Thursday, January 30, 2025

Thursday Night Links

  • It’s important to note that most entrepreneurs must bootstrap and run businesses for a cash profit. Only a tiny sliver of the most privileged closest to the money printers, e.g., the Harvard and Stanford networks, can tap into venture capital willing to lose money in hopes of flipping an IPO on public investors, so it’s extremely rich for this most privileged caste — pun intended — to then demand serf labor on top of their superior access to undisciplined capital. On a cash basis, your local dentist who owns his practice is a more successful entrepreneur than most of these people. [The Tom File]
  • As they note, airline equities have destroyed capital in real terms since the industry was first deregulated: every airline chases economies of scale, and that means the industry tends towards overcapacity. (A feature they don't note in the piece, but that also affects these growth incentives, is that newer planes are cheaper and seniority-based pay means that newer hires are cheaper, too, so a small but growing airline has better unit economics than a mature one, but only as long as it can outgrow the actuarial headwind.) One of the reasons airline returns over time have been poor is that they sometimes go to zero, and being a winning trade some years is just the flipside of the same volatility that can wipe out shareholders. The biggest airlines today are run more conservatively than they were in the past, and that flattens both tails of the distribution, but the fundamental challenges of the business remain. [The Diff]
  • In the long run, model commoditization and cheaper inference — which DeepSeek has also demonstrated — is great for Big Tech. A world where Microsoft gets to provide inference to its customers for a fraction of the cost means that Microsoft has to spend less on data centers and GPUs, or, just as likely, sees dramatically higher usage given that inference is so much cheaper. Another big winner is Amazon: AWS has by-and-large failed to make their own quality model, but that doesn’t matter if there are very high quality open source models that they can serve at far lower costs than expected. [Ben Thompson]
  • Quite honestly, the tax rates didn’t really matter because when an internet company worked, it grew so fast and got so valuable that if you worked another three years, say, you’d make another 10 X. Another 5 percent higher tax rate washed out in the numbers. [Marc Andreessen]
  • Plumbing:  Plumbing is a critical service that should benefit from older houses with older pipes.  These can range from simple calls related clogged drains to more complex issues such as backed up sewers.  Plumbing services is severely fragmented, leaving industry-wide data hard to come by, but we can glean some insight from Chemed, which owns Roto-Rooter, the nation’s largest plumbing company.  Since 2005, Roto Rooter’s revenues have compounded at about 6% per annum, a rate above nominal GDP over the same period.  For reasons we have already listed, this rate of growth for plumbing services seems set to continue. [Lawrence Hamtil]
  • The utilities situation might be the most interesting. A utility dropping 20%+ in one day on a read-through in demand is something that may well have last happened almost a century ago. Credit problems, lawsuits, supply constraints, regulatory issues—utilities do have the occasional big swing, but it's almost never directly attributable to a drop in future demand. One reason for the drop is that the buyers pushing a select group of utilities higher were either utility tourists who found a low-vol way to get AI exposure without the large-cap growth factor weighting, *or* were AI tourists who found a way to chase the market's favorite theme within their asset class constraints. Neither group is going to do an especially great job of rapidly updating their estimates based on news like this. But if there's a section of the economy that best fits the now very well-trod Jevons Paradox argument, i.e. that declines in AI's cost will increase dollars spent on AI because we all use it so much more, that beneficiary should be the companies whose watt-hours turn into revenue by way of inference. [The Diff]

No comments: