open-source offerings.

Conclusion and Resources

• The presentation emphasizes the importance of creating a good developer experience and providing easy-to-use tools and resources for implementing agentic AI systems.
• NVIDIA encourages attendees to explore their online resources, such as the build.nvidia.com platform, which provides a generative AI playground and technical documentation for their offerings.
• The presenters invite attendees to visit the NVIDIA booth (1022) to learn more and engage with the engineers who developed the showcased technologies.