Investing Frameworks11 min read
The "Shovel Sellers of the AI Gold Rush": Samsung, SK Hynix and Micron Are Being Crowned Kings — But What You're Buying Is the Most Brutal Cyclical Stock in History
The "shovel seller" narrative sounds airtight: you don't know if the miners strike gold, but the person selling the shovels always wins. Except memory isn't an ordinary shovel — it's a commodity business that periodically collapses. Howard Marks' pendulum, Buffett's moat, and Graham's margin of safety will show you just how dangerous that label is.
A Story That Sounds Perfect: I Don't Care Who Mines the Gold — I Sell All the Shovels
In the tech-investing world of 2026, a seemingly airtight story is circulating. Who wins or loses the AI arms race is anyone's guess, but one thing is certain: they all need enormous quantities of high-bandwidth memory (HBM) to feed those power-hungry accelerators. And so SK Hynix, Samsung and Micron — the three companies that control the vast majority of global HBM supply — have been crowned by the market as "the shovel sellers of the AI gold rush."
The metaphor's power lies in its intuitive correctness. In the 1849 California gold rush, the people who truly got rich weren't the miners — they were the ones selling jeans and iron shovels. Applied to today: nobody knows whether OpenAI, Anthropic or xAI will have the last laugh, but whoever wins has to buy memory from these three. And the data is genuinely stunning: SK Hynix and Samsung control roughly 90% of HBM supply, Micron's HBM capacity is "sold out through the end of 2026," and DRAM spot prices spiked over 160% year over year. Orders booked into next year, capacity snapped up, margins soaring — it looks like the guaranteed golden shovel.
But there's a fatal sleight of hand hidden here. Real "shovel seller" businesses are great because they have pricing power and stability — you can only buy the shovel from one place. Memory, however, is the closest thing in semiconductors to a pure commodity: standardized, substitutable, with capacity that can be expanded almost without limit. When Samsung and SK Hynix see today's windfall profits, they will do the exact same thing — expand capacity like crazy. And the end of every expansion cycle is, always, a price collapse.
The Truth About Memory: It's Not a Shovel — It's the Most Brutal Cyclical Stock in History
To understand how dangerous this arena is, you have to look at its history. The memory industry follows an iron-law cycle: demand explodes → capacity falls short → prices soar → windfall profits lure every player to expand → capacity floods the market a year or two later → oversupply → prices collapse off a cliff → the whole industry posts losses → capex is slashed → the cycle begins again. This curve repeats, over and over, like a roller coaster.
This isn't theory. In the 2018-2019 memory winter, DRAM prices nearly halved in a little over a year and Micron's stock lost more than half its value from the peak. Another round hit in 2022-2023, when memory makers collectively fell into massive losses and were forced to slash output to defend prices. Before every collapse, the market was telling a story that "demand is structurally different this time" — PCs, then phones, then the cloud, and now AI. The protagonist changes; the shape of the cycle never does.
Today's AI narrative is genuinely stronger and more durable than past ones — I won't deny that. But "stronger" does not mean "will never end." When all three giants are using today's sky-high profits to build out HBM lines, what will the supply side look like in two years? If AI capex growth slows even slightly (a possibility we dissected in another article) while capacity floods in on today's optimistic assumptions, that familiar downward curve will appear again.
Howard Marks: The Pendulum Is Swinging to the "This Time Is Different" Extreme
The first master to summon is Howard Marks of Oaktree Capital. The core of his life's work is the study of "cycles" and "the pendulum." Marks says market sentiment swings like a pendulum, forever oscillating between "excessive optimism" and "excessive pessimism," rarely resting at the rational middle. And the investor's greatest mistake is believing, when the pendulum reaches its peak, that it will stay there forever.
View today's memory market through Marks' cycle lens: orders booked into next year, capacity fully sold out, spot prices up 160% a year, stocks at record highs, every analyst raising price targets, and "AI shovel sellers" being talked about on every street corner — this is a textbook cluster of "pendulum-at-the-top" signals. Marks' famous line: "When everyone believes something carries no risk, the price has usually been pushed to the point of maximum risk."
Marks won't tell you the memory cycle peaks tomorrow — no one can time it precisely. But he will interrogate you with his famous question: "Where do we stand in the cycle?" When an industry famous for its brutal cycles is broadly being priced as a "guaranteed-winner growth story," you must at least recognize this: at your point of entry, both sentiment and valuation are standing at the most dangerous end of the pendulum.
Buffett: Does a Commodity Business Have a Real Moat?
The second master is Buffett, and he would ask one killer question directly: does this business have a moat? Buffett has spent his life favoring companies with pricing power — Coca-Cola can raise prices, See's Candies can raise prices, because customers can't give up their brands. And a truly great business should still make money at the bottom of the economic cycle.
Memory is precisely the opposite. It's a highly standardized commodity: Samsung's DRAM and SK Hynix's DRAM are functionally almost identical to a customer, who buys whichever is cheaper. That means in an oversupply, the three giants can only wage a price war until they all post losses. This is exactly the type of business Buffett detests most — no pricing power, a bottomless capex pit, and earnings that swing wildly with the cycle. He has said explicitly that he avoids these "commodity-type" capital-heavy cyclicals, because their moat is fragile — or nonexistent.
HBM currently does carry higher technical barriers than ordinary DRAM, which gives the three giants temporary excess profits. But Buffett would remind you: a technological lead gets chased down, especially when rivals hold mountains of cash and see the same alluring profits. Today's moat may just be a ditch that the followers haven't yet filled in. Mistaking cyclical excess profits for a permanent moat is one of the most dangerous assumptions in all of valuation.
Graham: Buying at a Low P/E on Peak Earnings Is the Cyclical Stock's Death Trap
The third and coldest master is Benjamin Graham, the father of value investing. He would fix his gaze on one deeply deceptive number: the price-to-earnings (P/E) ratio. On cyclical stocks, the P/E is a thoroughgoing contrarian indicator — when a memory company's P/E looks "cheap" (say, single digits), it is usually precisely because its earnings are at the peak of the cycle, which is exactly the most dangerous moment to buy.
This is the cyclical stock's death trap. At the cycle peak, the company is minting money, the denominator (earnings, E) is inflated to the maximum, so the P/E looks tiny — like an enormous bargain. But you've bought right at the edge of the cliff where earnings are about to fall. When the cycle reverses and earnings halve or turn to losses, that "cheap" P/E instantly becomes astronomical — and the stock has long since collapsed. Graham would tell you that cyclical stocks must be valued on "average earnings smoothed across the entire cycle," not on a single peak year.
This is also the most important application of Graham's margin of safety to cyclicals: the question you must ask is not "how much does it earn today," but "in the next cyclical trough, what is the least it could earn — or how much could it lose," and then demand a sufficient discount on top of that conservative base. Paying a "seemingly reasonable" price on today's peak earnings, in the world of cyclical stocks, is roughly equivalent to walking voluntarily into the slaughterhouse.
What Ordinary Investors Should Do: Peel Off the "Shovel Seller" Label and See the Cycle Underneath
So — can you buy Samsung, SK Hynix or Micron? This article won't give you the answer. But it wants to help you ask the question correctly.
Step one: peel off the "guaranteed-winner AI shovel seller" label. It exists precisely to make you drop your guard. The truth is: you are buying an intensely cyclical commodity business — it's just that this time the demand engine happens to be called AI. Step two: force yourself into the cycle's point of view — use Marks' pendulum to gauge the emotional temperature, use Buffett's moat test to interrogate whether the pricing power is real, and use Graham's full-cycle earnings to value it, rather than being seduced by the low P/E on a single windfall year.
This is exactly the habit of mind Valuefocus is built to give you. When the market wraps a cyclical stock in a catchy label and dresses it up as a growth stock, our social sentiment radar helps you quantify how extreme the euphoria is and whether it has gone too far. Then the master frameworks take over — using Marks' cycle positioning, Buffett's moat and Graham's full-cycle margin of safety to reduce the "shovel seller" story into a cyclical-investing proposition you can rationally test. The label makes you comfortable; the framework keeps you clear-eyed.
Recent posts
%22%2F%3E%3Cpath%20d%3D%22M0%20610%20370%20370l300%20160%20400-330%20530%20360%22%20fill%3D%22none%22%20stroke%3D%22%23c99a62%22%20stroke-width%3D%2218%22%20opacity%3D%22.65%22%2F%3E%3Ccircle%20cx%3D%221070%22%20cy%3D%22200%22%20r%3D%2246%22%20fill%3D%22%23c99a62%22%2F%3E%3C%2Fsvg%3E)
%20-1.jpg?width=1600)
Frequently asked questions
Why are Samsung, SK Hynix and Micron called "AI shovel sellers"?
Because whichever AI company wins the model race, they all need to buy HBM (high-bandwidth memory) from these three memory makers. SK Hynix and Samsung together control roughly 90% of global HBM supply, with Micron close behind, and 2026 capacity is fully booked by AI demand. The market therefore views them as "guaranteed-winner shovel sellers in the gold rush." This article is for educational purposes only and is not investment advice.
Since demand is this strong, are memory stocks a safe bet?
Not necessarily. Memory is the most extreme cyclical commodity industry in semiconductors, and every historical cycle of "price windfall → capacity expansion → glut → price collapse" has ended brutally (e.g., the memory winters of 2018-2019 and 2022-2023). Strong demand doesn't mean the cycle won't end — as the three giants use today's windfall to expand like mad, the oversupply risk two years out is building. They should be valued as cyclicals, not growth stocks.
What framework best evaluates cyclical memory stocks?
Stack three master lenses: Howard Marks on "the pendulum and cycle positioning" — has sentiment swung to a peak of extreme optimism; Buffett on "the moat" — does a commodity business have real pricing power; Graham on "full-cycle earnings" — don't be fooled by a low P/E on peak earnings; value it on earnings smoothed across the whole cycle and demand a margin of safety.
The label makes you comfortable; the framework keeps you clear-eyed. Use Valuefocus to reduce "the AI shovel seller" into a testable cyclical-investing proposition:
Download Valuefocus