Stenobird

Episode

How 3 CEOs Use AI to Run $10B in Companies | This Week in AI

Podcast
This Week in Startups
Published
Apr 2, 2026
Duration seconds
1810
Processing state
processed
Canonical source
https://podcasters.spotify.com/pod/show/thisweekinstartups/episodes/How-3-CEOs-Use-AI-to-Run-10B-in-Companies--This-Week-in-AI-e3hb85g
Audio
https://anchor.fm/s/7c624c84/podcast/play/117857904/https%3A%2F%2Fd3ctxlq1ktw2nl.cloudfront.net%2Fstaging%2F2026-3-2%2F421278921-44100-2-a6e50bbe23b31.mp3
JSON
/v1/public/podcasts/this-week-in-startups/episodes/how-3-ceos-use-ai-to-run-10b-in-companies-this-week-in-ai
Markdown
/podcast/this-week-in-startups/how-3-ceos-use-ai-to-run-10b-in-companies-this-week-in-ai.md

Actions

  • POST https://stenobird.com/v1/public/podcasts/this-week-in-startups/episodes/how-3-ceos-use-ai-to-run-10b-in-companies-this-week-in-ai/transcription-requests
    Idempotently request low-priority transcript generation for this episode.
  • GET https://stenobird.com/podcast/this-week-in-startups/how-3-ceos-use-ai-to-run-10b-in-companies-this-week-in-ai.md
    Read the agent-friendly Markdown representation of this episode resource.

Summary

Three CEOs discuss the physical and structural bottlenecks of the AI revolution, from copper connectivity to tabular data gaps. The discussion explores how photonics and specialized models are essential to overcoming the limits of current LLM architectures.

Topics

  • Artificial Intelligence
  • Photonics
  • LLMs
  • Data Centers
  • Enterprise Software
  • GPU Scaling
  • Machine Learning Infrastructure
  • Tabular Data

Highlights

  • Main idea: LLMs have a massive blind spot in tabular data, which represents the majority of enterprise value
  • Failure mode: Copper-based interconnects in data centers are hitting a physical wall, limiting GPU scaling
  • Practical takeaway: Moving from copper to photonics can enable thousands of GPUs to act as a single unified brain
  • Main idea: The future of video is moving from static generation to real-time, interactive, and personalized experiences
  • Strategic insight: Success in the next era of AI requires focusing on specific high-value use cases like code generation rather than broad, unfocused models

Chapters

  1. 3:10 The Tabular Data Gap: Jeremy Fraenkel explains why enterprise AI needs foundation models for rows and columns, not just text.
  2. 7:50 The Shift in Video AI: Victor Riparbelli discusses the transition from generative video to interactive, real-time personalized content.
  3. 10:00 The Power of Code Generation: A look at why Anthropic's focus on coding capabilities is currently outperforming broader competitors.
  4. 12:20 The End of Moore's Law: Nick Harris discusses the decline of traditional scaling and the need for new computing roadmaps.
  5. 14:40 Copper vs. Photonics: The physical limitations of copper cabling and how optical interconnects can revolutionize GPU clusters.
  6. 21:20 The Economics of Real-time AI: Analyzing the massive compute costs and potential of generating personalized movies on demand.
  7. 25:40 The Infrastructure Race: How big tech companies are racing to build custom chips and proprietary power infrastructure.