What We’re Reading (Week Ending 19 October 2025) - 19 Oct 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 19 October 2025):
1. Why AI Is Not a Bubble* – Derek Thompson
The dot-com bubble was genuinely insane. The internet companies didn’t have real revenue, and the telecom firms didn’t have real users. In its first fiscal year, Pets.com earned less than $700,000 in revenue and spent nearly $12 million on advertising. Telecom firms laid so much fiber-optic cable that as late as 2005, 85 percent of broadband capacity in the U.S. was going unused.
Today’s AI boom is nothing like that. The modern hyperscalers are among the most profitable enterprises ever. The eight biggest tech firms—the Magnificent Seven plus Broadcom—now account for 37 percent of the S&P 500 and are expected to grow profits by 21 percent this year. Return on equity for the S&P 500, around 18 percent, is the highest since at least 1991—and achieved with less leverage than the late 1990s. These are not fragile companies playing with borrowed money…
…So, you could boil down the whole “is AI a bubble?” thing to one simple question: Where’s the cash?
The answer is that three years ago, it was nowhere, and now it’s surging. According to Azhar, generative AI revenue has grown by ninefold in the last two years…
…What we’re really interested in is revenue that comes in from new businesses and customers. This comes from three sources.
- The first source is internal: Hyperscalers using their own AI to make money from their existing businesses, such as Meta using its AI to sell $1 billion more in ads (or, just as good for cash flow, using AI to save $1 billion).
- The second source is external: Companies like OpenAI and Microsoft getting money from companies that use their AI, such as a legal AI firm building a bespoke model off of GPT-5.
- The third source is novel: Using AI to create a business that doesn’t yet exist, like Tesla is trying (and mostly failing) to do with its fleet of Optimus robots.
The first two revenue sources seem to be firing on all cylinders. Microsoft and Amazon’s cloud divisions are surging as enterprise customers integrate generative-AI tools. Meta is using AI to sell more ads and to cut costs. Internal AI use (Meta’s ad tools, Microsoft’s Copilot, Amazon’s logistics optimization) plus external adoption (customers building on GPT-5 or Claude) means two of Azhar’s three revenue streams are already working.
The most important test for whether AI is a bubble is what you could call the “Triple-Digit Test.” The TDT says: If AI revenue grows more than 100 percent annually (or, even better, 200 percent) for the next few years, there probably won’t be a huge bubble pop. So far, that’s happening—or, at least, the biggest AI investors claim that it is happening.
- Microsoft says its AI business has surpassed a $13 billion annual run rate, up 175 percent year-over-year.
- Amazon claims its AI revenue is growing at “triple-digit percentages” year over year.
- The VC firm Menlo Ventures estimates OpenAI, Anthropic, Scale AI, and Perplexity are all doubling or tripling annual revenue, which means they’re growing at 100 percent or more…
…AI’s revenue problem is really a white-collar workforce creativity problem. It’s notable that AI’s profitability doesn’t just depend on how fast frontier labs innovate. It depends on how creatively their customers use the tools, as Smith points out. But most of the world’s companies are not prompt-engineering wizards. Researchers at Harvard and Stanford recently found that many firms are misusing AI so badly that workers spend more time fixing AI-generated “workslop” than doing their jobs. If that pattern persists, AI will increase frustration rather than productivity, and a slow adoption curve will blow up the Triple-Digit Test and make these companies vulnerable to a major correction in valuation or investment.
If everybody builds the same thing, where’s the moat? Something I still can’t figure out is how all of these companies are going to make money when they’re all building similar products. Anthropic’s Claude technology isn’t that different from OpenAI’s GPT technology, and when several companies build an interchangeable product, competition tends to drive down prices, which is great for consumers and bad for firms outlaying a trillion dollars to build that thing. Meanwhile, cheap Chinese or open-source models that are “99.8 percent as good for a tenth of the price,” as Smith writes, could commodify the industry overnight. In that world, the real winners wouldn’t be OpenAI or Anthropic but the downstream companies that use cheap AI to build their own moat to get rich.
2. Is AI eating Vertical Market Software? – Best Anchor Stocks
Much like some people already believed, AI has not “eaten” VMS software (at least not yet). Mark Leonard started the call with a powerful story that demonstrate his integrity and also the fact that some people might be running ahead of themselves with their forecasts:
In 2016, Geoff Hinton made a long-term forecast. For those of you who don’t know him, Geoff is known as the godfather of AI and is a Nobel Prize winner for his work in the field. And long-term forecasting is very difficult. I talked about this before, and I’m happy to send you some sources/information if you’d like to delve into that further.
Geoffs’s forecast in 2016 was that radiologists were going to be rapidly replaced by AI, and specifically, he said people should stop training radiologists. In the intervening nine years since he made that forecast, the number of radiologists in the US has increased from 26,000 (these are US board-certified radiologists) to 30,500 or a 17% increase. Now that’s outpaced the population growth in that period. So the number of radiologists per capita is up from 7.9 to 8.5. Now, Geoff wasn’t wrong about the applicability of AI to radiology. Where he was wrong was that the technology would replace people. Instead, it has augmented people. The quality of care delivered by radiologists has improved. And the number of practicing radiologists has increased.
So I told you this story to make two points. Firstly, you and I will never know a tiny fraction as much about AI as Geoff did. And secondly, despite his deep knowledge of AI, he was unable to predict how it would change the structure of the radiology profession…
…Management also discussed how the company leverages LLMs and how their strategy protects them from worsening unit economics. Both things are related, so let’s start first with how Constellation has structured its access to LLMs to avoid being “price-gouged:”
So we’ve essentially created our own centralized sort of platform that essentially removes the various factions that are currently going on where to a certain extent, you have to largely be within this cloud provider to have access natively to this LLM and so on and so forth. So there’s these turf wars being kind of created across the various cloud providers and whatnot.
And so with our strategy has been to really play a very neutral sort of Switzerland type role, where by centralizing things through strategic relationships, either directly with the model providers or with the platform providers and so on and so forth. We’ve managed to negotiate, I think some, some, some really aggressive deals and remove the element of these sort of factions. They’re all willing to kind of play nice with us in the sandbox. So that puts us in a very unique position where sort of technically we have access to 15,000 sort of unique models. And that’s because we’re essentially sort of coalescing sort of anything that otherwise couldn’t be or reside within other platforms. The other piece that sort of I had touched on very briefly, and Paul sort of alluded to as well, is sort of using a on prem based assets where and when possible.
So to the extent that the LLM needs to be or the AI model needs to be hyper specific or, you know, a specific trained one that it resides with a pre-existing best of breed provider, then sure, that may make sense to kind of tap into that one, but for basic, let’s say sort of translation service, summarization, service, and a myriad of other hosts of functionality and whatnot, you know, the on prem one is plenty sufficient and capable of doing it’s, you know, its own sort of thing.
This flexibility basically means that CSU will benefit from price wars across the different LLMs (which they expect will happen) and will also be able to take advantage of their on-prem infrastructure to lower costs for consumers (when able to).
3. An Interview with Gracelin Baskaran About Rare Earths – Ben Thompson and Gracelin Baskaran
GB: We have so much supply on the market right now, and that’s really coming from China, they keep overproducing and it’s actually forcing western companies out of operation. So to put this into context for you, in the last three years, lithium prices have fallen by 85%, nickel prices by 80%, and cobalt by 60%. So companies are struggling to operate at a time when we know we need a lot of these materials because the economics of it aren’t checking out.
And is this overproduction on purpose? Is it to knock out all these western companies to result in China dependence?
GB: I can tell you one thing is Chinese companies aren’t operating profitably by-and-large either, but they are willing to absorb long-term losses in order to gain a strategic monopoly on a lot of these sectors.
I’ll give you an example: Chinese companies in Indonesia five years ago were producing about 500,000 tons of nickel a year, now they’re producing over 2.5 million tons a year, and what’s happened is nickel prices have fallen so much that BHP, an Australian company, has closed their operations in Australia and Glencore, a Swiss company, has closed theirs in New Caledonia.
So that dominance, that willingness to absorb loss, has given them a dominance and the ability to weaponize minerals and cut us off…
…But I want to go back to your question about rare earths, there’s two things that are important. First of all, rare earths are not actually rare, they’re everywhere, but finding them in these large scale quantities that are again economically viable is actually much harder, number one.
Number two is you can mine rare earths in a lot of places, but we don’t actually process rare earths or we historically have not, which it means that no matter where the rare earths are mined — I mean, even this year until February, the rare earths that we mined in California still went to China for that processing phase, so that allowed them to build that dominance. But it’s a very small market. I need a ton of lithium, I need a ton of cobalt, rare earths are actually a small market…
…What is it about them that makes them so useful?
GB: They are the most powerful permanent magnet, which is actually really important. If you put a really good permanent magnet next to a fridge, you would basically pull the fridge off, so for defense technologies in particular, there’s nothing really that you can substitute at this point.
Got it. So is it just the magnetic properties or are there — for example, what’s their use in chips? I know particularly as chips have become more advanced, there’s questions about rare earths in there. Is that a magnetic thing or are there other properties as well?
GB: Rare earths are actually used in advanced semiconductors including memory chips and logic chips and actually when you look at the most recent export restrictions that China has applied, they’re actually reviewing the semiconductor end use on a case by case basis…
...Right. So what’s the trade-off there? Because you see numbers and you reference this before, I think China actually mines 60 to 70% of rare earths, but the actual processing is well over 90%. So what’s the bigger hole for us here? Is it the actual acquiring the rare earths or is it the processing/refining?
GB: Our big chokehold is processing because I can get rare earths from other places. So for example, now we are putting US government financial support, not just at mines, for example, Mountain Pass here in the United States, but we’re also providing financing to a project in Brazil that has rare earths. You can source feedstock from a variety of places, but it doesn’t actually matter when it goes back to China because then China can cut us off and we don’t have any of it. So we’ve got to build those processing capabilities here or else it’s like we never have access to them anyway.
So how does that happen? Is this an issue where basically there needs to be some combination of tariffs? Does China imports need to be blocked? There need to be a guaranteed price floor? How do you make the economics work? And you mentioned that these massive permitting issues when it comes to mines, is it better or worse when it comes to building these processing facilities?
GB: So really what we need is an all-of-the-above approach, and here’s what I mean by that. Again, it varies by commodity, the reason you need a price floor is this is when that US, the Department of Defense and MP Materials deal was signed earlier this year, which had a lot of support mechanisms. NdPr, neodymium-praseodymium oxide, which is one of our key rare earth compounds, was about $54 a kilogram. So at that price point, by 2030, there would only be eight projects outside of China that could even break even with their production costs because it was so commercially unviable. What you need in that case is you do need a price floor because what I don’t want is I don’t want my Western companies to go bankrupt or have to stop operating because prices are so low, and we’ve already seen it happen. In 2023, the United States opened its only cobalt mine and it closed it in the same year because prices had fallen so much. So we complained, we’re like, “Oh, I don’t want to lean into the Congo”, but we couldn’t keep our own mine open. We don’t want that to replicate for rare earths, so part of the story is a price floor story and the reality is you shouldn’t need a price floor forever.
What we saw after that deal, General Motors signed offtake, you saw Apple sign offtake, and already those prices have gone from $54 to about $84 or $85, the price floor is $110. So I’ve already closed what my fiscal responsibility by over 50%. As there’s more demand for a reliable supply chain and companies are now willing to pay that premium. I’m a Midwesterner, and what we saw after the rare earth export restrictions in April was that Ford actually had to stop manufacturing its Explorer model in Chicago because it couldn’t access these materials, so of course now we’re willing to pay a bit more to know that I won’t have to stop producing. Price floor is one part, but I need more than that.
So other mechanisms that become really important is I need concessional financing. Capital markets, because a lot of the risks that we’ve talked about, often tend to view this sector as too risky to lend to, but when the US government provides cheaper financing, we also see that banks are more willing to invest in that because they see it as a key de-risking mechanism.
The third thing I would add is government offtake helps because you’re not going to be able to sell everything to an American firm and so what we’ve seen this government do is say, “Okay, well we want a stockpile”. The recent budget in the US included $2 billion for a stockpile because if there is a supply chain disruption, I want to have enough to cover our national and economic security insurance, so they can backstop that by buying some of it…
…Given the massive risk that is here, let’s sketch out that risk. What happens if China actually just cut off rare earths tomorrow? What happens?
GB: Our manufacturing stops. Even in April 4th when those restrictions hit, US government officials said, “Maybe we get to June before we run out”…
…GB: I can manufacture for days, but the US at the end of the day has less than 1% of the world’s nickel, cobalt, we have about 2% of the world’s rare earths, we have less than 1% of graphite, we are not going to win this race alone no matter how we cut the cake. The question is how do we form — I mean, think about it — we used to form strategic alliances over oil, our relationship with Saudi was the defense for oil agreement that kept our economy open for a long time. The question is, “How do we work with our partners in a way that our supply chains are as close to us as possible?”, but we can’t do it alone, God didn’t give us the rocks.
Which mineral is the hardest problem to solve of these?
GB: I would say that the most complicated mineral is probably actually rare earths, and there’s a few reasons for that.
Is there one specific rare earth in particular?
GB: The United States Geological Survey just undertook its review of what is a critical mineral, and of the 55 or so minerals, samarium is ranked number one. The reason samarium is number one is when I take out a ton of rock from the ground, there’s a different percentage of every mineral in that ton, and samarium is such a small percentage of that rock that and I need more of it than that percentage is in there. So samarium is our most critical, which means that it is a high likelihood that there’s a failure of that supply chain. Niobium is right up there and rhodium is up there, and rhodium is a platinum group metal. So you pull it out with platinum, tiny percentage. So that’s what I mean by geology, I can’t will myself to have more Samarium in a ton of rock.
4. My friend became a millionaire at 17, and I got two book recommendations – Thomas Chua
“Actually… something huge happened.”
He told me slowly, almost reluctantly. His dad’s boss had given his father a red packet for Lunar New Year—a pretty standard gesture in Singapore. Bosses give employees red packets during the festive season as a bonus, usually cash.
Sometimes, though, they don’t give cash.
Sometimes they give hope.
In his dad’s case, his boss had given him a lottery ticket—the Singapore Sweep, with a top prize of over $2 million.
His dad won.
My jaw dropped. My teenage brain couldn’t comprehend that level of luck.
Over $2 million. The boss had fought to get the ticket back—or at least demanded a portion of it. I never learned exactly how it ended, but his dad quit his job not long after, so I assumed he kept everything. The family became estranged from relatives who’d expected generosity with the windfall, who’d wanted their own slice.
My friend and his siblings each received a tidy six-figure sum from their dad.
His family became millionaires overnight.
Looking back now, I realize that the Chinese New Year was probably their last normal one as a family. The last time money was just money, not a test of relationships. The last time people showed up because they wanted to, not because they wanted something…
…At seventeen, that kind of windfall looks like freedom. Looks like every door opening at once. No more worrying about tuition, about scholarships, about starting life in debt. Just pure possibility.
But freedom from what, exactly?
My friend hadn’t built anything yet. Hadn’t struggled for anything. Hadn’t earned the quiet confidence that comes from overcoming something you weren’t sure you could overcome. He hadn’t had the chance to discover what he was capable of when things got hard.
And here’s the thing about struggles: they don’t just test you. They build you.
5. National Bank of Detroit – Joe Raymond
Long story short, Buffett, Munger, and Guerin acquired control of Blue Chip Stamps in the late ’60s. The main appeal of the stock was the cheap price in relation to the large amount of deferred revenue from stamp sales. By taking control of the company, Buffett & friends could invest this “float” in securities.
Blue Chip had $89 million of stamp-related float in March 1972, $134 million of securities, and $74 million of common equities…
…Buffett needed to keep Blue Chip’s balance sheet liquid enough to handle stamp redemptions, but he knew he could do better than short-term debt instruments. Instead, he bought a group of solid companies at reasonable valuations.
Nearly two-thirds of the stock portfolio was made up of 10 banks…
…Blue Chip got out of most of these stocks within a decade. Nevertheless, I thought it would be fun to go through each of these banks and see how things played out over the long run.
It turns out this is a great way to learn about bank investing…
…This post will be dedicated to Blue Chip’s biggest position in 1972 – National Bank of Detroit – which is an interesting (and moderately successful) story…
…In March 1972, Blue Chip owned 218,380 shares of NBD (3.64% of the total outstanding) worth nearly $11 million. This equated to about 8% of the securities portfolio, 15% of the stock portfolio, and 24% of Blue Chip’s common equity.
All of this is to say this was a sizable bet.
The average price in 1971 (when Buffett was buying) was $50 per share…
…So, NBD was a dominant regional bank with a 12%+ ROE trading at a discount to book value. Loans to deposits was less than 60%, with the rest invested in conservative securities.
The 10-year track record was satisfactory…
…By the mid-80s, many states had passed reciprocity laws allowing bank holding companies from approved neighboring states to buy or merge across state lines.
Merger mania ensued; NBD did its fair share, making dozens of acquisitions from the mid-70s to mid-90s.
Despite the feverish M&A activity, results weren’t bad.
Book value per share grew from $57.24 in 1972 to $237.66 by 1995 (6.4% CAGR). The company also paid substantial and growing dividends over this period.
Annual BVPS growth adjusting for dividends came in around 11-12%…
…In 1995, NBD completed an all-stock merger of equals with First Chicago Corporation. The two banks had complementary business lines in adjacent geographies. The surviving entity operated under the combined name First Chicago NBD.
Then in 1998 First Chicago NBD merged with Banc One – a Columbus, Ohio based bank. Every one share of FCNBD received 1.62 shares of Banc One and the combined company was renamed Bank One (with a “k” instead of a “c”)…
…But Bank One’s fortunes started to turn south in the late ’90s shortly after the merger.
Earnings fell sharply in 1999 as growth slowed and anticipated cost savings failed to materialize. The credit card division from Banc One imploded due to bad loans and regulatory scrutiny. The stock fell by 50%. Analysts described Bank One as “the sick man of big banking.”
In 2000, a young executive by the name of Jamie Dimon was brought in to right the ship.
And right the ship he did.
Dimon wrote off billions in bad loans and goodwill. He centralized operations and established new risk controls. Tech systems were updated and unified. The credit card business was rebuilt.
By 2003, Bank One stock had tripled from its 2000 low.
In 2004, JPMorgan Chase and Bank One decided to merge…
…Each share of Bank One received 1.32 shares of JPM. Dimon was made President and COO for a year before taking the CEO title in 2005 and Chairman in 2006…
…Every one share of National Bank of Detroit Buffett purchased in 1971, if he had held for the next 54 years, would have turned into 14.58 shares of JPMorgan Chase today (as a result of multiple stock splits and stock-for-stock mergers).
NBD traded for an average price of $50 per share in 1971 whereas JPM trades for $310 per share today. As such, every $1,000 invested in NBD 54 years ago would be worth a little over $100,000 today.
Buffett’s $11 million stake would have grown to more than $1.1 billion.
Astute readers will note that this “only” equates to a 9% annual return. The buy-and-hold investor would have also received growing dividends over the decades, pushing the total annual return into the low-teens.
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 Amazon, Meta Platforms, and Microsoft. Holdings are subject to change at any time.
