1
ChipsChapter 1 of 6
GPU
Supply-constrained and margin-rich.
TL;DR
A GPU is the accelerator that trains and serves most large AI models.
“A GPU is the accelerator that trains and serves most large AI models.”
Read full chapterLearning path6 terms · ~18 min read
Why AI is Expensive
Six terms for the angry question at the VC meeting.
2
MemoryChapter 2 of 6
HBM3e
The memory bottleneck.
TL;DR
HBM3e (High Bandwidth Memory 3e (Enhanced)) is a JEDEC HBM memory generation from 2024, with 1,180 GB/s of bandwidth per stack.
3
ConceptsChapter 3 of 6
FP16
How compute capacity gets measured.
TL;DR
FP16 is a 16-bit floating point format widely used in neural network training and inference.
“FP16 is a 16-bit floating point format widely used in neural network training and inference.”
Read full chapter4
ConceptsChapter 4 of 6
Pretraining
Why frontier training runs are so costly.
TL;DR
The phase where a model learns from internet-scale text via next-token prediction · typically 15-30T tokens and millions of GPU hours.
“Pretraining is the moat.”
Read full chapter5
PricingChapter 5 of 6
Output Tokens
Why serving costs never go to zero.
TL;DR
The tokens the model writes back to you · priced 3-5× higher than input because decoding is sequential.
“Output token economics drove every major model architecture decision since 2024. MoE, reasoning tiers, speculative decoding · all optimizing one thing.”
Read full chapter6
EconomyChapter 6 of 6
AI Capex
The spending that all this requires.
TL;DR
The billions hyperscalers and AI labs spend each year on GPUs, datacenters, and training clusters · the #1 driver of AI spending narrative.
“Nobody knows yet · but BenchGecko tracks it daily.”
Read full chapterWhat you learned
By the end you can explain AI capex, HBM bottlenecks, training and inference economics · and how to read the AI Bubble Index.
Keep learning
Next path · 7 terms
The AI Bubble Explained
Seven terms that decode whether AI is overpriced, fairly priced, or criminally underpriced. Read in order.