Compute Is Cheap, Data Is Expensive

## Chapter 0 — Why AI Needs Special Hardware

## Chapter 1 — The Roofline Model

## Chapter 2 — Why Everything Becomes Matrix Multiply

## Chapter 3 — GPUs: Latency-Hiding Machines

## Chapter 4 — TPUs: Dataflow Machines

## Chapter 5 — Scaling One Chip

## Chapter 6 — Scaling Many Chips

## Chapter 7 — Precision, Sparsity, and Tradeoffs

## Appendix — References by Depth