Compute Shortages Stall LLM Token Growth As Hyperscalers Decline

agents are still extremely useful. however, if you just do the business math on a sector exponential growth, it has to come from exponential growth in demand and supply. both demand (token/hyperscaler spend) and supply (compute/DC) now have ceilings attached to them

i am not yet short, but im focused on scaling out of speculative longs and being more selective. adding more SAAS and biotech.

i am usually early. trends take a few quarters to filter down to earnings, so realistically the music won't stop till December ish esp after midterms.

the software companies which go up a lot are *currently private*. they grew v fast w highly accessible capital. the companies which go down a lot are big companies that are already public from 2022. now software will grow. not most current software stocks.

+a Democrat win is likely medium term bearish for AI. Democrats have been openly opposed to datacenter construction. cutting the possible ceiling of DC buildouts in half should lead to predictable consequences. Long term Dems are better on immigration, but rn the trade is capex.

ive spoken of the "inequality trade": companies with much high productivity gains from AI will devour ones that don't+spend more. imo this is still the case, but creative destruction requires several years, while mcap weighed indexes bleed down

one last point: you cant have an entire AI hardware supply chain all be bottlenecks for extended periods long term. smth has to give: bottleneck thesis kaput -> price rerates down, costs passed to users (reducing demand), profits cannibalised etc