peaks_and_valleys

CPA turned full time investor. Begrudging AI realist

[peaks_and_valleys @and_valleys]

@JeffBeckMadDog Do you think the units expiring are currently under market? In this market with new leasing spreads pretty high for most retail reits couldn't that be a good thing?

[peaks_and_valleys @and_valleys]

@DanielHerrold Thank you!

[peaks_and_valleys @and_valleys]

@DanielHerrold Reit

[peaks_and_valleys @and_valleys]

@realEstateTrent Real estate is also a bet on locations not just rates. The delta in returns between someone who bought a commercial building is nowhere Texas vs South Florida in the 80s is also astronomical.

[peaks_and_valleys @and_valleys]

@ylecun @ZhugeEX Furthermore a us person is able to borrow against this future income which reduces net worth but increases purchasing power and standard of living

[peaks_and_valleys @and_valleys]

@ylecun @ZhugeEX This is highly misleading. The median us persons income is substantially higher than the median Frenchman. The discounted present value of these incomes are not included in these statistics thus dramatically understating us wealth

[peaks_and_valleys @and_valleys]

@MarkAllenMulti Most deals purchased in 2020 were ok. By mid 2021 and later you could underwrite super low exit cap rates and the deals still wouldn’t have worked unless you assumed 5+ percent rent growth in perpetuity

[peaks_and_valleys @and_valleys]

@atelicinvest In a rational, non manipulated market the storage levels are not kept near zero and are refilled

[peaks_and_valleys @and_valleys]

@atelicinvest Essentially companies and speculators with storage capabilities would be bc melting against refiners/consumers for the last few barrels of oils driving prices higher until the price is high enough to cause enough demand destruction to rebalance the market.

[peaks_and_valleys @and_valleys]

@augustagarei @BeardyBrandon @robbiehendricks That’s fair. A lot of hateful and jealous people out there.

[peaks_and_valleys @and_valleys]

@realEstateTrent I think people misunderstand each other because they don't realize that people can have widely different values. So they go into conversations with a set of assumptions about basic values the other person holds that simply aren't true. All good points

[peaks_and_valleys @and_valleys]

@atelicinvest @dividendgrowth1 Not everyone is monetizing it equally at the end consumer level. Anthropic has been best at doing this so far so any infrastructure they are using has a good ROI. Google on the other-hand processes tons of tokens but the benefit is more indirect. Ditto meta

[peaks_and_valleys @and_valleys]

@orrdavid The winners are clearly companies that have installed capacity already including the hyperscalers. But not all installed capacity can benefit from increasing gpu rental rates. For example coreweave mostly has fixed price contracts. Just another variable to consider

[peaks_and_valleys @and_valleys]

@orrdavid I definitely agree that in a bottlenecked world the hardware companies are not the way to play things. I was just arguing it’s not a given in that world that all downstream ai companies benefit. In particular the model companies lose imo

[peaks_and_valleys @and_valleys]

@SouthernValue95 My biggest pushback would be that if companies are really spending that much per knowledge worker you would think that the total number of knowledge workers needed would go down pretty dramatically as workers get more efficient and/or have their job automated

[peaks_and_valleys @and_valleys]

@orrdavid This can have second order impacts if the world perceives model progress to be slowing down, ie the cost of capital may go up because people stop believing in a high likelihood of agi being achieved

[peaks_and_valleys @and_valleys]

@orrdavid If you believe that scaling laws still apply and the only way to make the models better is for more compute(still good evidence of this; top models now use way more compute per prompt than before) then bottlenecks will slow down progress.

[peaks_and_valleys @and_valleys]

@orrdavid So inference pricing power will go up but labs will have to make uncomfortable choice between serving inferior models that use less compute at scale or better models that limit number of total customers that can be served.