Episode

S12 E9: Mitesh Agrawal, Positron

Podcast
Code Story: Insights from Startup Tech Leaders
Published
Mar 10, 2026
Duration seconds
2079
Processing state
processed
Canonical source
https://codestory.co/podcast/e9-mitesh-agrawal-positron/
Audio
https://pdst.fm/e/pscrb.fm/rss/p/audio4.redcircle.com/episodes/67420aa5-5f7e-4eef-b88e-be37899ca2e3/stream.mp3
JSON
/v1/public/podcasts/code-story/episodes/s12-e9-mitesh-agrawal-positron
Markdown
/podcast/code-story/s12-e9-mitesh-agrawal-positron.md

Actions

  • POST https://stenobird.com/v1/public/podcasts/code-story/episodes/s12-e9-mitesh-agrawal-positron/transcription-requests
    Idempotently request low-priority transcript generation for this episode.
  • GET https://stenobird.com/podcast/code-story/s12-e9-mitesh-agrawal-positron.md
    Read the agent-friendly Markdown representation of this episode resource.

Summary

Mitesh Agrawal explains how Positron is tackling the memory capacity bottleneck in AI inference by developing a new silicon architecture. The discussion covers the transition from supercomputing at Lambda to building purpose-built hardware for massive AI models.

Topics

  • AI Inference
  • Silicon Design
  • Hardware Architecture
  • Machine Learning Infrastructure
  • Semiconductor Engineering
  • Startup Scaling
  • Memory Capacity
  • Deep Tech

Highlights

  • Main idea: AI model growth is creating a critical memory capacity bottleneck during the inference stage
  • Technical challenge: Moving beyond traditional SRAM architectures to solve memory-on-chip limitations using DRAM
  • Practical takeaway: Building a high-efficiency team (under 20 people for Gen 1) is a viable strategy for complex hardware startups
  • Failure mode: Avoiding the trap of over-solving for scale before establishing real-world product value and usage
  • Philosophical lesson: Entrepreneurial longevity requires finding deep passion in the work to endure the inevitable low points

Chapters

  1. 1:00 The Vision for AI Hardware: Mitesh discusses the potential impact of AI technology and the mission of the Positron team.
  2. 7:20 The Inference Bottleneck: Identifying the thesis that increasing model sizes necessitate new approaches to vector-matrix multiplication and memory.
  3. 10:20 Architectural Innovations: Exploring the trade-offs between SRAM and DRAM and how Positron is innovating on memory technology.
  4. 16:50 Scaling Engineering Teams: How to maintain high capital and people efficiency while driving high-speed silicon design.
  5. 20:10 The Strategy of Ambition: Navigating the risks of R&D in the silicon industry and the importance of aiming for massive growth.
  6. 26:20 Economic Scale in Hardware: Contrasting the challenges of servicing billion-dollar purchase orders versus small-scale deployments.
  7. 38:50 Advice for Entrepreneurs: The importance of passion, work ethic, and pursuing large-scale ambitions in the tech industry.