korrents

← People and their mental models

Dylan Patel's mental models

3 claims Dylan Patel made fit 3 mental models. Most often: Bottlenecks, Cheaper means more, not less, Speed is a feature. Everything they said here.

Models they name

Their own words name the idea.

Cheaper means more, not less

Make something cheaper to use and people use so much more of it that the total goes up (Jevons paradox).

Others who think with this model

David Heinemeier HanssonDoug O'LaughlinLinus BlomqvistZeke Hausfather

Open the model page

Cheaper models raise the demand for compute rather than lowering it: H100 rental prices went up in the weeks after DeepSeek launched.

  1. Dylan Patel Founder and chief executive of SemiAnalysis But the funniest thing I think that comes out of this is Jevons paradox is true. AWS pricing for H100s has gone up over the last couple of weeks, since a little bit after Christmas, since V3 was launched, AWS H100 pricing has gone up. DeepSeek, China, OpenAI, NVIDIA, xAI, TSMC, Stargate, and AI Megaclusters | Lex Fridman Podcast #459youtube.com · 3 Feb 2025 · 3:17:11 into the videoAll korrents from this video
    Their wordsBut the funniest thing I think that comes out of this is Jevons paradox is true. AWS pricing for H100s has gone up over the last couple of weeks, since a little bit after Christmas, since V3 was launched, AWS H100 pricing has gone up.

    Where this was said

    Watch from 3:17:11 plays here

    ↗DeepSeek, China, OpenAI, NVIDIA, xAI, TSMC, Stargate, and AI Megaclusters | Lex Fridman Podcast #459youtube.com

    3 Feb 2025 · video · 5h 06m · spoken · official transcript

Speed is a feature

Fast software feels simple, and waiting makes everything feel complicated.

Others who think with this model

Craig ModGuillermo RauchCasey MuratoriDavid Heinemeier HanssonJohn Maeda

Open the model page

Nobody wants a slow model, which is why labs will not trade inference speed for cheaper memory even though they easily could.

  1. Dylan Patel Founder and chief executive of SemiAnalysis They could release claw slow mode and have an increase in tokens per dollar by a significant amount. Um they could probably like reduce the price of Opus 46 by you know 4x 5x and reduce the speed by another by maybe just like 2x like the curve on inference throughput versus speed is there already just on hm um and yet they don't um because no one actually wants to use a slow model Dylan Patel — The single biggest bottleneck to scaling AI computeyoutube.com · 13 Mar 2026 · 1:17:38 into the videoAll korrents from this video
    Their wordsThey could release claw slow mode and have an increase in tokens per dollar by a significant amount. Um they could probably like reduce the price of Opus 46 by you know 4x 5x and reduce the speed by another by maybe just like 2x like the curve on inference throughput versus speed is there already just on hm um and yet they don't um because no one actually wants to use a slow model

    Where this was said

    Watch from 1:17:38 plays here

    ↗Dylan Patel — The single biggest bottleneck to scaling AI computeyoutube.com

    13 Mar 2026 · video · 2h 30m · spoken · machine transcript

Models we see in what they say

Our reading: their claim applies the idea without naming it. The claim is theirs; filing it here is ours.

Bottlenecks

A system moves only as fast as its narrowest point; speed up anything else and nothing changes.

Others who think with this model

Jacob TrefethenAlexandr WangCasey HandmerDavid Heinemeier HanssonJensen HuangLip-Bu Tanand 4 more

Open the model page

The bottleneck on America's AI buildout is transmitting power, not generating it: in parts of the US, moving electricity costs more than making it.

  1. Dylan Patel Founder and chief executive of SemiAnalysis Interesting thing is certain regions of the US transmitting power cost more than actually generating it because the grid is so slow to build. And the demand for power, and the ability to build power, and re-ramping on a natural gas plant or even a coal plant is easy enough to do, but transmitting the power's really hard. DeepSeek, China, OpenAI, NVIDIA, xAI, TSMC, Stargate, and AI Megaclusters | Lex Fridman Podcast #459youtube.com · 3 Feb 2025 · 3:45:47 into the videoAll korrents from this video
    Their wordsInteresting thing is certain regions of the US transmitting power cost more than actually generating it because the grid is so slow to build. And the demand for power, and the ability to build power, and re-ramping on a natural gas plant or even a coal plant is easy enough to do, but transmitting the power's really hard.

    Where this was said

    Watch from 3:45:47 plays here

    ↗DeepSeek, China, OpenAI, NVIDIA, xAI, TSMC, Stargate, and AI Megaclusters | Lex Fridman Podcast #459youtube.com

    3 Feb 2025 · video · 5h 06m · spoken · official transcript