What We’re Reading (Week Ending 31 August 2025) - 31 Aug 2025
Reading helps us learn about the world and it is a really important aspect of investing. The late Charlie Munger even went so far as to say that “I don’t think you can get to be a really good investor over a broad range without doing a massive amount of reading.” We (the co-founders of Compounder Fund) read widely across a range of topics, including investing, business, technology, and the world in general. We want to regularly share the best articles we’ve come across recently. Here they are (for the week ending 31 August 2025):
1. Monetary policy is not about interest rates, it’s about the money supply – Steve H. Hanke and John Greenwood
The ongoing feud between President Trump and Fed Chairman Jerome Powell centers on interest rates. This tells us more about the near-universal view of what constitutes monetary policy than it does about Trump or Powell. While Trump and Powell might quibble over the proper level for the Fed funds rate, they both think monetary policy is all about interest rates…
…Why the obsession over interest rates? One reason hinges on the fact that for over the past 30 years or so, macroeconomic models are neo-Keynesian extensions of dynamic stochastic general equilibrium (DSGE) models. These put interest rates front and center…
…But that’s not what monetarists, who embrace the quantity theory of money, tell us. Unlike the neo-Keynesian macroeconomic models that exclude money, the quantity theory of money states that national income or nominal GDP is primarily determined by the movements of broad money, not by changes in interest rates…
…First, let’s consider the case of Japan between 1996 and 2019. Throughout this period, the Bank of Japan’s (BOJ) overnight policy rate lingered at negligible levels, averaging 0.125%. As a result, most economists concluded that monetary policy in Japan was very “easy”. But monetarists, who focused on Japan’s anemic broad money (M2) growth of only 2.8% per year, concluded that monetary policy was “tight”…
…Japan’s inflation averaged a de minimis 0.2% per year in the 1996-2019 period. It is clear that the monetarists were correct…
…Let’s consider the U.S. between 2010 and 2019. During most of this decade, the Fed funds rate was held down at 0.25%. In addition, the Fed engaged in three episodes of quantitative easing (QE). Many concluded that this amounted to very “easy” monetary conditions. They warned that inflation would result. In fact, broad money growth (M2) remained low and stable at 5.8% per year. In consequence, inflation also remained low, averaging just 1.8% per year between 2010 and 2019. As was the case with Japan, interest rates turned out to be a highly misleading indicator of the stance of monetary policy. The growth in the money supply was a much better guide to economic activity and inflation than the course of the Fed funds rate…
…The reason why central bank policy rates are a misguided mechanism for steering and forecasting the course of the economy is because interest rates are, in large part, symptoms of past money growth, not necessarily drivers of future money growth. Changes in the quantity of money, on the other hand, directly fuel spending, and therefore correctly signal the direction of spending and inflation…
…By ignoring the quantity theory of money and employing neo-Keynesian macroeconomic models, central bankers are often wrong-footed. They think that by managing policy rates, they are controlling monetary policy when in reality, they are just reacting to changes in the quantity of money that occurred in a prior period.
2. Global Crossing Is Reborn… – Praetorian Capital
Let’s start with total datacenter spend for 2025. Insiders think it’s going to clock in at around $400 billion…
…What’s a datacenter made of?? There are three main components; the building and land at roughly a quarter of the cost, all the power systems, wiring, cooling, racking, etc. at about 40% of the cost, and then the GPUs themselves at about 35% of the cost. I am sure I’m off by a few percent in these categories, but I’m relying on AI and we all know it’s still imperfect. I’m assuming that the building depreciates over 30 years, the chips are obsolete in 3 to 5 years, and then the other stuff lasts about 10 years on average. Call it a 10-year depreciation curve on average for an AI datacenter. Which leads you to the first shocking revelation; the AI datacenters to be built in 2025 will suffer $40 billion of annual depreciation, while generating somewhere between $15 and $20 billion of revenue. The depreciation is literally twice what the revenue is…
…With nothing to go on, I’m going to take an optimistic guess here, and say that ultimately, the margins get to positive, and then gradually creep up towards 25%. Why 25%?? I have no idea. It just sounds right because electricity is really expensive and you need a lot of expensive tech nerds to manage the equipment. Honestly, no one really knows where gross margins eventually land, so let’s just run with it, so that we can do some simple math…
…By my math, you need $160 billion of revenue at that 25% gross margin, which gives you $40 billion of gross margin against $40 billion of depreciation. Now, remember, revenue today is running at $15 to $20 billion. You need revenue to grow roughly ten-fold, just to cover the depreciation. Except, no one does anything to break even in business. For a new technology like this, with huge obsolescence risk, what unlevered ROIC would you demand?? Would you want a 20% ROIC?? That’s still dilutive to the ROIC for most of the largest capex spenders. Even at that dilutive ROIC, you’d need $480 billion of AI revenue to hit your target return…
…$480 billion is a LOT of revenue for guys like me who don’t even pay a monthly fee today for the product. To put this into perspective, Netflix had $39 billion in revenue in 2024 on roughly 300 million subscribers, or less than 10% of the required revenue, yet having rather fully tapped out the TAM of users who will pay a subscription for a product like this. Microsoft Office 365 got to $ 95 billion in commercial and consumer spending in 2024, and then even Microsoft ran out of people to sell the product to. $480 billion is just an astronomical number…
…While we all remember Pets.Com and the hundreds of other Dot Com startups that flamed away, it was companies like Global Crossing, spending tens of billions on fiber, that facilitated all of this. That fiber, amazingly, is still in use. Global Crossing went bankrupt along the way, as did many of its peers. They overestimated what people would pay for this fiber, not that it would eventually be used or valuable.
Today, I watch in awe (stupefaction really), as companies continue to throw endless resources at AI, I remember back to the Dot Com bubble and Global Crossing—fiber was the datacenter of that cycle, and Corning was the NVIDIA of its day (it lost 97% of its share price in the two years after it peaked).
3. Bitcoin TreasuryCos & The Roaring 20s – Be Water
The Bitcoin Treasury craze is either genius or madness—and very possibly some combination of both…
…This is not the first time leveraged financial vehicles promised to democratize access to scarce assets using leverage and the accretive magic of mNAV premiums: the 1920s investment trust and holding bubble followed a similar script in the run-up to the 1929 Crash…
…During the Roaring Twenties common stocks occupied a cultural position remarkably similar to Bitcoin (and arguably the S&P) today—they were viewed as the revolutionary investment of their era, and there was widespread belief that supply of stocks was too scarce to meet surging demand.
In the 1920s, mutual funds were introduced under the name “investment trusts,” and—like Bitcoin treasury companies—formed to capitalize on this scarcity. A major difference between modern mutual funds and these trusts was that the trusts were leveraged: like Bitcoin treasuries, they invested using borrowed money that was considered “safe” because—like MicroStrategy—they issued preferreds and long-term debt securities to the public to buy portfolios of stocks. Galbraith:
The most notable piece of speculative architecture of the late twenties, and the one by which, more than any other device, the public demand for common stocks was satisfied, was the investment trust. The investment trust did not promote new enterprises or enlarge old ones. It merely arranged that people could own stock in old companies through the medium of new ones…
…Like Bitcoin Treasuries, the 1920s trusts had the added appeal of mNAV premiums that seemed to offer something for nothing.
Just as Bitcoin treasury companies today boast of their mNAV and ‘bitcoin yield,’ a key feature of the 1920s bubble was the tendency for investment trusts to trade at significant premiums to mNAV during their heyday. Galbraith:
The measure of this respect for financial genius was the relation of the market value of the outstanding securities of the investment trusts to the value of the securities they owned.
Normally, the securities of the trust were worth considerably more than the property it owned—sometimes even twice as much. There should be no ambiguity on this point: the only property of the investment trust was the common and preferred stocks, debentures, mortgages, bonds, and cash that it held. (Often, it had neither an office nor office furniture; the sponsoring firm ran the investment trust out of its own quarters.)
Yet, had these securities all been sold on the market, the proceeds would invariably have been less—and often much less—than the current value of the outstanding securities of the investment company. The latter, obviously, had some claim to value that went well beyond the assets behind them…
…As with today’s Bitcoin TreasuryCos, this persistent mNAV premium created a powerful financial engine for both the trusts and the underlying stocks they were buying: the ability to conduct immediately accretive share issuances. When a trust trades at a premium to its underlying stock values, it can issue new units at the inflated market price and instantly increase the NAV for its existing shareholders.
This reflexive accretion mechanism created a self-reinforcing feedback loop similar to today’s “Bitcoin Leverage Loop”. The cycle worked as follows:
- Investor optimism drove a trust’s price to an mNAV premium.
- The trust would issue new units at this premium price, which was immediately accretive to the NAV per share.
- The new capital raised was used to purchase more stocks, adding buying pressure to the overall market and increasing the value of the trust’s own portfolio.
- The rising NAV and apparent success of the strategy further fueled investor optimism, widening the premium and allowing the cycle to repeat.
- Meanwhile, investors in the trusts and individual stocks amplified their exposure to a sure thing by using margin loans to leverage their positions, adding extra “juice” to the trade and further driving up NAVs and mNAVs for the trusts…
…Goldman Sachs Trading Corporation (GSTC) was perhaps the proto-MicroStrategy of the day. Launched by the influential Goldman Sachs partner Waddill Catchings in December 1928, it was, at its inception, the largest investment trust yet established—boasting an initial capitalization of $100 million. Its units, offered to the public at $104, was immediately oversubscribed and quickly soared in value, doubling to $226 within a short period and trading at a massive premium to the underlying value of its stock holdings…
…In Brad DeLong and Andrei Shleifer’s The Stock Market Bubble of 1929: Evidence from Closed-end Mutual Funds, they noted:
If [investment trust mNAV premia] indeed reflect excessive investor optimism rather than skill at management, there will be a tendency for funds to pyramid on top of one another. If each fund can be sold for 50 percent more than its own net asset value, promoters can more than double their profits by establishing a fund that owns funds that hold stocks, rather than just establishing funds that hold stocks…
This prediction is confirmed by one of the largest funds: the Goldman Sachs Trading Corporation. This was a closed-end fund organized in December 1928 with a net asset value of around $100 million. In 1929, one of its largest holdings was the Shenandoah Corporation, another closed-end fund organized by Goldman Sachs. Another large holding was in its own stock.
Nor is this all. In the same year, Shenandoah organized a new closed-end fund called the Blue Ridge Corporation and became a large investor in its stock. All these funds traded at premia; at the top of the pyramid, the Goldman Sachs Trading Corporation traded at a premium to a premium to a premium to net asset value…
…If history serves as any guide, we can expect Bitcoin treasury companies to begin investing in other Bitcoin treasury companies before this cycle concludes.
4. Whatever Happened to the Self Driving Semi? – Chris Paxton
There are almost three million semi trucks in the United States alone, to the point that trucker is the most common job in 29 states. Most of these are driving 400-600 miles per day along long, straight, predictable highways — a use case that, at a glance, seem perfect for autonomy.
And yet, on-road autonomy looks guaranteed to start not with semis but with taxis, operating over much shorter distances in much less of the United States…
…Fully-loaded trucks are massive, with a legally-mandated maximum of 80,000 lbs. This makes everything a truck does notably less responsive. Planning becomes more difficult; learning methods are less effective, too, when there’s not a clear, immediate mapping between input and output.
If we want to discuss how serious a problem this is, we should look at stopping distance; i.e. how long it takes a semi truck to come to a complete stop because, say, there was an accident on the road ahead of it.
Stopping distance for a fully-loaded semi truck traveling at 65 mph is approximately 525 feet to about 600 feet. Even though most US highways have higher speed limits, trucking companies usually limit speed to 65 mph for safety and fuel efficiency reasons; it seems reasonable to expect that autonomous truckers would do the same. But note that this is under ideal conditions; stopping distances can as much as double on icy roads.
Now, a good long-ranged lidar could have 1000 feet of range. Aurora has a particularly good in-house lidar, with about 450 meters (~1500 feet) of range – much farther than many other options. But maximum range isn’t effective range, which is far more important. This is hard to estimate — it varies depending on conditions, on objects, and of course on the quality of the particular classifiers being used to interpret objects. This quantity is notably shorter than the maximum range on practically any sensor, by as much as about half; and we’ll also need to classify if this was a spurious detection (a plastic bag blowing onto the road, a cardboard box) or a serious issue.
And that’s setting aside other concerns: what if there’s a patch of black ice ahead on the road? The lidar can’t detect this at all, and it’s a huge issue for highway driving. There was a famously horrific 133-car pileup in Fort Worth, Texas in 2021, caused by black ice, which led to 65 injuries and six fatalities.
5. SITALWeek #459 – Brad Slingerlend
Investing is a form of storytelling. CEOs spin tales about their companies and try to rally the workforce to manifest them over a long time horizon. Investors decide if they too believe the stories or not. Most of the time, the stories are fiction, fantasy, or even fairy tales. Occasionally, visionary entrepreneurs pen a nonfiction, or even a compelling fiction that turns out to be so predictive of the future that it serves as prior art for reshaping reality (think of the Steve Jobs Reality Distortion Field!). There are also stories about economics, politics, and the world at large that influence the stories about companies and investments. Investors create their own stories about businesses as well, and the resulting investment ideas can end up in either a canonized history book or a throwaway dime novel. Even trying to unravel the truth of past stories can be fraught, as hindsight is only as good as the incomplete and unreliable human narratives on which history is based…
…Today, it’s not clear how much, if any, impact investors’ stories have on the daily prices of stocks. And, in some cases, it appears to me companies are losing complete control of their own narratives as well…
…And, now, we have something very different happening: all of that volume in the market, previously programmed in some form or another by humans guiding machine learning algorithms (or retail investor brains programmed by social media news cycles, etc.), is slowly being taken over by LLMs and agentic AI. I suspect autonomous AI trader bots are writing their own signal algorithms and creating their own stories. They are telling those stories to each other and executing trades. We can see clues that this shift is happening in a recent study that found meaningful drops in trading activity during ChatGPT outages. I think that tidbit of information gives us, well, the rest of the story as to what will soon define the stock market on a day-to-day basis (if it’s not already the dominant force, which I suspect it is). This agentic investing evolution will create even more noise and less signal in the daily price of any given stock. Again, this turn of events spells good news for us active investors who still think we can find stories that, with any luck, will turn out to be superior nonfictional investments.
Disclaimer: None of the information or analysis presented is intended to form the basis for any offer or recommendation. We currently have a vested interest in Microsoft and Netflix. Holdings are subject to change at any time.
