"The value of a business is the present value of the cash it will generate between now and Judgment Day, discounted at the appropriate interest rate."

- Warren Buffett

Every year coal sets a new record.

In 2025 the world burned 8.8 billion tonnes of it. An all time high. Because of the events in the Middle East, 2026 will set another all time record. While the financial press wrote eulogies, the black rock kept showing up to work.

I want to tell you about a company that takes this hated, dying, politically toxic rock and does two things with it that should not be possible. It sells it for enormous profit to a world that cannot get enough of it. And it turns it into parts for fighter jets.

This is Core Natural Resources. Ticker CNR.

Two legends got married

Core is not one company. It is two.

In January 2025 two coal giants merged. CONSOL was the king of cheap thermal coal in Northern Appalachia, with one secret weapon, a marine export terminal in the Port of Baltimore. Arch owned the one of best metallurgical coal mines in America. The kind of coal that is geologically rare and impossible to replace.

Neither one alone had the full picture. Together they do. Core digs up Arch’s premium steelmaking coal and ships it to the world through CONSOL’s port. One business balances the other.

And the people running it are not amateurs. When firms merge they always promise cost savings that never appear. Core did the opposite. They promised 110 million in synergies. Then raised it. Then raised it again. It now sits at 150 to 170 million a year. Management delivered more than it said. Remember that. It tells you who you are betting on.

A coal miner that builds aircraft

Before the coal, let me break the stereotype completely.

Core has a division called Innovations. It takes coal mined in West Virginia and turns it into a special carbon material. That material is used to make the tools and molds needed to build military aircraft and spacecraft. This is not a small experiment. The business serves more than 40 customers. Its products have been used in the F-35 fighter jet, the B-21 bomber, and NASA programs.

Sit with that. The market sees a dying coal miner. Some of that same coal is ending up in fighter jets. Nobody is paying a cent for it.

Subscribe now

Part One. The thermal coal nobody admits they need

The thesis nobody in polite company will say out loud. Demand for thermal coal is not falling. It is rising. And it will keep rising for two decades.

There are 8.3 billion people on this planet. Roughly 750 million have no electricity at all. Billions more have power that cuts out every day. They do not care about your ESG report. They want lights, refrigerators, a fan when it is 45 degrees outside.

By 2050 the world must electrify another 2 billion people, because the world’s population is going to grow. That is building the entire power grid of China and Europe from scratch in 25 years. You do not do that with solar panels.

China burned more coal in 2024 than the rest of the world combined, and has over 200 gigawatts of new coal capacity under construction. India has 1.4 billion people and uses a fraction of the power per head that the West does. Neither has committed to a phase out. None.

Now add a new buyer. Data centers. AI is devouring electricity faster than anyone predicted, and it needs power at three when the sun is not shining. In much of the world that means coal.

That is CONSOL’s world. The crown jewel is the Pennsylvania Mining Complex, three connected mines holding 529 million tons, around 20 years of life. The coal runs near 13,000 BTU, hotter than most American coal, and Core owns 94 percent of it outright instead of leasing. That is an edge by itself.

Why does CONSOL win? Lowest cost plus the port. In commodities the cheapest producer always wins, because when prices fall the expensive mines die first while the cheap one keeps printing cash. And the Baltimore terminal is the only major East Coast export terminal served by two competing railroads, so the railroads cannot squeeze them. The Baltimore terminal throws off around 60 million a year, and Core also owns 35 percent of a second export terminal, DTA, for even more export capacity.

When the home market is weak they export. When export is weak they sell at home. They can sell wherever the better market is.

Part Two. Arch and the rarest rock in the world

Now the second thesis.

Metallurgical coal is not thermal coal. It is a different animal. You do not burn it for power. You use it to make steel. There is no way to make steel in a blast furnace without it.

And here is the key. Unlike thermal coal, which sits under half the planet, high quality metallurgical coal exists in only a handful of places on Earth. The United States and Australia dominate the premium grades. It is geologically scarce. And there is no economic substitute that can scale. The supply of the good stuff is fixed by geology.

That combination, scarce supply and no substitute and steady demand, is the holy grail of commodity investing. Prices spike violently every few years when supply gets tight. So what does Arch bring to the table? One of the best metallurgical coal portfolios in America.

The trophy is the Leer Complex in West Virginia. Three pieces. Leer, with 29 million tons of reserves. Leer South, with 57 million tons. And an undeveloped expansion called Leer West, with another 84 million tons sitting in the ground waiting. Together that is more than 30 years of life. Leer and Leer South produce High-Vol A coal. This is the special stuff, the binding agent that lets steelmakers mix cheaper coals together. It sells at a premium for exactly that reason.

Then there are the supporting mines. Beckley, a premium Low-Vol mine with about 23 million tons. Itmann, another premium Low-Vol operation with 27 million tons. And Mountain Laurel with about 16 million tons.

Let us sell that coal

Read more

"In particular, if you have a corporation where a man has risen to be CEO, and he now has hundreds of millions of dollars in the stock of the company... And then he has his directors vote him a great stock option grant annually to preserve his loyalty and his enthusiasm to the company, I think that's demented. And I also think it's immoral."

- Charlie Munger

I made a mistake.

Cardlytics was my investment before Cannibal Stocks existed. The ticker is CDLX. Most people reading this have never heard of it, and that is the first interesting thing about this company. You have probably used it. You just did not know.

Here is the business in four sentences. You open your banking app and see an offer, twenty percent back at a restaurant you already visit. You tap it, you pay with your card, the cashback lands automatically. Behind that tap sits Cardlytics, the wire connecting the advertiser to your bank, and the bank already knows what you buy. They do not advertise themselves. They have no logo you would recognize. They are already built into the payment systems of some of the largest banks in America, which means millions of people use this company every week and have no idea it exists.

When I bought it, here was my logic, and I want to be honest about every piece of it.

I thought the moat was enormous. Banks will not hand customer transaction data to anyone. Google tried. Facebook tried. Both got told no. Cardlytics got in because the data never leaves the bank. A competitor starting today would have to go bank by bank, contract by contract, for years. That felt unbreakable to me.

The stock was cheap. All-time low cheap. I was buying the whole company for under one times revenue. The math in my head was simple and seductive. The cost base is fixed, somewhere around 250 to 300 million a year, whether revenue is 250 million or a billion. So the moment they become meaningfully profitable, and they are already around breakeven, the market hands a small-cap a high multiple overnight, and you are sitting on a brutal return. That was the entire thesis. And I was buying while everyone else was running for the exits. I cloned the idea from Cliff Sosin, and I was paying less than he did. What could go wrong.

Here is what I got wrong.

It turns out the banks can do this themselves. I do not regret that part. If you do your homework, you think hard, you read everything, and you still get the moat wrong, that happens. It happens to the best investors alive. I am young. I will misjudge a moat again. That is the cost of doing this work, and I am at peace with it.

While I was sitting in this position, watching the stock grind lower, I noticed something else. Something that had nothing to do with banks or moats or competition. Every single quarter, no matter how badly the company performed, no matter how far the stock fell, management got rewarded. More shares. More grants. Again. And again. And again.

So I asked people. Smarter people than me, more experienced people, people who have been doing this longer. I asked them, is this normal? Is this how it is supposed to work? Every one of them told me the same thing. Yes. That is standard. That is just how the industry does it. Stock-based compensation, equity grants, it is normal, do not worry about it.

Bullshit.

Give a gift subscription

What I Found Underneath

Quick thing first, because the whole post rests on it. A company is a pie. Your shares are your slices. When the company prints new shares for someone else, your slices do not disappear, but the pie now has more slices in it. Your piece got smaller. That is dilution. Nobody reaches into your pocket. You just own less of the same company than you did yesterday.

Now look. In 2018 Cardlytics had about 22 million shares. Today it has about 58 million. They more than doubled the slices. And while they kept printing, the stock fell from about 150 dollars to under one. Price down 99 percent. Slices up 156 percent.

What does that do to you? If you owned one percent in 2018, you own about 0.39 percent today. You sold nothing. Your piece got cut by more than half while you watched.

So where did all these shares go?

A big chunk went to pay the people running the company. From 2018 to 2024, Cardlytics handed out about 251 million dollars worth of stock as pay. The entire company today is worth about 50 million. They paid out five times the value of the whole business in stock. Not with money they earned. With your ownership.

Then the deals. In 2021 they bought two companies for about 856 million combined. Within three years they admitted those companies were nearly worthless and wrote off more than 655 million of it. Hundreds of millions gone, and more shares printed to pay for it.

But the part that got under my skin was the accounting.

Cardlytics reports a number called Adjusted EBITDA. In 2024 the real result was a 189 million dollar loss. Their homemade number was a 2.5 million profit.

How do you flip a 189 million dollar loss into a profit? You take the bad stuff out. Mostly the 40 million in stock they paid themselves. They erase the cost of their own pay, then call the business profitable. Their excuse? Stock is “non-cash.” No money left the building, so it does not count.

That is the trick. If you would feel richer getting salary plus stock than salary alone, then the stock is pay. And pay is a cost. You do not get to call it free just because no cash moved. Somebody paid. You paid, through dilution. Munger had a word for this kind of accounting. Corrupt.

Then the detail that ended my doubt. On January 12, 2026, with the stock around 90 cents, after the collapse, after everything, the company handed a new executive one million fresh shares. They printed more.

Why would anyone be comfortable doing that? Buffett answered it in 1995.

“We have had conversations with managers where they tell us how fortunate they feel because the stock is down and they can issue options cheaper. If they were issuing those to the third parties, I’m not sure whether they’d have exactly the same attitude.”

There it is. A low price is a disaster for you. For a manager paying himself in shares, it is just a cheaper way in. The lower it falls, the more shares it takes to hit the same payday, and the more of your company quietly changes hands. Munger said it plainer, about a manager who thinks printing shares costs nothing:

“Imagine hiring a manager who thinks that way and paying them money to behave like Judas in your very midst.”

I got the moat wrong. I can live with that. But the moat could never have taught me this. The price tells you what the market thinks of the company.

The share count tells you what management thinks of you.

Share Cannibal Stocks

The Silence of the Lambs

There is a reason for the title.

The shareholders of Cardlytics did what shareholders are trained to do. They held. They trusted. They waited for the next quarter. And every quarter, more shares were printed and handed to the people at the top. They did not scream. They stood quietly in line while their piece got smaller, and smaller, and smaller.

That is what most of us are. Lambs. And the silence is mistaken for consent.

I was one of them.

Not anymore.

Today I put on the mask.

Cheers, Sandro

Get 20% off forever

Note: After years of dilution, the final act was a reverse split. On June 3, 2026, Cardlytics announced a 1-for-10 reverse split, which will begin trading on a split-adjusted basis on June 8. The main reason is to push the share price higher and remain compliant with Nasdaq listing requirements. It improves the stock's appearance, but it does not improve the underlying business. It simply buys more time.

A reverse split is financial cosmetics. It makes the stock price look higher, which can attract buyers who focus on the price rather than the business itself.

I won’t bombard your inbox every day.

But there are a few things happening behind the scenes that I think are important, and that you should know about. So this time I am making a small exception and writing a short update. Going forward, these smaller updates will be posted in our Substack Chat or Substack Notes instead.

So don’t hold it against me. I promise I am not trying to turn your inbox into CNBC. 🙂

The first update is that over the last few days I have been working on a new tool that analyzes SEC EDGAR filings. The Cannibal Radar.

Every trading day I want to know which companies announce a new buyback program, increase an existing authorization, extend a repurchase program, or do anything else related to share repurchases.

This will become a new source of ideas and future posts.

The Cannibal Radar has already been running for several days. There are still a few things I need to improve and automate, so don’t criticize me if everything isn’t perfect just yet. But even in its current form, it is already proving to be extremely useful and hopefully no cannibal will escape our attention.

Most of the companies showing up on the radar are either just starting their buyback journey or are increasing and extending existing repurchase programs. Those announcements often go unnoticed by the market.

Sometimes they don’t matter. Sometimes they matter a lot.

In fact, one such example appeared today.

This company is actually very familiar to me. I spent a lot of time analyzing Arch and Consol in the past, and CNR is essentially the merger of those two businesses. Because of that, I already have a decent understanding of the company, and a full write-up will probably be coming sometime next week.

But there are even more important developments taking place right now.

Let’s talk about that.

Read more

One year ago I published my first pitch. Twenty ideas later, this is the scoreboard.

Not a victory lap, and not a list of winners cherry-picked after the fact. A ranking. Every idea I put in front of you, sorted from my highest conviction to my biggest mistake — including the ones I actually own, and the ones I only traded and would never marry.

That distinction matters, so let me be clear about it up front. Some of these are companies I hold because I believe in them for the next decade. Some were special situations — Helium, Sable — that I rode hard and sold, brilliant trades built on businesses I have no interest in owning. A good return tells you nothing about which is which. The ranking does.

Share

How to read this

The tier is my conviction, not the share price. S means highest conviction. A is strong. B is solid. D is a miss. A stock can be up 100% and still sit in D, and another can be down and still sit in A — because price is what happened, and conviction is what I think happens next. Performance is measured since the first pitch. The only companies I currently hold in my portfolio are the S-Tier names.

With that out of the way, from best to worst.

S-Tier — Highest Conviction

This is where I have the least doubt, and it’s no accident that it’s dominated by two industries the market still hates. Met coal and offshore drilling. The setup in both is the same. New supply is brutally hard to bring online, the survivors have cost positions that don’t lose money, and the industry’s ugly history — bankruptcies, dilution, destroyed capital — is precisely what keeps the opportunity alive. Everyone remembers the pain. Almost nobody noticed that the supply side was permanently damaged.

Alpha Metallurgical (AMR) is one half of my met coal core. If prices stay high, it’s probably the better bet than Warrior — cheaper valuation, more aggressive buybacks, better capital allocation. It’s going to return all of its free cash flow to shareholders, and the market still treats it as garbage. If you're a cannibal and Mr. Market treats you like garbage, that's about as good as it gets. The main negative is its dependence on railroads.

Warrior Met Coal (HCC) is the other half. Better margins, a fantastic management team, and a moat most investors underestimate. If everything goes wrong, Warrior is the one that still doesn’t lose money. I genuinely can’t tell you which I prefer between the two, so I own both. The anomaly is that both will return 100% of free cash flow to shareholders. The only knock on Warrior is its preference for dividends. At these prices, buybacks would create far more value over time.

Transocean (RIG) is the offshore thesis in one name. The industry is consolidating fast, and there’s a realistic path where one player controls roughly half the world’s ultra-deepwater fleet. Once the debt comes down, I expect management to start handing cash back — and that’s when the market re-rates the whole thing. The first-year return was already excellent. I think the bigger story is still ahead. On the negative side, the Valaris merger is being financed with shareholder dilution. If management creates more value per share than it gives away, and eventually adopts a met coal-style capital return strategy, returning virtually all free cash flow to shareholders, this could turn into a brutal investment.

Valaris (VAL) is the same trade — the $35 billion fleet nobody wanted — and the market is only now waking up to what it’s worth. I already view these two companies as one position. At this point, I have very little left to add.

A-Tier — Strong

A-tier is where conviction is high but something — size, politics, a single variable — keeps it just short of the top. These companies have a high probability of eventually earning an S-Tier ranking and making their way into my portfolio.

Mongolian Mining (975.HK) has incredibly low production costs, decades of Chinese demand, a railway due in 2028, and a company that has started buying back its own stock. Premium metallurgical coal with more than a billion tonnes of reserves and resources. An asset that I believe will become one of the most sought-after in the world over time. Even India is looking for ways to access this coal, but transportation is challenging because it would have to move overland, making logistics costs significantly higher. The reason it isn’t S-tier is one word: politics. When Mongolia’s Democratic Party was in power, it tried to nearly destroy this company.

Athabasca (ATH.TO) is a genuine cannibal that returns nearly every spare dollar through buybacks, and you can already see it in the share price. A major tailwind is developing right now. Canada wants a new pipeline to the Pacific Coast. If that happens, Alberta can export more oil directly to Asia instead of relying on U.S. buyers.That would narrow the massive discount Canadian oil trades at today. Athabasca wouldn’t need to produce a single extra barrel to benefit. A higher realized price means higher cash flow. Higher cash flow means more buybacks. And the best part is that every dollar goes back to shareholders through a cannibal that almost nobody is paying attention to.

Kaspi (KSPI) isn’t a cannibal, but management acts like one — entirely in shareholders’ interest. An enormous moat, a lot of growth, and the best cloner alive. They don’t reinvent anything, they find what works elsewhere and execute it better inside a platform that already runs millions of daily lives. At around 7x earnings, the valuation makes no sense to me. If management keeps doing what it’s done for a decade, this could end up the single best idea on the list. The only problem? My wife has banned me from investing in Kazakhstan.

Apparently companies that just emerged from bankruptcy are acceptable levels of insanity. 😅

Bitcoin (BTC). I don’t treat it as an investment. I treat it as money. Viewed purely as an asset, the fixed-supply argument honestly puts it alongside met coal and offshore. It’s still down, and it’s still hated. Both of those are features, not bugs. Because here’s the thing I’m certain of: every fiat currency eventually trends toward its intrinsic value, which is zero.

I lived through it. The hyperinflation that destroyed the Yugoslav dinar in the early 1990s wasn’t a chart to me — it was daily life. If you’ve never seen what happens when money dies, watch this. Eighteen minutes will teach you more about why Bitcoin matters than any white paper.

B-Tier — Solid

These are good-to-great businesses where I either expect more modest upside, have an open question, or simply prefer a wider margin of safety before getting excited.

Whitehaven (WHC) has brilliant capital allocation and top-tier thermal coal. My hesitation is that in the great years, a large slice of free cash flow leaks out through royalties and taxes. And some of the mines are not in the best condition.

CNX Resources (CNX) has unbelievably low costs and a massive buyback the market doesn’t appreciate yet. A bigger cannibal than Alpha, even — it just hasn’t shown up in the price. I also think patience makes sense here. Once the situation in the Middle East stabilizes, natural gas prices could come down, giving us a chance to pick up shares at a lower price.

EOG Resources (EOG) is one of the best-run operators on the planet. The only problem is that it’s already a giant, which caps the upside compared to the smaller, more hated names elsewhere on this list.

Sable Offshore (SOC) was the easiest trade of the year for anyone willing to buy when it was hated, and the return shows it. Sable Offshore has a tailwind from California because the state is reducing reliable energy supply while energy demand remains high, increasing the value of local oil production. But the next chapter needs enormous capital, much of it likely from issuing shares, and that dilution is exactly why I can’t fall in love with the long-term story.

Lumine (LMN.V), Constellation Software (CSU.TO), and Topicus (TOI.V) are all fantastic businesses, and I would not be surprised if they are much larger companies ten years from now. They probably belong closer to the A-tier than the B-tier. What cooled me off is that none of them are cannibals. We are unlikely to ever see buybacks. On top of that, after speaking with people close to Topicus, I learned that growth has become a bit more challenging than I originally thought, even in areas where I believed competition was almost nonexistent. That doesn’t break the thesis, but it does change it. These are still excellent businesses. The problem is that they are not extremely cheap, so for now they go into the "too hard" pile.

AutoNation (AN) and AutoZone (AZO). I think Mr. Market is making a clear mistake pricing both this cheaply. The fleet keeps aging, cars get more expensive to replace, and lower share prices just let both retire more stock per dollar of buyback. AutoZone is the better company; AutoNation is the better investment. They’re in B only because the explosive growth is mostly behind them. AutoZone has already delivered roughly a 450x return since its IPO. That's what happens when a cannibal retires 9 out of every 10 shares outstanding. Madness.

D-Tier — Miss

The honest part of the list.

Cardlytics (CDLX) was probably the biggest mistake of my investing life — and the most valuable. The real lesson wasn’t that I misunderstood the moat. It was dilution. Good management buys stock when it’s cheap. Bad management pays itself with stock while shareholders get diluted. Buffett called it “sell low, buy high.” Cardlytics followed that playbook perfectly. Roughly $49 million of new shares were issued in 2024 while the company continued losing money. The stock fell from $157 in February 2021 to about $0.65 today, a decline of roughly 99%. Management got richer. Shareholders got poorer. Munger had a line for companies like this.

A colony of rats in a grain warehouse.

Kansai Paint (4613.T) returned a lot of cash, which I liked — until I realized most of it came from selling assets, not recurring free cash flow. That makes the long-term economics genuinely hard to judge, so it lands here by default rather than conviction.

The Helium Basket (Avanti, Helium Evolution, Pulsar) was, alongside Sable, the best trade of the year. I bought early, when nobody was looking, and sold near the top after the crowd arrived. As a trade, it was close to perfect. As businesses, these have weak economics, uncertain paths to profitability, and depend almost entirely on expectations rather than performance. Could one of them work? Sure. But they don’t deserve any more of my time — which is exactly why a great trade still ends in the D-tier.

The Rats Ate the Grain in Those Other Companies

From today, only cannibals live in the portfolio. Everything else was a ride. For a year I showed you what I look for. Now I show you what I hold.

Most companies issue shares. Mine retire them.

The rats ate the grain in those other companies. Mine eat themselves.

Out there, plenty of companies dilute you while their prices keep climbing. Be careful what you own.

Year two begins.

Cheers, Sandro

Get 30% off forever

“We are not happy when things we're buying go up in price. We want them to go down and down and down."

- Warren Buffett, 2004 Meeting

Two years ago I flew back to the country where I was born. Bosnia. Most people couldn’t find it on a map. It’s famous for “ćevapi”, the second most popular dish on earth. You can’t grill proper “ćevapi” without coal. Strange how coal keeps showing up in my life.

They make “ćevapi” here in Germany too. But restaurants here cannot use charcoal grills because of smoke, smell, ventilation rules, and German bureaucracy, so they cook them on electric grills instead. The result? The “ćevapi“ in Germany taste like shit. 🙂

Anyway.

In Bosnia I was giving a presentation about my new book. Afterwards, I met the director of the Sarajevo Stock Exchange and some great people. Then I started hunting.

What I found was a graveyard. Nobody buys stocks in Bosnia. Nobody. There are trading days with zero volume. Zero. No one in the entire country buys or sells a single share.

It wasn’t always like this. Back in 2008 the Sarajevo Stock Exchange went vertical. Every ticker on the board exploded. People took loans to buy whatever they could grab. Brokers slept in their offices. Pure mania.

Then it cracked.

And to this day the market is dead. Stocks are hated. If you want to buy something, you call your broker on the phone like it’s 1985. Corruption is high. Trading is difficult. But prices are absurd.

I found a company trading at $0.65 per share. It has just 21,182 shares outstanding, giving it a market value of less than $15,000. It is a declining business, but the assets behind the company are worth roughly $5 million. Most of those assets are tangible. Buildings, land, and investment properties.

Price per share is $0.65. Book value per share is $225. Good luck finding that on the New York Stock Exchange. I actually tried to buy the whole thing. There was just one problem. The stock had not traded since 2017. I am fairly sure most of the shareholders are dead.

That was not even the interesting part.

The interesting part was that although everything in Bosnia is dirt cheap, I found a company, a boring cement business. Cyclical. Nothing special. And yet it traded at more than 30 times earnings. The stock is up seven, maybe eight times in few years.

Why?

They started putting shareholders first. Dividends. At first they paid out 20 or 30 percent of profits. Then they kept raising. Today they pay out 100 percent. Every cent. A handful of investors saw what was happening and started fighting for shares. The most boring company on the Bosnian market became the most wanted. The market handed a cement plant a tech multiple.

This is where Alpha enters the story.

If you like to focus only on signals, here is it is. The company is returning 100% of its free cash flow to shareholders. Dividends are nice. But retiring shares is another level.

The rest of this post is the long term thesis and what I believe could realistically happen even this year.

What happens when there are no shares left to buy?

Read more

“Really big effects, lollapalooza effects, will often come only from large combinations of factors.” — Charlie Munger

Munger had a word for those rare moments that make people rich. Lollapalooza. Many forces hit at once, in the same direction, so they don’t add up. They multiply. One plus one becomes eleven.

Think of Coca Cola. It’s not one thing. Sugar, caffeine, cold, the fizz, and a brand you’ve trusted since childhood. Each force small on its own. Stacked together, they make a drink billions of people crave every day. That’s the first lollapalooza, the product itself.

But there’s another side. Charlie Munger didn’t just drink Coca Cola, he invested in it too. And the same thing happened there. For decades the stock paid dividends and bought back its own shares. Marketing built the brand, the brand built sales, sales built profit, and profit flowed back to shareholders. Forces stacking in the bottle, and forces stacking in your pocket.

A lollapalooza is five, six, seven things lining up at once, every one pushing the same way.

I don’t know what happens next. Nobody does. But I know what a setup looks like. I’m staring at one.

Alpha Metallurgical Resources. Ticker AMR. Digs metallurgical coal, the kind that makes steel. No real substitute. No skyscraper, no car, no bridge without it.

Force by force.

One. China

May 22, 2026. A gas explosion tears through the Liushenyu mine in Shanxi. Over 80 dead. China’s worst mining disaster in fifteen years.

China makes more than half the steel on Earth. More than the US, Europe, India, Japan and everyone else combined. Shanxi alone digs about a quarter of all Chinese coal.

After the blast China shut mines across the province for inspections. Roughly 125 million tonnes of annual capacity frozen overnight. To me that’s the biggest event in the met coal market this year.

On May 26 it came out that inspectors found hidden tunnels, fake walls and doors, falsified maps, untracked illegal miners, mines with no gas monitoring at all. When a government finds fraud like this, the crackdown grows. Some mines could stay shut through July and August. If tunnels or ventilation in the good mines were damaged, months or more.

And it lands right as China enters peak summer demand. China has huge reserves, but most isn’t premium hard coking coal. That quality is the bottleneck. When China runs short, it doesn’t sit there. It goes shopping on the seaborne market, the ocean trade where everyone else buys. That market is small and thin next to China’s domestic one. When the biggest steelmaker on Earth shows up in a small pool, you don’t get a ripple. You get a wave.

Two. India

It’s a Muslim holiday right now. Much of India’s steel is idle. India is the largest seaborne met coal buyer in the world, and it’s sitting out.

Normally, when your biggest customer goes quiet, prices fall.

Instead, prices are rising. Read that again. The market climbs while the number one buyer sleeps. That tells you how tight this already is.

Every year the slow season ends around late May and India steps back in June or July. Will it again? I don’t know. But if it does, it walks into a market already stretched, already rattled by Shanxi, already climbing without it. A few extra premium cargoes bought while India buys, and prices jump.

Three. Kingston Wildcat

While this unfolds, Alpha is doing something concrete. Firing up a new mine, Kingston Wildcat.

It produces Low-Vol coal. The highest quality, highest priced met coal there is. Mills pay a premium because it makes the cleanest coke. Summer 2026 futures run around 250 dollars a tonne, and those contracts are being signed right now.

About 500,000 tonnes in 2026. Full capacity from 2027, a million tonnes a year. That one mine, at today’s prices, can add up to 250 million in new annual revenue. Alpha already runs 19 others.

Prices spike exactly as Alpha adds its best coal. You can’t plan that timing.

Four. The company eats itself

This is the heart of what I write about.

Since March 2022 Alpha has aggressively bought back its own shares. About 1.2 billion dollars, roughly 7 million shares, shrinking the company by about 32 percent.

Let that land. Nearly a third of the company is gone. Every share you still hold now owns a much bigger slice of the same business, and you never lifted a finger. That’s a cannibal. A company that uses its own cash to eat itself and makes every remaining owner richer.

Now stack that on rising coal. When met coal climbs, Alpha’s profit doesn’t rise gently. It goes bananas. Costs barely move while the selling price jumps.

Five. Mongolia

The doubters say relax, China will just buy more from Mongolia next door. Cheap, overland, done.

Not that simple. Mongolia is landlocked. Almost all its coal crosses into China by truck and rail through one region, about 90 percent already going that way. The main mine runs at the limit of the current route. The new railway that would add real capacity isn’t done. Construction runs into 2028. You can’t wish a railway into existence this summer.

So Mongolia can’t flood China overnight. And the US? China puts roughly 28 percent tariffs on American coal, so that door is mostly shut. That leaves seaborne, mainly Australia and Canada, the same thin pool. The same one India is walking back into.

Every escape route leads back to the same small pool.

Six. Insiders are buying

Watch what insiders do, not what they say.

Kenneth Courtis, an Alpha director, just bought another 15,000 shares in the open market, spending roughly $2.8 million. He now owns almost one million shares.

Mohnish Pabrai, one of the most respected value investors alive, recently added another 80,000 shares. No interest in selling.

Not traders flipping for a quick gain. People who understand this company better than almost anyone, locking shares away.

Seven. Share supply is absurdly small

Look at the share count. The whole company is 12.71 million shares. Tiny. The float, the shares actually free to trade, is 10.9 million, and maybe less today since those numbers are from May 15. Institutions hold over 91 percent. Insiders nearly 25 percent. The free float is a puddle.

Now the shorts. Around 1.59 million shares sold short in mid-May, over 18 percent of the float betting it falls. Short ratio 6.4, meaning more than six full days of normal volume just to buy back what they owe.

Think about it. A puddle of shares. Seven forces pushing up. A crowd of shorts who eventually must buy back into that same puddle. When they rush the exit at once they have to buy, buying pushes price up, that forces more to buy. Fuel on a fire already lit.

The Lollapalooza Setup

Shanxi rips out supply, maybe for months. India wakes up hungry. Alpha fires its best mine into the spike. The company eats itself through buybacks. Mongolia can’t rescue China in time and US coal is tariffed out. The smartest insiders are buying and refusing to sell. All of it aimed at a company with under 12.7 million shares, a puddle of a float, and a fat pile of shorts who may have to buy it all back.

Seven forces. Every one pointing the same way.

Here’s the thing. Everybody lies in the coal market. The miners lie, the traders lie. And right now the information coming out of China is almost nothing. So don’t listen to the talk. Watch the met coal futures. Price doesn’t lie. In the coming weeks the futures will tell you the truth before anyone says a word.

One thing I’ll watch closely is whether Alpha starts pulling away from Warrior. The two usually move together, both met coal, same industry forces. If Alpha dramatically outperforms while Warrior stays calm, something is moving beneath the surface.

This is one of those moments. I watch it every day. Now you know what to watch too.

Like and restack this post, or I’ll officially appoint the AMR short sellers as your financial advisors. They seem very generous with other people’s money. 😄

Cheers, Sandro

Get 20% off forever

“Heads, I win. Tails, I don't lose much.” - Mohnish Pabrai

Two men retired in September 2025. Three days apart.

One was Mark Leonard, who built Constellation Software. The market noticed. The stock fell 17 percent in a week.

The other was Nick DeIuliis, who built CNX into one of the cleanest cannibals in America. Nobody noticed.

Get 20% off forever

Today’s trend is that the biggest companies on Earth are fighting over electricity.

Microsoft is paying to restart a nuclear reactor at Three Mile Island. A twenty year deal. Meta locked up the Clinton nuclear plant in Illinois for twenty years. Amazon went to Talen and built a structure that now runs to almost two gigawatts through the year 2042.

These are the richest companies in the world signing contracts longer than most marriages, just to be sure they have power. Artificial intelligence eats electricity.

In 2023 American data centers used about 176 terawatt hours of power. Lawrence Berkeley Lab thinks that number could more than triple by 2028. The IEA says data centers will be responsible for nearly half of all the growth in US electricity demand for the rest of this decade.

These machines run twenty four hours a day, every day. They cannot wait for the wind to blow or the sun to rise. They need power that is always on. There are only a few things that do that at scale. Nuclear is one.

The other is natural gas.

The IEA says it plainly. Gas is the most scalable always on resource the United States has, and it remains the dominant near term fuel feeding these data centers. Texas alone is tracking over 400 gigawatts of large load interconnection requests. Almost ninety percent of it is data centers.

So picture the perfect company for this moment.

A lowest cost producer of natural gas. Sitting right on top of the demand. Cheaper than everyone in the country. Profitable in every single year, good or bad. Run by management that takes the cash and uses it to swallow its own shares.

A cannibal. In a hated industry. Exactly the kind of business I build my portfolio around.

That company exists. Its name is CNX Resources.

Read more

247 workers were underground.

82 are dead. 2 are missing. 128 are in hospitals.

This is the deadliest Chinese mining disaster since 2009.

The owner is now in a police cell. Beijing has accused him of “major illegal acts.” The mine blueprints handed to the rescue teams did not match the real underground layout, so the rescuers went in blind. Within 72 hours, every mine in the county was shut. By Sunday, Beijing had ordered the entire country into emergency inspections.

The market reacted fast. Dalian coking coal futures hit their daily limit on Monday. Singapore premium hard coking coal jumped to 247 dollars per tonne. The forward curve lifted out to 2028.

My read is that roughly 6 to 8 million tonnes of premium coking coal are about to disappear from the global market over the next six weeks. The accident mine itself is gone for at least a year. Some of the rest could stay shut even longer.

That is enough to break the seaborne balance.

Let me walk through why.

Read more

“I’ve seen gluts not followed by shortages, but I’ve never seen a shortage not followed by a glut.”

- Nassim Taleb

A powerful statement from a man I deeply respect. But there are forces in the modern economy that prove he may not always be right.

Let me show you one.

In 2023 I started looking at companies that make DRAM memory chips. Historically this industry had been a graveyard for investors. Brutal boom and bust cycles. New producers entered easily. They innovated, cut prices, destroyed competitors, and then the whole cycle restarted from zero.

But something changed.

The technology evolved. The chips became extremely complex. So complex that today only three companies control roughly 95 percent of the global market. Samsung, SK Hynix, and Micron. Back in 2023 these companies were hated. They traded at maybe 8 times earnings. Nobody wanted them.

Then demand exploded. Supply could not keep up. Prices rose in 2024. Climbed higher in 2025. Exploded in 2026. The only new entrant trying to break in is China’s CXMT, and for now they hold just a few percent of the market. And those three giants? They are not really competing anymore. They talk to each other. Which means they can now do almost whatever they want with pricing.

I felt this personally. I bought a new MacBook. In Europe the base model with 16GB of RAM costs around 1,500 euros. But I am an intensive user. Research, Substack, photos, video editing. I need at least 32GB. How much does that cost? Around 3,000 euros. The memory guys are effectively charging me the price of an entire second MacBook just for more RAM.

There was supposed to be a glut by now. Massive profits were supposed to attract massive new supply. That is what cyclical industries always do.

But the glut never came.

TrendForce estimates conventional DRAM contract prices rose 90–95% in 1Q26, with another 58–63% increase expected in 2Q26. At the same time, Micron stated:

“We continue to expect supply-demand conditions for both DRAM and NAND to remain tight beyond calendar 2026.”

And now ask yourself something.

What if offshore drilling and met coal are no longer normal cyclical industries either?

Read more

“Pay attention to cannibals, spinoffs, and cloners.”

Charlie Munger said it. Three categories where pattern recognition beats IQ. Three places where smart money keeps finding ten baggers while the rest of the market hunts for the next Nvidia.

Cannibals you know. The reason this newsletter exists.

Spinoffs you know. Joel Greenblatt wrote the book.

Cloners might be the most misunderstood category of the three.

Get 20% off forever

Why Mohnish Bought Kaspi

When Mohnish Pabrai disclosed Kaspi.kz as a top ten position in his Wagons ETF in March 2026, I spent some time trying to figure out what he saw.

The numbers were obvious. 7 times earnings. 50 percent ROE. Founder owning 23 percent. A super app with 16 million monthly users in a country of 20 million people.

But Mohnish does not buy stocks only for the numbers. He buys for the framework underneath the numbers. Mohnish did not buy only a cheap fintech.

He bought a cloner.

Microsoft Has Never Had an Original Idea

For more than a decade Microsoft has sat in the top five most valuable companies on Earth. Most people assume that means Microsoft invents things.

Microsoft invents nothing.

Mohnish laid it out recently.

“You look at Microsoft. Windows is copied from Apple. Word is copied from WordPerfect. Excel is copied from Lotus. Bing is copied from Google. Warren Buffett tells me, do you know what Bing stands for? Bing stands for “But It’s Not Google.” Everything that has worked for Microsoft, including now AI, has been copied from other places.”

Microsoft spends billions every year on internal research. PhDs. Labs. Nothing comes out of them. The most valuable software company in history has never produced a single category-defining product from its own minds.

Everything came from somewhere else. And that is not an insult. Cloning is better than inventing. If you know how to clone a product or service and improve it at the same time, you have an incredible advantage.

The inventor has to guess what people want. The cloner already knows.

The best cloner alive today is not in Seattle.

He is in Almaty. He runs a 16 billion dollar super app. I think he has never had an original idea in his life.

That is why I think Mohnish owns the stock.

His name is Mikheil Lomtadze.

The Boy Who Was Built to Clone

Georgia. 1980s. Soviet collapse on the horizon. A family with no money for food but money for tutors.

In his own words.

“My parents had to work really hard but everything that we had went into our education. I always had tutors in math and physics and English. We might not have food on the table but I always had a teacher. I might have secondhand clothes but I definitely would have a teacher.”

Food was optional. A teacher was not.

That is not how poor families raise inventors. That is how poor families raise cloners. The message is clear. You are not smart enough alone. Find someone who already knows. Sit at their feet. Absorb everything.

He did it for the rest of his life.

He read about Guivy Zaldastanishvili, the first Georgian to graduate from Harvard. Decided to copy the path. Applied to HBS. Rejected. Flew to Boston, sat with the head of admissions, made his case. Applied again. Accepted.

He did not invent a new way in. He found someone like him who had already done it and refused to stop.

His first company was a clone of Andersen Consulting. He built the global accounting franchise in Tbilisi while still a student. When Andersen collapsed in 2002, the franchise rolled into Ernst Young. He took a proven developed-world model and operated it where the originals had not yet entered.

This is the cloner’s signature move. Find a proven model. Operate it where the original cannot go. Collect the difference between developed-world quality and emerging-market pricing.

Twenty years later he would still be doing the exact same thing. Just at a different scale.

Everything Kaspi Is, Lomtadze Cloned From Somewhere Else

The super app. Cloned from WeChat. His own confession.

“The only mobile application in the world that enjoys such engagement is WeChat in China.”

He built the only WeChat outside China by copying the model exactly. 16 million monthly active users in a country of 20 million people. 11 million daily active users.

He did not invent the super app. He cloned it into a market Tencent could not enter and pushed engagement to a level only the original beats him on.

QR payments. Cloned from Alipay and WeChat Pay. China removed Visa and Mastercard from the equation by building closed-loop QR networks. No interchange fees. No third party middlemen. The payment goes from consumer wallet straight to merchant wallet.

Lomtadze cloned the architecture exactly. In his own words.

“The only place in the world where QR payments are more popular than in Kazakhstan is China.”

By 2023 Kaspi was processing 78 percent of all payment network transactions in Kazakhstan. Visa, Mastercard, and every other network combined had 22 percent. Kaspi’s Payments segment now produces a 65.7 percent net margin. For comparison, Visa runs around 55 percent. Mastercard around 46 percent.

The cloner produces fatter margins than the originals he copied.

Postomats. Cloned from Amazon Lockers. In his own words.

“Pick up in the lockers like Amazon Locker style.”

Over 10,000 Postomats across Kazakhstan today. Density per capita higher than the United States. Higher than Amazon’s home market.

Alaqan palm payments. Cloned from WeChat and Amazon. His admission on stage at the launch.

“Palm scan based payments are getting popular in China on WeChat. They are developing this tech just like Amazon in the States.”

He named both originals out loud. Then he shipped the clone in three months. 500,000 customers registered in Almaty before either original had reached that scale outside their home cities.

BNPL. Klarna invented Buy Now Pay Later for consumers. Lomtadze cloned the mechanic and adapted it. Buy Inventory Now Pay Later. A wholesaler delivers goods to a small shop. The shopkeeper scans the supplier’s Kaspi QR. Kaspi pays the supplier in full. The shop repays Kaspi in 30 days from its Kaspi Pay sales.

E-commerce. He admitted the entire methodology in one paragraph.

“For example we want to launch e-commerce, so I will be reading about e-commerce. I will be reading about Amazon, or I will be reading about MELI annual report, I will be reading about Rakuten which we visited, HBS grad, marketplace in Japan. Before launching we did a 4 day trip to Japan.”

He reads Amazon. He reads MercadoLibre. He flies to Tokyo for four days to meet Rakuten. Then he builds his version.

The whole company in few words from the founder himself.

“We are definitely learning like crazy. Always.”

Share Cannibal Stocks

Many Clones. One Result.

Add up everything Lomtadze cloned.

WeChat super app. Alipay closed-loop QR. Amazon Lockers. WeChat palm payments. Klarna BNPL. Reichheld NPS. Amazon customer obsession. Buffett discipline. Munger ability to kill bad ideas.

Microsoft did exactly the same thing for forty years. Cloned Apple. Cloned Lotus. Cloned WordPerfect. Cloned Google. Today Microsoft sits at the top of the global market cap rankings.

Lomtadze is the same playbook. At an earlier stage. In a market the West has barely discovered. The question is no longer what Mohnish saw.

The question is what the Kaspi stock is worth. And what am I doing with the stock?

Read more

• That is how many shares of Constellation Software exist as of May 2026.

• That is how many existed in 2015.

• That is how many existed in 2010.

• That is how many existed in 2006, on the day of the IPO.

Twenty years. Not a single new share.

In that time, the company grew from a billion dollar market cap to 51 billion CAD. Free cash flow is up 26 fold. The stock price is up 130 fold.

All per share. All clean. Because the share count never moved.

1999. Mark made the only real mistake of his career.

Constellation did not need money. They had healthy cash flow and a clear runway. But Mark wanted to diversify the investor base. He agreed to a secondary share issuance. The dilution was permanent. He has called that issuance one of the biggest mistakes of his life. He estimated it cost shareholders billions over the decades.

Since that day, Constellation has not issued a single new share. Sit with that.

Twenty seven years. Hundreds of acquisitions. More than sixteen billion dollars deployed. More than sixty four thousand employees. Six operating groups. Over one thousand acquired businesses. All financed from operating cash flow and modest debt.

Not a single share issued as a weapon.

Other software CEOs would say it is impossible. Mark says everything else is unacceptable.

Constellation buys between 80 and 130 companies per year.

Most cost less than ten million dollars. They pay one or two times revenue. They target IRRs of 25 to 30 percent on small deals.

How is this possible at this scale?

They have a database of 50 thousand VMS firms they track. Not 50. Fifty thousand. They have dozens of people across their hierarchy authorized to buy companies. Each of them has spent years learning Mark Leonard’s discipline.

When somebody offers private equity a VMS firm with 30 percent EBITDA margins and stable growth, private equity probably wins. They can lever it, model a higher exit multiple, accept a lower IRR.

When somebody offers Constellation that same firm but in chaos, Constellation wins. Because they are the only ones who know how fast it can be fixed. Because they have data from a thousand similar fixes. Constellation does not buy champions. Constellation buys companies that will become champions. Under their hands.

That is the difference nobody copies. Because nobody has twenty five years of data and a decentralized machine of dozens of people trained to use it.

Now we get to the part that hurts most managers. Constellation pays no stock based compensation.

Not one RSU.

Everything is paid in cash. Then 75 percent of the after tax bonus, for executives, must go into buying Constellation shares on the open market. Those shares sit in escrow for an average of four years. For non employee board members, all after tax compensation must go into shares. No exceptions.

Mark himself has taken nothing since 2015. No salary, no bonus, no travel reimbursement. He pays for his own flights out of his own pocket. He has said he refuses to free load on Constellation shareholders.

He and his family hold around 7 percent of the company. That is it. That is his compensation.

Think about the effect of this. Every executive is a shareholder. Every board member is a shareholder. Everyone thinks in twenty years because their shares are in escrow. Nobody dilutes anybody.

A typical SaaS firm issues 15 to 20 percent of its shares in SBC every year. The founder knows he will not be there in five years. Nobody is aligned with anybody. Michael Burry pointed out that much of Big Tech profit comes from stock based compensation. Companies pay employees with shares instead of cash. The cost shows up as dilution, not as a real cash expense.

So earnings look strong. Over the last 10 years, the Nasdaq is showing 42 percent more profit than actually exists. But the share count keeps rising. Alpha, Warrior and Transocean all use share based compensation too. That is how most of the industry operates.

Constellation is the diametric opposite.

Now the part that troubles my kind of investor most. Constellation has never bought back a single share.

Never.

Mark refuses for two reasons. The first is practical. The shares rarely trade below intrinsic value, so buying makes no sense.

The second is moral. Mark asked outside lawyers about the ethics of buybacks as a public CEO. They told him officers and directors cannot legally repurchase shares while in possession of material non public information. And Constellation guards its acquisition trade secrets like proprietary treasure. Their entire edge depends on those secrets.

But Mark’s real reason runs deeper than legal advice. A buyback is a transaction in which management and the board, who know more than the sellers, buy as cheaply as possible. The bigger the discount you take for yourself, the more those who sell to you get hurt. Mark sees selling shareholders as the same partners as the ones who hold. He refuses to rob them.

The first time I read this, I had to pause.

My entire job is built around companies that recognize their stock is undervalued and act on it. The whole portfolio teaches me that buybacks are a mechanism for transferring value from the weak to the strong.

Mark Leonard says that is not fair. Mark took the second road. And he won.

2019. The stock price explodes. Mark thinks it is too expensive.

Another CEO would stay quiet, smile at conferences, sell some of his own stock, and disappear to the Maldives. Mark issues a public document. He discloses that his personal investment plan is structured to automatically sell shares above his estimate of intrinsic value and buy them below. He sends a signal to the market not to buy at premium prices.

The market ignores him. The price keeps rising.

Same year, they pay a special dividend. They try to calm the euphoria. To tell investors Constellation is not a miracle without limits, that there is a ceiling.

The market still ignores.

Sit with this idea once more. A capitalist publicly begging the market to stop buying him. There is not a single other case I can name. Not Buffett. Not Munger. None.

That is not market behavior. That is religious behavior. Mark Leonard is not a CEO. He is a steward.

September 25, 2025.

The press release was three paragraphs long.

Mark Leonard resigned as president of Constellation Software for health reasons. Effective immediately. Mark Miller, who joined the company in 2001 and had been chief operating officer, was appointed his successor. Leonard would remain on the board.

That was it. No farewell tour. No final letter. No biographical reflection. The same dry tone of every other Constellation announcement.

In the week that followed the stock dropped 17 percent.

Constellation, the most disciplined compounder of our era, lost roughly 14 billion CAD of market cap because one man stopped showing up. The market voted. The market is afraid. The market is asking the question every reader of this essay should be asking.

Was Constellation a system, or was it Mark Leonard?

In its first twenty five years Constellation deployed about six billion dollars in acquisitions. Free cash flow today is 2.66 billion dollars per year. For Constellation to reinvest all of its free cash flow at the same hurdle rates, the next three years require deploying more capital than the first twenty five combined.

Read that again, slowly.

This was already Mark’s problem. It is now Mark Miller’s problem. And it landed on his desk on day one. A study Mark’s team did on twelve high performance conglomerates showed the same trajectory for all of them. High ROIC in their early years. Mean reversion as they scale. Nobody escaped gravity.

Constellation is on the same path. ROICs are already lower than they were five years ago. The decline is not a question of whether. It is already underway.

What was Mark doing about it before he left?

Lowering the hurdle rate but only on large acquisitions. Small deals still target 25 to 30 percent IRR. Big deals can drop to 17 or 20 percent. Experimenting outside the VMS world. At the 2022 shareholder meeting he revealed he had come close to buying a thermal oil business for a billion dollars. The sector could not get financing. Prices were depressed. There were tax advantages and a clever structure. In the end it did not happen.

Do you understand the weight of that one sentence?

The most disciplined software investor of the 21st century considered buying an oil field.

That evolution is now Mark Miller’s inheritance. He is taking over a company that has already accepted it must change. He must execute the change. The first non VMS deal, when it comes, will be the most scrutinized acquisition in Constellation’s history.

If he gets it right, a new decade of compounding opens up. If he fails, Constellation becomes a slow capital returner instead of a compounder.

What am I doing with this?

Constellation trades around 2440 CAD today. After the Leonard resignation news the stock stayed depressed. AI fears compound the pressure. Market cap is 51 billion CAD. At 38 billion USD with 2.66 billion USD of free cash flow, the multiple is 14.3 times cash flow.

That is the cheapest Constellation has been on a free cash flow basis.

A compounder growing 15 percent per year, with 20 percent ROIC, with a thirty year culture of discipline, at 14 times cash flow.

The market is pricing in two risks. Mark Miller might fail to maintain the standard. AI might erode the moat across the portfolio.

I think both risks are overstated, and the second is already priced.

But for my portfolio. Met coal and offshore drilling positions still offer more return per dollar of risk. Constellation will not 5x in three years. But under Miller, if the discipline holds, it will deliver 12 to 15 percent per year for the next decade. Possibly more if the first non VMS deal is well placed. Right now, with Leonard gone and the market in fear, the entry point is more attractive than it has been in a long time.

Compounding is not a strategy. Compounding is not math. Compounding is faith. Faith in discipline, faith in long horizons, faith in your own intelligence against the fog of the market.

Mark Leonard believed. Constellation Software was his cathedral. Now the cathedral has a new caretaker.

The walls Mark built are still standing. The data library still lives in the operating groups. The hurdle rates are still magnetic. The escrow accounts still hold the shares of men who must think in decades.

Whether the cathedral grows or merely stands depends on what comes next.

I am watching. So is the market. So, presumably, is Mark Leonard, from wherever he is recovering, with the same patience he taught everyone around him.

Cheers, Sandro

Get 20% off forever

“Show me the incentive and I’ll show you the outcome.” - Charlie Munger

In the last days of his life, Charlie Munger was buying one stock.

Between November 17 and November 24, 2023, that stock traded between $259 and $273. I assume his final purchase was made somewhere in that range. A few days after that final trade, he was gone.

But he left us one last clue.

His final transaction.

The man who gave the world the incentive framework spent his last moves applying it. To one company. To one misprice. In silence.

Every commodity market has a price. Every commodity index claims to discover that price. The two are not always the same number.

A few days ago, Alpha Metallurgical Resources reported a Q1 2026 net loss of 11 million dollars. The company should have reported a profit of 25 million. The 36 million dollar gap between what Alpha actually earned and what its headline numbers said it earned is not an accounting issue. Not fraud. Not management hiding anything.

It is a coal mine in Alabama. Owned by Warrior. Labeled wrong.

The mine is Blue Creek. The label is Premium High Vol A. The cost to Alpha shareholders is 107 million dollars a year on a stock trading at 180 dollars. At today’s price, that is 600,000 shares of Alpha float the buyback program cannot retire this year.

This is the most expensive label in metallurgical coal. And it sits on a mine Alpha does not even own.

Two Ingredients

Most coal is used to make electricity. That is thermal coal. Metallurgical coal makes steel. Without met coal, there is no steel. Inside the coke oven, met coal does one of two jobs.

Think of a bread oven. Flour gives bread its body. Yeast makes it rise. Both required.

A coke oven works the same way. Some coals provide strength. Other coals provide flexibility. Strength coals are called low vol and mid vol. Flexibility coals are called high vol. Every coke blend on Earth uses both, mixed in roughly a 70/30 ratio.

Two ingredients. Two jobs. Every coke blend on the planet.

Below the line, four answers:

• Which mine carries the wrong label — and why both the SEC filings and the company’s own marketing already admit it

• Who the villain is, who the victim is, and why the villain is also bleeding three hundred million a year

• Why this exact misprice is what Munger was compounding into in his final days

• When the gap closes — and the three forces that close it

Read more

In late March 2026, Mohnish Pabrai disclosed Kaspi.kz as one of the top ten holdings in his Pabrai Wagons ETF.

Three weeks later, on April 20, Tencent entered Kazakhstan’s fintech sector for the first time in its history. Around 518 million dollars for a 3.2 percent stake in Kaspi.

On the same day, the founder bought more.

Three buyers. Same stock. Same quarter.

The stock is Kaspi.kz. Market cap around 16 billion dollars. Trading at 7 times earnings while Nvidia trades at 45.

The man who paid 650 thousand dollars to have lunch with Warren Buffett. The Chinese company that invented the super app. And the founder himself.

All three of them looked at this stock at 7 times earnings and bought more.

What Is This Thing?

Kaspi is the operating system of Kazakhstan.

Across Kazakhstan and Türkiye the platform now serves more than 25 million consumers and nearly 1 million merchants. The average user makes 77 transactions per month inside the app. There is one other app on Earth with this kind of engagement. WeChat. Built by the company that just bought into Kaspi.

Inside the app you pay bills. Buy electronics. Order groceries. Book flights. Apply for a business license. Get married. Sell your used car over coffee. File your taxes. Renew your driver’s license. Get a loan. The entire life of a Kazakh citizen lives inside one icon.

The numbers from the 2025 annual report tell you the scale. Total payment volume of around 87 billion dollars per year. Marketplace of around 13 billion. Fintech originations of around 23 billion. Revenue including Türkiye was 8 billion dollars in 2025. Net income was 2.1 billion. Return on equity over 50 percent.

The only companies on Earth with returns on equity this high are called Visa, Mastercard and Moody’s.

The stock trades at 7 times earnings.

How.

One Word.

Kazakhstan.

Say it to your wife. “Honey, I’m putting our entire savings into Kazakhstan.”

Watch the look on her face.

That is the entire reason this stock trades at 7 times earnings instead of 30.

Real Kazakhstan went from a 90 percent cash economy to a 90 percent cashless one in less than ten years. The government digitized everything. You register a business in three minutes. You transfer car ownership with a tap. Try doing that in Germany, where I live.

The country leapfrogged the developed world. Kaspi was the rail it ran on.

The discount is not about the business. The business is exceptional. The discount is about a stereotype that has nothing to do with the country that exists today. This is the gap. The Kazakhstan in Wall Street’s head versus the Kazakhstan that exists on the ground. That gap is your edge.

The Founder Who Will Not Sell

When the Tencent transaction closed, a founder bought more.

He already owned around 23 percent of the business. He took the chance to add to his stake personally. Senior management did the same. Insiders at a company trading at 7 times earnings are buying. And they are buying alongside Tencent and Pabrai.

This is not a man who needs more money. This is a man telling you, with his own checkbook, that the stock is mispriced.

Lomtadze took a small failing bank in 2007 and built it into a 16 billion dollar super app. Harvard Business School wrote two case studies on him. He survived a bank run, an attempted coup in January 2022, and Russia’s invasion of Ukraine the very next month. The man does not break.

His policy is to return capital to shareholders. The dividend, just resumed after the Hepsiburada acquisition closed, runs at roughly 8 percent of the current share price.

A super app. Growing 20 percent. Paying you 8 percent.

Find me another one.

The Bad Money Story

Ten years ago, Kaspi had a credit card product. It generated roughly one third of the company’s net income. Hundreds of millions of dollars per year, straight to the bottom line.

Lomtadze killed it.

Not because it was unprofitable. Customers did not love it. He called the revenue “bad money.” Cash the company should not be earning because it was not making customers’ lives better.

He shut it down. Walked away from a third of his profits.

You cannot manufacture this. You cannot reverse engineer this culture. You cannot fake it. It either exists, or it does not.

The Part Tencent Could Not Stop Thinking About

There is one specific dynamic about Kaspi that, once you understand it, makes the 7x multiple feel almost criminal. It is the single asset that turns a cheap stock into an inevitable one.

It is not the dividend. It is not the founder ownership. It is not the user engagement.

It is a specific catalyst inside the Türkiye business, combined with one ambition the CEO put into writing in his 2025 letter, plus a statement made publicly by Xi Jinping in Astana that quietly rewrote the geopolitical risk equation on this stock. Almost no Western analyst has internalized it yet.

If you are a free subscriber, this is the decision point.

The first half of this post was the setup. The second half is the trade. The 100 million user ambition. The Türkiye engagement gap that nobody is modelling correctly. The Xi statement in full. The Kaspi launch that tells you the product machine is still accelerating. The peer multiple breakdown showing exactly how much room this has to re-rate. The honest risks, including what Elon Musk just dropped on the entire super app industry. And the one sentence that captures the whole thesis.

Paid subscribers get all of it. Plus every position update going forward. One cup of coffee per month.

If you made it this far, you already know this is not a normal Substack.

Welcome to Cannibal Stocks.

Get 20% off forever

Read more

Alpha Metallurgical’s Q1 2026 looks like that report from a distance.

It is not.

If you only read the press release, you will see a loss of 11 million dollars and conclude that another coal company is bleeding through a war. If you read the earnings call carefully, you will see a company shifting its weight from one foot to the other.

That shift is the entire story. Let me show you why.

The coal market just split into three

Right now there are three benchmark prices for metallurgical coal trading.

• Australian Premium Low Vol sits at $239.80 per ton.

• US East Coast Low Vol sits at $195 per ton.

• US East Coast High Vol A sits at $159 per ton.

Three completely different prices.

The gap between Australian premium and US Low Vol is 23 percent. The gap between US Low Vol and US High Vol A is also 23 percent. Andy Eidson called these “historically unusual divergences” on the call this morning. That is a careful way of saying the market is broken.

A simpler way to think about it

Imagine three butchers on the same street.

• The first sells filet mignon. People line up. The price is $239 and rising.

• The second sells ribeye. Steady demand. Steady price at $195.

• The third sells ground beef. Way too much of it. Nobody really wants it. The price is $159 and the butcher is sweating.

That is the coal market in May 2026.

Filet is Australian Premium Low Vol. The world cannot get enough of it. Ribeye is US East Coast Low Vol. Solid, useful, in demand. Ground beef is US East Coast High Vol A. Drowning in supply. The market could still rebalance toward the end of the year, but Daniel E. Horn was very clear about what needs to happen.

“I think we have to see some demand improvement and some continued supply discipline.”

He also added:

“As demand improves, that will help somewhat with the rebalancing.”

What is actually happening at the company

Set the loss aside for a moment. Look at the trend.

Q1 2025 loss was 33.9 million. Q1 2026 loss is 11 million. EBITDA went from 5.7 million a year ago to 30 million this quarter. Realized price per ton went from 118 to 124. Margin per ton went from 8.27 to 16.41.

Alpha sits on 317 million in cash, 50 million in short-term investments, 184 million of unused credit. Total liquidity of 476 million. Long-term debt of 12 million. A two billion dollar company holds half a billion in liquidity and almost no debt. When the cycle turns up, they will buy. When the cycle turns down, they will buy. They have built a fortress and they intend to use it.

The buyback proves it. In Q1 they spent 22.9 million repurchasing shares. Since 2022 they have spent 1.2 billion buying 7 million of their own shares at an average of $166. There are now only 12.7 million shares left outstanding.

A company that has eaten more than a third of itself in three years. And keeps eating.

What the market sees and what is actually there

Most readers of this morning’s release will see a loss in a coal company during a war and move on. That is what the market is pricing.

What the market is not pricing is a low-cost producer that just told you, in a few quiet sentences scattered through a 30-minute call, that it has reorganized its entire portfolio toward the most expensive product on the menu while its competitors flood the cheapest one.

The most honest description of the high vol coal market I have heard in two years came from Alpha’s own commercial officer this morning. It explains why Alpha walking out of that fight is the smartest decision they could make right now.

I want to walk you through it. And then I want to show you the single insider transaction in March that, in my opinion, tells you exactly how someone who sits in the boardroom interprets all of it.

Read more

One doubled.

The other halved.

Same industry. Opposite fates. If you understand why, you understand more about offshore drilling than most institutional analysts paid to cover it.

The Numbers

Transocean took EBITDA from 244 million a year ago to 440 million this quarter.

Valaris went from 181 million to 67. Down 63 percent.

Why?

Utilization. Transocean is running 87 percent of its rigs. Valaris is running 58. Their drillship fleet sits at 38 percent. The lowest active number in the industry. Three of those drillships are preservation stacked. That is the polite term for parked indefinitely.

Day rates tell the same story. Valaris drillships earn 436 thousand per day. Transocean ultra deepwater earns over 480.

Then this. Valaris quietly announced they will no longer hold earnings calls or provide guidance. That is not an operational decision. That is the language of a company that has accepted what it is becoming.

Lucky Valaris Saved by Transocean

Valaris shareholders should send Anton Dibowitz a thank you card. Their business is cracking. The weaker rigs sit idle. Pricing is softening. They are running out of options on their own. Falling into Transocean’s lap is the best outcome they could realistically hope for.

But Transocean wins too.

This is not just a rescue. Transocean takes four real things from this deal. Valaris carries less debt. The merger drops leverage of the combined entity the day it closes. The new firm has a bigger footprint and stronger negotiating leverage with customers on rates and terms. One competitor disappears from the industry, which means less price war for everyone still standing. And Transocean is the better operator. The numbers above prove it. Put the Valaris fleet under that umbrella and the combined company can run materially better than Valaris ever did alone.

A weaker, less indebted competitor swallowed by the stronger one. Balance sheet repaired, market power increased, competition removed.

All in one move.

What the Transocean Headlines Sing About

Revenue 1.08 billion dollars. EBITDA 440 million at a 40 percent margin. Free cash flow 136 million.

If you read this newsletter, you know headline numbers are never the whole story. We read what is underneath.

Read more

All true. But none of it is the point.

Blue Creek is finished.

Walter J. Scheller said it like this on the call.

“Blue Creek was delivered ahead of schedule and fully in line with our capital expenditure guidance, on budget and fully paid out of cash from operations without incurring any funded debt.”

Total spend $1,022.9 million. Not a single cent of new debt. Delivered ahead of schedule. The final $66 million tranche was paid in Q1.

Built a $1 billion mine without a bank.

The number that matters

“When a company is selling a product with commodity-like economic characteristics, being the low-cost producer is all-important.”

— Warren Buffett, Shareholders letter 2000

Cash cost of $96.17 per ton FOB port.

Q1 2025 cash cost was $112. A 14 percent drop in one year. Fifteen dollars per ton lower. Three dollars of that came from the new 45X tax credit. Strip out the subsidy and the underlying cost is $99 per ton.

Ninety nine dollars per ton FOB port.

That is the number you need to remember. The question I always ask about a cyclical commodity company is what happens in a brutal bear market?

If PLV drops to $130, Warrior makes money. If it drops to $110, Warrior makes money. If it drops to $100, breakeven. Look at the history. PLV has almost never traded below $100 in the last twenty years. This is the definition of a company that cannot lose money.

And it gets better. Blue Creek is just starting. Once it ramps to its full 6 million tons per year, cash cost goes lower. I think we see numbers in the mid eighties by 2027.

Why diesel prices do not hurt Warrior?

Read the recent commentary on coal stocks and you keep hearing the same worry. Diesel is up. Oil is up. Surface miners are getting squeezed. Trucking is more expensive. Margins are tightening.

This is where Warrior is different from almost every other met coal producer. When asked directly about diesel exposure on the call, Walt was almost dismissive.

“We do not do a lot of trucking of coal. We do truck a little bit to the barge load-out, so we are not a high user like strip mines or surface mines. We just do not use a lot of diesel.”

Warrior runs longwall underground mines. The coal moves on belts and rails, not on diesel trucks. For a typical surface miner, diesel can be 10 to 15 percent of total cost. For Warrior it is a rounding error.

This is what a real moat looks like. Warrior is the turtle. Slow, low to the ground, hard to crack. The diesel storm passes over the surface miners. Warrior just keeps producing at $99 a ton and lets the market price adjust around it.

I am not going to tell you what Warrior is worth. I am going to show you what it pays you while you wait. You have seen the business. Now you will see what it pays you.

Read more

If you could only own one for the next ten years, which would it be?

I thought I knew. I have written about both. I have defended both in comments sections at 11pm. I opened a spreadsheet just to confirm what I already believed.

Two hours later I closed the laptop and sat in the dark.

The company I thought was the winner had just lost on paper. And the one I thought was expensive was quietly the better business by almost every measure that matters over twenty years.

I am going to show you exactly what I saw. Ten rounds. Ten metrics. One uncomfortable answer. But before we start, one quote. Read it like your portfolio depends on it. Because it does.

“If the business earns 6% on capital over 40 years and you hold it for that 40 years, you’re not going to make much different than a 6% return—even if you originally buy it at a huge discount. Conversely, if a business earns 18% on capital over 20 or 30 years, even if you pay an expensive looking price, you’ll end up with a fine result.”

- Charlie Munger

Over twenty or thirty years, return on capital is signal. Everything else is noise.

That is the fight behind every round that follows.

I own two stocks the market has decided to hate.

Same father. Same playbook. Same incentives. Same culture of managers who buy shares with their own money.

One earns 34% on capital. The other 20%. One trades at 19x cash flow. The other at 25x. One is hated because the market thinks AI will kill its industry. The other sits in a continent about to spend trillions on software.

A compounder growing at 20% doubles your money every 4 years. A compounder growing at 30% doubles your money every 2.5 years.

Right now the market is terrified. The AI selloff has gutted entire sectors. Fear everywhere you look. And in that chaos, Mr. Market is handing out wonderful businesses at fair prices.

Ten rounds. Let’s go.

Round 1. The Industry

Lumine sells telecom and media software. Billing systems. Network management. Content delivery. Streaming platforms. A trillion-dollar industry that is barely growing. Global telecom revenue grows at 2.8% a year. Operators are squeezing suppliers on price. Ericsson, Nokia and AWS are circling.

If you sell software to telecoms, you are selling to customers fighting for their own survival.

Topicus sells to regulated European verticals. Education. Healthcare. Social services. Local government. Financial services. Legal. Real estate. That market is projected to grow from $45.5 billion to $83.4 billion by 2030. A 12.9% CAGR.

And here is the thing about regulated industries. Nobody disrupts them quickly. You cannot move fast and break things when the thing you are breaking is a hospital’s patient database or a municipality’s tax system. The regulations themselves are the moat.

Lumine sells to an industry that is shrinking its vendor list. Topicus sells to industries that are legally required to buy software.

Score. Topicus 1. Lumine 0.

Round 2. Return On Capital

This is the round that should decide everything.

If Munger is right, if return on capital really is the whole story over twenty years, then one clear winner here should make the other rounds academic.

Which should make the other eight rounds irrelevant. Should.

Who wins?

Read more

“The real money is going to be made in the growing businesses, and that's where the focus should be.”

- Warren Buffett, 2012 Meeting

Let me tell you how close Warren Buffett came to missing the best investment of his life.

It was 1972. A candy shop in California. The owners wanted 25 million dollars. Warren and Charlie thought it was expensive. Twelve times cash flow. For two guys who had built their careers buying dollar bills for fifty cents this was pain. They almost walked away over 100,000 dollars. One hundred thousand. On a 25 million dollar deal.

“If See's Candies had asked $100,000 more, Warren and I would've walked. That's how dumb we were at that time.”

- Charlie Munger, 2003 Meeting

But they bought it. By 2026 that candy shop had returned roughly 3.15 billion dollars in pre-tax profit to Berkshire. A 130-bagger. From a candy shop.

“The ideal business is one that earns very high returns on capital and can keep using lots of capital at those high returns. That becomes a compounding machine. If you can put $100 million into a business that earns 20% on capital, ideally it would be able to earn 20% on $120 million the following year, $144 million the following year, and so on. You could keep redeploying capital at these same returns over time.

But there are very, very, very few businesses like that. We look for them, but they don't exist."

- Warren Buffett, 2003 Meeting

Today I want to tell you about a European software company that looks exactly like that animal.

It trades at 123 times earnings.

And I just bought it.

My name is Sandro. I run Cannibal Stocks.

You know my style. I buy hated companies in hated industries. Alpha Metallurgical Resources at 1.3 billion in free cash flow on a 2 billion market cap. Transocean at about 25 cents on the dollar of replacement value. Ugly cheap. So cheap that even if I am wrong about everything else the price saves me.

So when I tell you I just bought a stock at 123 times earnings you should probably check if someone hacked my account. 🙂

Nobody hacked my account.

What happened is that I spent two weeks inside this business and I found something I have been looking for my entire investing life. A machine that takes one dollar and turns it into twenty cents of cash every year. Forever. Into a market so big that the machine cannot eat through it in my lifetime. Or my kids’ lifetime.

The stock is called Topicus.

And the 123 PE is a lie.

Let me show you.

Read more

For those who are caught up. At the end of that post I said I was not buying Constellation. Too big. The 140x is behind us. I said I was looking for the next Constellation.

I found it.

Its name is Lumine Group. It trades in Canada under the ticker LMN.V.

Let me show you what I found.

One Billion Devices

Before I tell you about Lumine I need to tell you about two phone calls.

The first call happened in 2018. Someone at Nokia picked up the phone and called Lumine. Nokia had a problem. They owned a video streaming software business called Velocix that they had inherited through the Alcatel-Lucent merger. It was a good business. Content delivery. Caching. Streaming personalization. Customers in 14 countries. But Nokia did not want it. It did not fit their strategy. They needed someone to take it off their hands.

Private equity would buy it. Gut it. Flip it in five years. Nokia did not want that. They wanted a permanent home for the engineers and the customers.

Lumine said yes. They bought Velocix. Restored the original brand name that Nokia had buried. Let the team run independently. Gave them capital and a playbook of 250 best practices.

By 2024 Velocix won the NAB Product of the Year Award. A business Nokia considered dead weight became an award winner.

Then came the second call. December 2023. Nokia picked up the phone again. This time the business was much bigger. Their Device Management and Service Management Platform. Software with roots in two prior Nokia acquisitions called Motive and mFormation.

This platform manages over one billion devices across 150 deployments worldwide. Broadband routers. IoT sensors. Multi-vendor device fleets. The software that telecom operators use to remotely manage every piece of hardware connected to their networks.

One billion devices.

Nokia wanted 202 million US dollars. Five hundred employees would transfer to Lumine.

Why did Nokia call Lumine a second time? Because the first deal worked. Velocix thrived. The engineers were happy. The customers were retained. The brand was restored. Nokia trusted Lumine.

Reputation compounds just like capital. I wrote that in Part One. This is what it looks like in practice.

The 90% Monopoly

Now let me tell you how Lumine was born.

In 2014 a man named David Nyland walked into Constellation’s Volaris Operating Group with a mandate. Build a communications and media software vertical from scratch. Nyland had the resume for it. Software developer at Accenture. Two CEO roles. Acquisitions across North America and Europe. He knew how to buy and build.

For nine years he did exactly that. Quietly. Deal by deal. Small telecom software companies. Billing platforms. Fraud management systems. Mobile optimization tools. By 2022 he had assembled a portfolio generating 228 million in revenue.

Then Constellation did something it had only done once before. They spun Lumine off as an independent public company. But the spin-off came with a gift. An acquisition so good it justified the entire separation.

WideOrbit.

WideOrbit is the system of record for over 5,000 broadcast stations and cable networks in the United States. It processes more than 35 billion dollars in advertising revenue every year. It commands 90% market share among US local TV stations for ad trafficking and sales.

Ninety percent. In an industry where the software is so embedded in daily operations that switching would shut down the station.

94% of revenue is recurring. Renewal rates are nearly 100%. Gross margins 70%. And here is the number that matters most. Before the acquisition WideOrbit’s estimated return on invested capital was 53%.

Constellation paid 490 million. Three times revenue. Eleven times EBITDA. Expensive by Constellation standards. But when you are buying a 90% monopoly with a 53% return on capital you are not overpaying. You are stealing.

The structure was genius. Lumine stock was used as partial payment. The founder Eric Mathewson joined the board and kept running WideOrbit as CEO. His wealth is now tied to Lumine’s success. Everyone has skin in the game.

One deal. Lumine nearly doubled in size overnight.

The CEO Who Came Back Twice

I have saved the best story for last.

In November 2023 Lumine bought two small business units from a Nasdaq-listed company called Synchronoss Technologies. Standard carve-out. 41.8 million dollars. Nothing remarkable.

Lumine did what it always does. Restored the heritage brand names. Openwave. RazorFlow. SpatialNetworX. Names that had been buried inside Synchronoss through years of corporate acquisitions. Lumine resurrected them. Gave the teams autonomy. Applied the playbook.

Jeff Miller the CEO of Synchronoss watched this happen from the outside. He watched his former divisions come back to life under new ownership. Two years later in December 2025 Miller called Lumine and said take everything.

All cash. Nine dollars per share. 258 million dollar enterprise value. Lumine took Synchronoss private. Delisted it from Nasdaq. Absorbed the entire company into the portfolio.

Miller himself said the former businesses “have flourished under Lumine Group’s stewardship.”

Think about what just happened. A CEO sold two small pieces of his company to Lumine. Watched them thrive. Then sold the entire rest of the company because he was so impressed. That is a referral. The most expensive referral in corporate history.

And it reveals something about Lumine that no financial metric can capture. When you are the buyer that makes businesses better instead of worse, sellers come to you. You do not have to chase deals. Deals chase you.

The Number That Changes Everything

The stock is down 58% from its all time high.

I spent days going through Lumine’s filings, Constellation’s shareholder letters, RBC equity research, and every deep dive I could find. What I found does not make sense.

There is a single metric that Constellation’s entire empire was built on. The metric Warren Buffett called the difference between a great business and a mediocre one at the 1995 Berkshire Annual Meeting. The metric that determines how fast intrinsic value compounds per share.

Lumine scores higher than Constellation itself.

What follows is the full breakdown. The number. The formula Mark Leonard buried in his shareholder letters. A free cash flow model through 2035.

And my position.

Read more

You will find two. Maybe three. That is it. No Twitter account. No LinkedIn. Nothing.

This man built a 300-bagger. And the world has almost no idea what he looks like.

His name is Mark Leonard. He is the founder of Constellation Software. He took the company public in 2006 at 17 dollars a share. By May 2025 that share was worth 5,300 dollars. A 34% annual return for nearly two decades. One of the greatest track records in stock market history.

Then the stock crashed 50%.

Today it trades around 2,300 dollars. And even after that crash it is still up 140 times from the IPO. Most investors would kill for a 10-bagger in their lifetime. Leonard delivered 140. After the crash.

This story is about the man. And about what he built. And about what happens next.

Give a gift subscription

The Terrible Banker

Leonard started his career as an intern at Barclays. A mentor pulled him aside early and said something that would change his life.

“You are going to make a terrible banker.”

The mentor told him to go study the firm’s clients. Find something else. Anything else.

So Leonard wandered into venture capital. And there he met a man named Steve Scotchmer. An advanced materials engineer who had successfully built and run his own business. They searched together for a company that Leonard could invest in and Scotchmer could run.

They never found one.

Scotchmer’s bar for quality was impossibly high. Nothing met his standard. But something more important happened during that search. Scotchmer became Leonard’s mentor. He fed him Buffett. He fed him Munger. He taught him what a truly great business looks like.

Leonard went back through his venture firm’s portfolio with fresh eyes. And he zeroed in on a category of business that checked every single box.

Except one. They were too small to matter on their own.

The Dentist’s Secret

Think about your dentist.

When you walk in the door there is software running the entire practice. Appointments. Patient records. X-rays. Billing. Insurance claims. Every single part of the operation.

That software was not built by Microsoft. It was not built by Google. It was built by a tiny company you have never heard of. A company that spent 20 years sitting next to dentists learning exactly what they need. Tweaking. Adjusting. Customizing. Building trust that took decades to earn.

That is vertical market software. VMS for short. One product. One niche. One deep relationship. Leonard described it in a single sentence that explains his entire empire.

“Vertical market software is the distillation of a conversation between a vendor and the customer that has gone on frequently for decades.”

Now think about the economics of that business.

The dentist is not switching. Ever. The software runs everything. Switching would mean weeks of chaos. Retraining staff. Losing data. Risking patient records. And for what? The software costs less than one percent of the practice’s revenue. The dentist does not even think about it.

Constellation retains somewhere in the mid 90s percent of its customers every single year. The average customer relationship lasts 20 to 30 years. The margins are high. The revenue is recurring. The capital requirements are low. And here is the part that really got Leonard excited. These markets are tiny. So tiny that there are usually only two or three players. Nobody else wants in. The addressable market is too small to attract big money.

Venture capital does not care about a 5 million dollar software company that runs bowling alleys. Private equity is not interested.

But Leonard was.

He found the one flaw. These businesses were incredible. But they were small. They could not absorb much capital individually.

So he came up with an idea. What if you did not just buy one? What if you bought hundreds?

The Machine That Never Stops

In 1995 Leonard raised 25 million dollars. Mostly from pension funds. He built a holding company called Constellation Software. And he started buying. The rules were simple. Buy vertical market software companies. Buy them directly from founders whenever possible. Pay fair prices. Never sell.

Take the cash flow. Buy more. Rinse and repeat. For 30 years.

Today Constellation owns more than 600 of these businesses. They have sold exactly one. Leonard still regrets it. The typical deal is 3.3 million dollars. These are companies so small they are invisible to the rest of the financial world. But inside each one is a machine that prints sticky recurring revenue from customers who will never leave.

And the supply of deals is endless. Some estimates put the total number of vertical software companies worldwide at 70,000. Constellation contacts them at least twice a year. They have a large team making phone calls and sending emails to founders. The message is always the same. If you ever decide to sell please think of us.

Most of these founders are between 40 and 80 years old. Every year a few percent want to retire. Others want to move on. Some need the cash yesterday. That means thousands of potential deals flowing into the pipeline every single year without Constellation having to chase anything. About half of revenue comes from government agencies. Free cash flow has compounded at 26 percent annually since the IPO.

What separates Constellation from every other software company? This is clean free cash flow. Real cash. No stock based compensation tricks. No adjusted metrics. Constellation pays its people in actual money. When you see a software company reporting 20 percent free cash flow margins remember that most of them are hiding massive equity compensation costs. Constellation is not hiding anything.

So why has nobody replicated this? After 600 acquisitions Constellation has a proprietary dataset that tells them exactly what a broken software company is worth and exactly how to fix it. They win the deals nobody else wants because nobody else has the pattern recognition. Add 30 years of reputation as a permanent home for founders and you have a moat that gets wider with every deal they close.

The Anti-CEO

Now let me tell you about the man himself. Because this is where it gets strange.

Leonard asked for his salary to be reduced to zero. Then he stopped taking a bonus. Then he started paying his own travel expenses. He explained that he used to fly economy and stay in cheap hotels because he did not want to freeload on Constellation’s shareholders. But as he got older and wealthier he wanted more comfortable travel. So instead of charging first class to the company he just paid for it himself.

More than 100 employees owned over a million dollars in Constellation stock by 2015. Leonard wanted that number to grow five times by 2025. Every employee above a certain pay threshold must invest 25 to 75 percent of their after-tax bonus into Constellation shares. Those shares are locked up for four years. More than 3,000 employees participate. Insiders today own nearly 1.5 million shares worth about 2.6 billion dollars.

There is no stock based compensation. The share count has barely moved since the IPO. Zero dilution.

Here is a story that tells you exactly who this man is.

In 2011 Constellation’s two private equity sponsors wanted out. To find buyers and keep the company independent Leonard made a personal guarantee. If Constellation did not compound at least 10 percent per year he would transfer a portion of his own shares to the buyer.

His own shares. Out of his own pocket.

It was never necessary. But the fact that he offered tells you everything.

He considers his second capital raise in 1999 one of his biggest failures. They did not need the money. They raised it just to diversify the investor base. The dilution was permanent. It has cost shareholders billions. It was the last time Constellation ever issued equity.

The stock kept going up anyway.

And about buybacks? Leonard has moral reservations. He believes management has more information than selling shareholders. His securities counsel told him that directors cannot legally do buybacks while holding material undisclosed information. To reveal enough to make buybacks legal would mean exposing trade secrets. That would hurt long-term shareholders more than the buybacks would help.

So he simply does not do them.

This is not a man playing the Wall Street game. This is a man playing a completely different game. And winning.

The Crash

On a Tuesday morning in September 2025 Constellation Software released a short statement. Mark Leonard was stepping down as President due to health issues. Effective immediately.

Thirty years. Gone in a press release.

His replacement was Mark Miller. One of the founders of the very first company Constellation ever acquired back in the 1990s. A man who had been inside the machine since day one. If anyone understood the culture it was him.

But the market did not care about culture. The market cared about one thing. The man who had compounded capital at 34 percent a year for three decades was no longer running the show.

Then in March 2026 Leonard announced he would not stand for re-election to the board. His last day would be May 15. The founder was not just stepping back. He was stepping out.

At the same time a wave of panic swept through every software company on the planet. Artificial intelligence was going to change everything. The narrative was simple and brutal. Why would anyone pay for niche software when AI could build a custom replacement in minutes?

Constellation fell 50 percent. From 5,300 dollars to 2,300. Billions in market value gone. Other software names fell 70 percent or more.

Thirty years of compounding. And half of it erased in months.

Now here is where most people stop thinking. The stock is down. The founder is gone. AI is coming. Sell everything.

But stop for a second.

Before he left Leonard addressed the AI question directly. He told the story of a prediction made years ago by an AI specialist who said artificial intelligence would eliminate radiologists. The specialist told young people to stop studying radiology entirely.

Over the following decade the number of radiologists grew 17 percent. Most of them are thriving. AI did not kill the profession. It made radiologists faster and more accurate.

Now think about what Constellation actually owns. These are not generic consumer apps. These are systems that run hospitals. That manage water treatment plants. That operate transit authorities and courts and schools. Systems built through decades of conversation with customers who depend on them every single day.

Your dentist is not going to ask an AI to rebuild the software that manages patient records and insurance claims and appointment schedules and regulatory compliance across three jurisdictions. Not this year. Not in five years. Probably not ever.

Will some of Constellation’s thousand-plus subsidiaries feel pressure from AI? Of course. But AI is far more likely to become a tool inside these products than a replacement for them. The same way it became a tool for radiologists instead of a replacement.

Constellation now trades below 15 times free cash flow. The cheapest it has been since 2014. For most of the last decade the market valued it between 25 and 30 times.

The greatest compounding machine in Canadian stock market history. At a decade-low valuation.

The Question

Can this machine keep running without the man who built it?

Here is what gives me confidence. Leonard spent 30 years making himself unnecessary. The decentralized structure was not an accident. It was the plan. Six operating groups run independently. Capital allocation decisions are pushed down to the people closest to the deals. The incentive systems are locked in. Over 3,000 employees own the stock. The playbook is not in Leonard’s head. It is in the culture.

But I will not lie to you. Losing a founder of this caliber is never nothing. Buffett is 93 and Berkshire investors still get nervous. Leonard is 66. He is stepping away not dying. There is a difference. But it is the biggest transition in Constellation’s history and the market is right to ask hard questions.

Here is what I think.

Constellation Software is one of the greatest compounding machines ever built. Mark Leonard is one of the most underrated capital allocators alive. The model is proven. The culture is deep. The track record is 140 times your money even after a 50 percent crash.

But at 50 billion CAD I think I am too late. The 140-bagger already happened. I am not here to buy what already compounded. I am here to find what is about to.

And that brings me to a new idea.

“Pay special attention to cannibals, cloners, and spin-offs.”

That is Charlie Munger’s advice.

Constellation spun off two companies. Most people know about Topicus. But there is another one. Smaller, less known, and with a growing presence in Europe.

A company born directly from Constellation’s DNA. Same playbook. Same incentive structure. Same acquisition discipline. But smaller. Much smaller. Thirty businesses today versus six hundred. A CEO who is in active conversation with hundreds of founders. A pipeline of over a thousand targets. And Constellation still holds 60 percent of the equity.

Mark Leonard said one thing about this company that stopped me cold.

“I hope my grandkids are still holding Lumine shares 50 years from now.”

Read that again.

This is the man who built a 140-bagger. A man so careful with words that he writes one shareholder letter a year and says almost nothing in public. He has never said anything like this about any other company. Not even Constellation itself.

I spent the last few weeks inside Lumine’s filings. The acquisition math. The capital allocation. The management incentives. The runway ahead. What I found changes how I think about this entire sector.

That deep dive is for paid subscribers.

If you want to see what I see, the link is below. And if you enjoyed this piece, give it a like and restack. I write about stocks nobody has heard of, so this is basically my only chance of convincing the algorithm I exist. 😅

Cheers, Sandro

Get 20% off forever