← Multimodal Sources 🕐 2 min read
Multimodal Sources

Nvidia CTO Michael Kagan: Scaling Beyond Moore's Law to Million-GPU Clusters

> The requirements for performance started to grow up at a much higher coefficient.

Show: Training Data · Publisher: Sequoia Capital · Host: Sonya Huang, Pat Grady

Episode URL: https://pscrb.fm/rss/p/traffic.megaphone.fm/CPUAI5059499822.mp3

Publish date: 2025-10-28
Duration: NAs
Default source credibility: HIGH — Sequoia partners interview frontier-lab founders + F500 AI buyers. VC-hosted — portfolio-company framing on recommendations; named guest metrics stay HIGH. Peer-tier to No Priors in quality.

  • NVIDIA’s success is tied to expanding the market rather than competing with others, emphasizing a win-win culture.
  • The Mellanox acquisition enabled NVIDIA to scale GPU performance beyond single chips to large clusters, crucial for AI workloads.
  • AI workloads demand exponential performance growth, necessitating innovations in networking and software to manage large-scale GPU clusters.

Extracted quotes

# Credibility Speaker Org Timestamp Topic Quote
1 HIGH Michael Kagan (CTO) NVIDIA 03:45 01-ai-native-landscape The requirements for performance started to grow up at a much higher coefficient. The models started to go up in terms of size and capacity, 2x every three months, which requires now 10x or 16x a year performance growth versus old school of 2x twice every other year.
2 HIGH Michael Kagan (CTO) NVIDIA 14:21 01-ai-native-landscape We are the fastest growing Ethernet business, you know, let alone NVLink and InfiniBand. But just Ethernet business is the fastest growing business ever.
3 HIGH Michael Kagan (CTO) NVIDIA 38:17 01-ai-native-landscape The slope is somewhere in the range of 10x or a few orders of magnitude a year. And that’s what we are doing, by the way, now. Since about two or three years ago, we accelerated our product introduction from every other year to every year.

Per-quote detail

1. Michael Kagan — NVIDIA (03:45)

The requirements for performance started to grow up at a much higher coefficient. The models started to go up in terms of size and capacity, 2x every three months, which requires now 10x or 16x a year performance growth versus old school of 2x twice every other year.

  • Stat: Models grow 2x every three months, requiring 10x-16x annual performance growth.
  • Credibility: HIGH — Specific metric with denominator and unscripted interview.
  • Topic tag: 01-ai-native-landscape

2. Michael Kagan — NVIDIA (14:21)

We are the fastest growing Ethernet business, you know, let alone NVLink and InfiniBand. But just Ethernet business is the fastest growing business ever.

  • Stat: NVIDIA’s Ethernet business is the fastest growing.
  • Credibility: HIGH — Specific claim about growth and unscripted interview.
  • Topic tag: 01-ai-native-landscape

3. Michael Kagan — NVIDIA (38:17)

The slope is somewhere in the range of 10x or a few orders of magnitude a year. And that’s what we are doing, by the way, now. Since about two or three years ago, we accelerated our product introduction from every other year to every year.

  • Stat: NVIDIA’s product performance grows 10x or more annually.
  • Credibility: HIGH — Specific metric with unscripted interview.
  • Topic tag: 01-ai-native-landscape

Extracted 2026-04-14T18:12:20 via scripts/podcast_mine.py (MLX mlx-community/Qwen2.5-32B-Instruct-4bit).