The shift to agentic AI demands far more than just a powerful GPU. To meet the needs of AI agents, NVIDIA rethought traditional processor design with the Vera CPU. By focusing on powerful custom Olympus cores rather than maximizing core counts, NVIDIA says Vera can deliver twice the performance of a traditional x86 architecture.

NVIDIA's strategy is outlined by Dion Harris, the Senior Director of HPC and AI Hyperscale Infrastructure Solutions for NVIDIA. In our discussion, Harris speaks with Data Center Richness editor Rich Miller to trace the rapid evolution of AI, and what each shift has meant for the hardware, power, and cooling inside AI factories.

A key theme is how NVIDIA's rapid rollout of powerful new hardware is driving big changes in the data center, especially the transitions to high-density racks, liquid cooling and 800-volt DC power distribution.

Dion also shares why NVIDIA is entering the market for Land, Power and Shell for Data Center Campuses.

Here's an overview of our conversation:

00:00 Welcome and the evolution of AI: from training to reasoning to agentic AI
04:26 Vera: NVIDIA's CPU built for AI agents
07:35 Hardware cadence, CUDA compatibility, and why older GPUs keep their value
11:21 Getting data centers ready: DSX and data center-scale power management
15:49 DSX Flex: tapping stranded grid power through flexibility
20:27 Why mixture of experts made liquid cooling essential, from Blackwell to Vera Rubin
25:54 Does the rack remain the rack? Scale-up domains and the shift to the pod
29:10 Scale-up, scale-out, and scale-across: networking AI across data centers
31:55 The transition to 800 VDC power
35:15 Closing thoughts

-----------------------------------------------------------------------

At Data Center Richness we demystify technology and explain why data centers matter, and what comes next for the Internet, cloud computing and AI. Host Rich Miller is a journalist and analyst who has been covering the data center sector for 25 years.

Subscribe to our channel for more insights:
https://www.youtube.com/@DataCenterRichness?sub_confirmation=1

VISIT OUR SITE:

Data Center Richness Substack:
https://datacenterrichness.substack.com/

LINKS FROM THIS SHOW

NVIDIA Data Center Solutions
https://www.nvidia.com/en-us/data-center/

NVIDIA Developer Blog: Vera CPU
https://developer.nvidia.com/blog/nvidia-vera-cpu-boosts-ai-factory-throughput-to-accelerate-agentic-workloads/

Dion Harris on LinkedIn
https://www.linkedin.com/in/dionharris/

Rich Miller on LinkedIn
https://www.linkedin.com/in/datacenterfrontier/

RELATED VIDEOS:

Engineering the Future With NVIDIA's Most Advanced Racks
https://youtu.be/SmjqVekN93o?si=CJzewN2dT6zA-EjU

Vera Rubin NVL72 Has Arrived
https://datacenterrichness.substack.com/p/coreweave-vera-rubin-nvl72-has-arrived

All video by Data Center Richness.

Supplemental music via Epidemic Sound https://share.epidemicsound.com/60o8qx