The Two Harvard Dropouts Who raised $800M to take on NVIDIA

Summary

Etched co-founders Gavin Uberti and Robert Wachen discuss building a full rack-scale AI inference system designed to massively increase token throughput and concurrency. They detail their technical bets on low-voltage inference and cluster-scale memory, the near-death fundraising struggles, and a contrarian talent philosophy. The conversation highlights why inference will become the world's largest market, the importance of TSMC's partnership, and the inevitability of token production economies of scale.

• Etched is building an entire rack solution, not just a chip, to optimize AI inference.
• Their key technical edges are low-voltage inference for thermal efficiency and cluster-scale memory for low-latency, high-bandwidth chip interconnects.
• The company overcame a capital crunch by raising an unusual $100M Series A through sheer persistence.
• TSMC's exceptional customer service and partnership are cited as the real driver of its continued dominance.
• Inference demand will grow to serve billions of concurrent agents, making token production the most valuable industry.
• The founders emphasize speed, vertical integration, parallelization, and a 'legends plus raw talent' recruiting model.
• Wall-clock time for long-horizon tasks will collapse, accelerating scientific breakthroughs and software development.