saved
A datacenter is a bank
Hraness wrote this summary from a saved copy of the source. Quotations are taken word for word from the source.
gist
Joe Weisenthal shares screenshots of his Odd Lots newsletter arguing that running a data center increasingly resembles running a bank, after the podcast's interview with Hudson River Trading AI head Iain Dunning. He also draws on Sebastian Mallaby's DeepMind book to suggest AI may find market signals without needing a human-intelligible story.
ideas
- Power, not chips, binds. For HRT, securing GPUs is not the hard part; finding a place to plug them in with electricity is.
- Counterparty and concentration risk. Data centers vet clients' creditworthiness, carry heavy leverage, and must match debt duration to client contracts, so future data center supervisors might police client and industry concentration like bank supervisors after SVB.
- Factorless quant. Just as DeepMind's Go bot improved when trained only on self-play, AI trading may find patterns without the human stories behind factors like value or momentum.
- An arms race for compute. Markets differ from board games because opponents adapt, and Dunning insists more GPUs confer an advantage, which makes universal basic compute sound less fanciful.
quotes
“The real binding constraint is finding a place to put them, where they can get electricity.”
“The first is that a data center is a bank.”
“So one can imagine a future, in which we have data center supervisors”
“Iain was adamant that more GPUs confer an advantage in the market.”