Dylan Patel — Deep dive on the 3 big bottlenecks to scaling AI compute
Source
Dwarkesh Patel
Author
Dwarkesh Patel
Date
Terms in this piece · Glossary
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
Why it matters
Explains the real supply-side constraints (logic, HBM memory, power, ASML/TSMC allocation) that determine GPU availability and cost — useful context windowThe maximum amount of text a model can consider at once — its working memory for the current conversation or task.Full definition → for anyone planning compute budgets or timing infrastructure decisions for AI-native products.
Key quotes
“You sort of have this dynamic that is quite interesting in that an H100 is worth more today than it was three years ago.”
Dylan Patel
“Again, this is the most complicated machine that humans make, period, at any sort of volume.”
Dylan Patel
“You go down the supply chain, everyone’s doing X - 1. In some cases, they’re doing X ÷ 2, because they’re not AGI-pilled.”
Dylan Patel
“But in reality, he’s screwed the pooch compared to OpenAI, whose approach was, “Let’s just sign these crazy fucking deals.””
Dylan Patel
“It’s fast timelines, the US wins; long timelines, China wins.”
Dwarkesh Patel
Chapters
00:00:00Why an H100 is worth more today than 3 years ago
00:24:52Nvidia secured TSMC allocation early; Google is getting squeezed
00:34:34ASML will be the #1 constraint for AI compute scaling by 2030
00:55:47Can't we just use TSMC's older fabs?
01:05:37When will China outscale the West in semis?
01:16:01The enormous incoming memory crunch
01:42:34Scaling power in the US will not be a problem
01:54:44Space GPUs aren't happening this decade
02:14:07Why aren't more hedge funds making the AGI trade?
