Compute is not oil. Chips are different and perf improves with time from both soft and new chips. Components like memory surging can instantly increase the rate of an entire gen of chips as they come due. Backwardization/Contango basically useless terms as u really don't have a a true compute curve. Go compare the cost to serve ai overviews to claude. 10GW of rubin for spacex in less than 12 months would be substantially more "compute" then all the hyperscalers and labs have today.
So, you can easily argue oversupply in compute with rental rates going up because its a fwd production capacity vs existing stock question. If all that is slated to be produced in 12 months doesn't look like it will remotely have demand based on today's trajectory, then ur oversupplied. It's that simple. U will assume less production then capacity which will mean lower prices for future supply. Today, at least to some, it seems obvious that current production capacity has jumped far ahead of the demand curve.
@insane_analyst Nearly wrote a similar tweet. As long as there is a training race, compute is locked up in training clusters
But what if...
You know people are going to think about the end even if it continues to recede into the distance
The NYTimes is REALLY going after $NVDA
It's interesting - can't recall the last time I saw a campaign like this
This is undeniably a smear campaign
You have to assume it is the Trump angle or perhaps they've really drunk the $ANT Kool-Aid
@WhullFee@dalibali2 Depends where you are - there is likely to be trade offs that can be made in different regions which help make it fungible
I imagine a lot of this is used for training now too, so it might be a bigger unlock than people expect
AGI Chronicles explains that $ANT has had a number of "success disasters," aka disastrously good success
They just had a new one: all they wanted was to close the flow of $NVDA GPUS to China... now they are about to crash spot GPU prices, where they have the most exposure 😂
AGI Chronicles explains that $ANT has had a number of "success disasters," aka disastrously good success
They just had a new one: all they wanted was to close the flow of $NVDA GPUS to China... now they are about to crash spot GPU prices, where they have the most exposure 😂
I think we see $CRWV and $NBIS even lower
I also think we see some of the tiny pure bare metal renters in the absolute toilet
Some of those latter will survive tho - so speculatively those might become interesting
Everything is still trading like a factor bet
$NVDA
"In total, Nvidia has participated in a record 129 venture-capital deals this year, as of Nov. 10, accounting for 27% of global VC deal value"
https://t.co/uaxIJP380z