S2S: Introduction

Last Updated 2025-06-18

  • Inputs - A look at the raw materials and basic inputs to the semiconductor manufacturing process.

  • Fabrication - Deep dive into the fabrication process that takes raw inputs and produces integrated logic and memory circuits.

We're on the threshold of a transformation. The Bronze Age, The Industrial Age, The Information Age - we're now entering The Intelligence Age. This transition will change everything, and its already begun at scale. Understanding the mechanics and foundational constraints is essential for accurately modeling even the near future. Up until recently, I was primarily focused on the highest layers of this transformation - the state of the art models, their capabilities, the players involved. But we're now entering a phase of the transition where every layer is beginning to feel the strain. We're no longer just capital constrained or compute constrained - we're energy constrained. We're becoming geopolitically constrained. We're reaching the part of the curve where things get very interesting.

I'm writing this series to get a much better understanding of the full stack. People far smarter than me have spent lifetimes living and working in a tiny slice of this stack, and the full chain is surely the most complex process to ever exist, so I'm not going to be pursuing perfect precision or accuracy. But I do believe in having a complete picture. Step too close to a problem, and you're lost in the details. Step too far away, and you've lost the ability to resolve with clarity. Calibrating the level of analysis is crucial to achieving useful understanding.

One additional overarching objective of this series will be to understand the existing and future supply chain from the perspective of US dominance. For the sake of mankind, the US must maintain a decisive advantage in the AI race. There is no other sovereign, company, or entity that we should prefer to have control over our collective future than the US and her people. Some may vehemently disagree on this point - they're wrong.

The Plan

This collection of primers will attempt to map the process steps involved in constructing AI, from the extraction of raw materials all the way to the training and deployment of frontier models, and everything in between. More specifically, my goal is to break this task into the following sections:

  • Raw Materials & Primary Inputs: What are the most important raw materials? What will be most in demand while scaling? Which companies and countries control these inputs?

  • Semiconductor Manufacturing: Who is producing the leading edge compute? Where? What does this process look like? What are the key bottlenecks?

  • Semiconductor Design: Who designs semis? Where? What software is used?

  • Model Architectures, Training, & Deployment: What are the primary challenges that frontier labs are facing from a software standpoint?

  • Energy & Compute Architecture: How are datacenters currently constructed? Where? By whom?

Then, once these descriptive components are complete, I can begin to move onto the more interesting prescriptive sections:

  • American Dominance: Is US dominance guaranteed assuming a complete China decouple? What are the biggest roadblocks to US victory?

  • Future Paradigms & Takeoff Forces: Are there major shifts in the foundational assumptions of compute? How does compute production change near the Singularity (if at all)?

Disclaimers

  • This piece is primarily for my own edification.

  • I'm not an expert by any stretch of the imagination. I'm quite certain that I'm wrong or missing important information on any individual topic discussed. Please don't hesitate to let me know where I get something wrong!

  • While I relied heavily on LLMs for researching this work. All of the words are my own (save for text in tables/etc). Not that this warrants praise or special consideration - if anything, it should be further warning as to this series' potential bias and inaccuracy!

Terms & Abbreviations

  • Semiconductor Manufacturing - SM

  • Front End of Line - FEOL

  • Back End of Line - BEOL

  • Integrated Circuit - IC

  • Chemical Vapor Deposition - CVD

  • Physical Vapor Deposition - PVD

  • Atomic Layer Deposition (ALD)

  • Rare Earth Elements - REEs

  • Extreme UltraViolet (EUV)

  • Direct UltraViolet (DUV)

  • High NA

  • IoT, Communications, Automotive, Power, & Sensor (ICAPS) - >=28nm process nodes

  • Field Effect Transistor (FET)

  • Metal oxide semiconductor FET (MOSFET)

  • FinFET

  • Gate-all-around FET (GAAFET)