Nvidia GTC: Future Trends in the Global Tech Industry?

Key Takeaway

Nvidia’s CEO Jensen Huang emphasizes that data is crucial for AI-driven industries. The company is developing new enterprise infrastructure for deploying agentic AI across hybrid data centers. Their Blackwell architecture, designed for accelerated computing and Gen AI applications, features multiple GPU models, including the GB200 NVL72 with 72 Blackwell GPUs. This architecture offers up to 4x faster training and 30x faster inference on large language models while maintaining energy efficiency. Additionally, Nvidia’s innovative chip-on-wafer-on-substrate technology allows two GPU dies to operate as a single unit. The upcoming Vera Rubin Computing System is expected to launch in late 2026.

Published: 2025-03-20 20:22:36

“Data serves as the essential resource driving industries in the AI era,” states Jensen Huang, Founder and CEO of Nvidia.

“Together with the world’s leading storage companies, we are creating a new category of enterprise infrastructure that organizations require to implement and scale agentic AI across hybrid data centers.”

Blackwell architecture: For Gen AI applications

The Blackwell architecture is Nvidia’s inaugural AI chip designed specifically for accelerated computing and Gen AI applications.

This platform includes various GPU models, such as the GB200 NVL72, which integrates 72 Blackwell GPUs within a single system.

“AI has experienced a tremendous advancement — reasoning and agentic AI require exponentially greater computing performance,” remarks Jensen.

The new architecture provides substantial performance enhancements over its predecessor, the Hopper platform.

Nvidia asserts that Blackwell can achieve up to 4x faster training and 30x faster inference on large language models while consuming the same energy.

Engineers have also combined two GPU dies using Nvidia’s chip-on-wafer-on-substrate technology, allowing the chips to operate as a single GPU even though they are physically two interconnected chips.