Summary

My guests today are Gavin Uberti and Rob Wachen, the founders of Etched. A few years ago, when they set out to build a better AI chip than the largest companies in the world, almost everyone I called told me it could not be done. They have since done it, taping out a working chip on their first attempt and becoming the first hardware company founded after…

Key takeaways

• Etched achieved a 40-day chip-to-inference deployment (vs. a competitor's 10 months) by parallelizing all non-chip work—software, rack, cold plates, production line—before silicon arrived, a strategy they call 'prefetching'
• Their two core architectural bets—low voltage inference (under half the voltage of any competing AI chip) and cluster-scale memory with 5x lower chip-to-chip latency—compound to deliver roughly an order of magnitude more user concurrency per megawatt versus Blackwell-class systems
• Etched's talent model pairs 'Legends' (e.g., Brian Leyler, who built Nvidia's HGX/DGX rack systems) with 'chips-on-shoulders' young engineers via project-based recruiting that maps the hardest unsolved technical problems to the single best person in the world who solved them

Why this matters

As inference scales to billions of concurrent users and long-horizon AI agents, tokens-per-watt and chip-to-chip latency—not peak FLOPS—become the binding constraints on AI product economics, making purpose-built inference silicon a foundational infrastructure bet for every enterprise deploying AI at scale.

Entities

Related reports

• Gavin Baker - Watts and Wafers

  Invest Like the Best with Patrick O'Shaughnessy

• Kareem Amin - The Unusual Approach to Company Building

  Invest Like the Best with Patrick O'Shaughnessy

• Alex Sacerdote - How to Invest Through Technology Cycles

  Invest Like the Best with Patrick O'Shaughnessy

Show notes

My guests today are Gavin Uberti and Rob Wachen, the founders of Etched. A few years ago, when they set out to build a better AI chip than the largest companies in the world, almost everyone I called told me it could not be done. They have since done it, taping out a working chip on their first attempt and becoming the first hardware company founded after ChatGPT to do so. They already have more than a billion dollars of customer demand for their first product, and have raised eight hundred mill

Themes

• openai
• inference
• investing