atlas built a data centre
i made the original video in 2020 when it felt entirely plausible that the bureaucrats were going to win. ayn rand was supposed to be the villain. howard roark was supposed to be a warning. the individual was increasingly treated as a social problem requiring supervision, correction and management. i never bought that horse shit. neither, as it turns out, did the market.
still, samson had his hair cut. america spent the better part of four decades cutting its own. factories disappeared. skills disappeared. industry disappeared. strength was given away. a nation that once made things became obsessed with explaining why making things no longer mattered.
five years later i find myself wondering whether howard roark escaped from the pages of the fountainhead and moved to memphis. he appears to have taken the name elon musk. when colossus was first announced it looked absurd. a private individual building national compute infrastructure off grid. on any normal accounting basis it was fucking crazy. exactly the sort of overreach business schools love to mock.
except railroads looked crazy. canals looked crazy. dark fibre looked crazy. then demand arrived.
not gradually. not according to forecast.
anthropic planned for ten times growth and got eighty. eighty. that’s not growth. that’s mutation.
and suddenly the model transcontinental railroad didn’t look so stupid.
somewhere in memphis a privately owned machine is drawing enough electricity to power a city. somewhere in kentucky corning is manufacturing glass strands thinner than human hair. and somewhere in st barts, sitting at a bar, i realised the world had already changed.
human hair. that’s a funny phrase.
samson’s hair was cut. now kentucky is making hair.
i know that’s childish. i know it’s ridiculous. but if that corning chart doesn’t give you a hard-on then i’m not sure we’re even in the same business.
for years we were told industry didn’t matter. until ai needed power. until ai needed fibre. until ai needed turbines. until physics showed up and demanded a seat at the table.
that’s the part most people miss. colossus didn’t just solve anthropic’s constraint. it solved spacex’s story. it solved anthropic’s story. it probably solved openai’s story too. one asset. three problems solved.
back in february i wrote a paper sitting in a bar. the argument was embarrassingly simple. find the constraint. own the constraint. hold the constraint while the market tells you you’re wrong. then watch.
that’s the trade. that’s always been the trade.
everyone wants to talk about models. everyone wants to talk about intelligence. everyone wants to talk about software. fine. i’m interested in the glass. i’m interested in the turbines. i’m interested in the electricity.
tokens are the commodity.
the rest is commentary.
hugh.
