Russel do you follow Gary Marcus (as a qualified counter balance to some of the more egrarious AI hype). Asking as I've always found his scaling arguments persuasive and the GPT-5 launch does nothing (imho) to weaken them. I mention this because I think the mounting evidence that throwing ever more raw compute at LLMs will hit a performance wall is exactly what could unravel the capex spend. FYI, I'm a CompSci guy in real life so have a reasonable grasp of the mechanics behind LLMs & transformer architectures.
Asking this mainly because you mentioned a capex taper would strengthen the treasury bull case. We haven't seen that yet but maybe it's coming (the costs openai have spent on GPT-5 will have been enormous, hard to see them doing it all again for GPT-6 if there are increasing fears over diminishing returns from extra compute)
Yes - more evidence seems to be building that there is a declining return on investment to ChatGPT... but it is seen as so strategic, does capex get switched, not cut? I dont know.
I thought DeepSeek would strengthen the argument to take a pause on capex... but the opposite occurred. I think it more psychological than financial.. todays winners are the true believers of the last cycle.
All these discussions miss the big picture of debt - I guess it can be repeated ad nauseam, but if your debt-to-GDP hits levels that are already overdrawn in US debt-to-GDP levels, the picture changes from all-in treasuries to "don't touch them with a 100-foot-pole".
The picture is as murky because no one the hell knows what Trump is going to do next to poop on your party.
Is he going to announce a 2000% tariff on potatoes? Is he reversing stance on his new 15% contribution from chipmakers who sell to China? (And why is this not weighing heavily on those overprices stocks - he's gonna take 15% of their margin as a "tribute" while his administration (rightfully) stated security implications for high-end inference chip sales to China. Now, with 15% slapped into the margin - making theim either more expensive to China or the seller lose margin - no security issues anymore?
How does one buyer trust the US administration to buy their debt-fueled crack-up boom that's likely the outcome of these delusional roundhouse kicks to everyone's face?
Why anyone bought their debt at real negative rates was already a mistery - if the long curve does not go towards 8% as a risk premium (minimum), who else will buy but a weaponized Fed?
It's looking like the 70s - 80s all over again - and for all the wrong reasons, too.
Russel do you follow Gary Marcus (as a qualified counter balance to some of the more egrarious AI hype). Asking as I've always found his scaling arguments persuasive and the GPT-5 launch does nothing (imho) to weaken them. I mention this because I think the mounting evidence that throwing ever more raw compute at LLMs will hit a performance wall is exactly what could unravel the capex spend. FYI, I'm a CompSci guy in real life so have a reasonable grasp of the mechanics behind LLMs & transformer architectures.
Adam Butler's post yesterday is a good read on this too: https://x.com/GestaltU/status/1954561703967867019
Asking this mainly because you mentioned a capex taper would strengthen the treasury bull case. We haven't seen that yet but maybe it's coming (the costs openai have spent on GPT-5 will have been enormous, hard to see them doing it all again for GPT-6 if there are increasing fears over diminishing returns from extra compute)
Yes - more evidence seems to be building that there is a declining return on investment to ChatGPT... but it is seen as so strategic, does capex get switched, not cut? I dont know.
I thought DeepSeek would strengthen the argument to take a pause on capex... but the opposite occurred. I think it more psychological than financial.. todays winners are the true believers of the last cycle.
All these discussions miss the big picture of debt - I guess it can be repeated ad nauseam, but if your debt-to-GDP hits levels that are already overdrawn in US debt-to-GDP levels, the picture changes from all-in treasuries to "don't touch them with a 100-foot-pole".
The picture is as murky because no one the hell knows what Trump is going to do next to poop on your party.
Is he going to announce a 2000% tariff on potatoes? Is he reversing stance on his new 15% contribution from chipmakers who sell to China? (And why is this not weighing heavily on those overprices stocks - he's gonna take 15% of their margin as a "tribute" while his administration (rightfully) stated security implications for high-end inference chip sales to China. Now, with 15% slapped into the margin - making theim either more expensive to China or the seller lose margin - no security issues anymore?
How does one buyer trust the US administration to buy their debt-fueled crack-up boom that's likely the outcome of these delusional roundhouse kicks to everyone's face?
Why anyone bought their debt at real negative rates was already a mistery - if the long curve does not go towards 8% as a risk premium (minimum), who else will buy but a weaponized Fed?
It's looking like the 70s - 80s all over again - and for all the wrong reasons, too.
Trump created a false impression that tariffs balanced the budget based on the June number. Tomorrow's July deficit will show otherwise.
Combined with unfavorable CPI numbers, we'll have another leg higher in yields.
Inflation does help balance budgets... Japanese budget deficit numbers have improved a lot