Fed guy
@jerome-powell.bsky.social
vibecession forever. jpow wishes this was his burner. (for clarity, I am not jerome powell)
In terms of behavior I think this is accurate for self-generated growth capex, but broader reindustrialization thru incentivizing financial investment requires lower cost of debt. Not denying importance of positive contribution margin, only magnitude
April 21, 2025 at 8:52 PM
In terms of behavior I think this is accurate for self-generated growth capex, but broader reindustrialization thru incentivizing financial investment requires lower cost of debt. Not denying importance of positive contribution margin, only magnitude
However higher interest rates raise the threshold of demand/profitability required to undertake investment
April 21, 2025 at 8:37 PM
However higher interest rates raise the threshold of demand/profitability required to undertake investment
Interesting! In fairness a variety of scenarios can play out based on what happens over the next few months / how nimble players are. Contra to my take is hyperscalers still own synergies for innovation and dry powder
January 28, 2025 at 6:35 PM
Interesting! In fairness a variety of scenarios can play out based on what happens over the next few months / how nimble players are. Contra to my take is hyperscalers still own synergies for innovation and dry powder
vs. training (library vs. detective analogy) and/or lower overall training spend. Recall 2015-2019 with rise of cloud computing, data center workload demand almost tripled, while electricity consumption remained flat. In the coming months, the 17% load growth over 5 yrs assumption will be reviewed
January 28, 2025 at 6:13 PM
vs. training (library vs. detective analogy) and/or lower overall training spend. Recall 2015-2019 with rise of cloud computing, data center workload demand almost tripled, while electricity consumption remained flat. In the coming months, the 17% load growth over 5 yrs assumption will be reviewed
In most scenarios, power infrastructure investments have the most to lose here. The goal of increasing efficiency is to decrease time/money spend, with electricity being one of the main input costs. Greater efficiency will target and reduce power demand, whether through shifts towards utilization...
January 28, 2025 at 6:09 PM
In most scenarios, power infrastructure investments have the most to lose here. The goal of increasing efficiency is to decrease time/money spend, with electricity being one of the main input costs. Greater efficiency will target and reduce power demand, whether through shifts towards utilization...
If both (1) and (2) are true, who are the real losers of the DeepSeek news? Hyperscalers are a clear no. 1, as the barriers to entry are now significantly reduced. Their moat is now capital: if efficiency can be gained across all models, $billions committed eventually translate to greater gains
January 28, 2025 at 6:05 PM
If both (1) and (2) are true, who are the real losers of the DeepSeek news? Hyperscalers are a clear no. 1, as the barriers to entry are now significantly reduced. Their moat is now capital: if efficiency can be gained across all models, $billions committed eventually translate to greater gains
(2) In contrast, lower barriers to entry and efficient model training can lead to increase in industry entrants and higher utilization (ie Jevons Paradox). If training efficiency is now such that $5m is needed to train vs $100m, which encourages 20 people to enter market, market size stays equal
January 28, 2025 at 5:59 PM
(2) In contrast, lower barriers to entry and efficient model training can lead to increase in industry entrants and higher utilization (ie Jevons Paradox). If training efficiency is now such that $5m is needed to train vs $100m, which encourages 20 people to enter market, market size stays equal
This mainly pertains to capex made by hyperscalers, but has rippled implications. Capital raised for infrastructure and power investments to service AI learning now has lower expected ROI - all else equal
January 28, 2025 at 5:55 PM
This mainly pertains to capex made by hyperscalers, but has rippled implications. Capital raised for infrastructure and power investments to service AI learning now has lower expected ROI - all else equal
Right direction, will probably be poor execution unfortunately
December 11, 2024 at 3:36 AM
Right direction, will probably be poor execution unfortunately