Explore how Etched and CEO Gavin Uberti are rebuilding AI infrastructure from the silicon up to solve hardware scaling issues and power the future of AI models.

Whoever produces the most tokens is going to be the most valuable company in the world, and to get there, we have to move from handcrafting tokens to building a massive, industrial-scale token factory.
Teach a comprehensive lesson based on the attached interview about Etched and AI hardware. Cover: the difference between AI training and inference, why existing GPUs are outdated for modern LLMs (pre-ChatGPT design), the economics of token production, and technical bottlenecks like memory movement (KV Cache, Prefill vs Decode). Explain Etched's architecture (complete rack vs chip), low-voltage efficiency, and custom networking. Use the ELI10 style: start with first principles and intuition, use vivid analogies (Lego, cooking, pop culture), and explain the 'why' behind each engineering choice. Include startup lessons on hiring, fundraising, and solving 'unsolvable' problems. Conclude with the requested takeaways, predictions, and the summary story. Maintain a conversational, humorous, yet technically accurate tone throughout.



Etched is a company focused on rebuilding AI infrastructure from the silicon up to address the limitations of current hardware. According to CEO Gavin Uberti, many of today's chips were designed before the rise of ChatGPT and are being retrofitted for tasks they weren't originally built for. By focusing on specialized silicon design, Etched aims to create a 'token factory' that can support world-changing models at a massive scale.
Gavin Uberti is the CEO and co-founder of Etched, a company dedicated to developing next-generation AI hardware. Uberti recognized that the current hardware landscape is hitting a wall because existing chips lack the necessary infrastructure to provide advanced AI insights to everyone. His vision involves moving away from handcrafted legacy systems toward specialized transformer chips that can handle the high demands of modern AI applications efficiently.
The power of GPT-4V was demonstrated when it instantly identified a tumor from a medical scan that had previously taken doctors six months to diagnose. However, the experience also highlighted a major bottleneck: the user immediately ran out of credits. This serves as a metaphor for the current state of AI, where world-changing models exist, but the underlying infrastructure and silicon design are not yet capable of supporting them at scale.
Cree par des anciens de Columbia University a San Francisco
"Instead of endless scrolling, I just hit play on BeFreed. It saves me so much time."
"I never knew where to start with nonfiction—BeFreed’s book lists turned into podcasts gave me a clear path."
"Perfect balance between learning and entertainment. Finished ‘Thinking, Fast and Slow’ on my commute this week."
"Crazy how much I learned while walking the dog. BeFreed = small habits → big gains."
"Reading used to feel like a chore. Now it’s just part of my lifestyle."
"Feels effortless compared to reading. I’ve finished 6 books this month already."
"BeFreed turned my guilty doomscrolling into something that feels productive and inspiring."
"BeFreed turned my commute into learning time. 20-min podcasts are perfect for finishing books I never had time for."
"BeFreed replaced my podcast queue. Imagine Spotify for books — that’s it. 🙌"
"It is great for me to learn something from the book without reading it."
"The themed book list podcasts help me connect ideas across authors—like a guided audio journey."
"Makes me feel smarter every time before going to work"
Cree par des anciens de Columbia University a San Francisco
