The Supercycle

In all of investing, there is no phrase more expensive than "this time is different." It is the sentence that marks the top of every bubble and every cyclical peak, spoken with total conviction, backed by a sophisticated and genuinely plausible argument for why the old rules no longer apply — and it is, with monotonous reliability, wrong, because the forces that create a cycle are laws of economics that no narrative can repeal. The memory-chip industry is the purest cyclical business in all of technology, and in 2026 the market has decided, with the help of an elegant and seductive argument, that its cycle is finally, permanently over. The company at the center of this conviction is Micron Technology, the last great American memory maker, and what has happened to it is a textbook in real time of how a cyclical peak disguises itself as a permanent plateau.

The numbers driving the euphoria are genuinely staggering. In its fiscal second quarter of 2026, Micron reported revenue of nearly $24 billion, up an almost unbelievable 196% year over year and 75% sequentially; its gross margin reached 75%, with guidance pointing toward the low 80s — margins more associated with software than with manufacturing physical chips; and its earnings exploded past every estimate. The engine of this boom is high-bandwidth memory, or HBM — the specialized, stacked memory that sits beside AI accelerators and that the artificial-intelligence build-out consumes in vast quantities. Demand for HBM has been so intense that Micron's entire 2026 capacity sold out, the HBM market is on track to roughly double to something like $62 billion in a single year, and Micron is shipping its newest generation, HBM4, into Nvidia's next-generation AI platform. On the strength of all this, Micron's stock surged — a single session saw it jump 19% — and in late May 2026 it crossed $1 trillion in market value, an almost inconceivable milestone for a company that, not many years ago, was a boom-and-bust commodity manufacturer worth a fraction of that.

Thirty years of the same movie

To understand why this should make a careful investor deeply wary rather than euphoric, you have to understand the history of the business, because the history is the whole point. For three decades, the memory-chip industry has followed a pattern so reliable it is almost a metronome. A shortage develops; prices spike; the makers — Micron, Samsung, SK Hynix — earn enormous profits; flush with cash and confidence, they pour billions into building new fabrication plants to capture the bonanza; those plants come online about two years later, all at roughly the same time, flooding the market with new supply; the supply overwhelms demand; prices collapse, often by half or more; the makers swing to losses and bleed red ink; capital spending is slashed; the excess capacity is eventually absorbed; a new shortage develops; and the entire cycle begins again. This is not a vague tendency; it is the defining, documented rhythm of the industry. DRAM prices peaked in late 1995 and then fell 51% in 1996 and another 65% in 1997. The same pattern, with different dates, has repeated through every decade since. Memory is the closest thing technology has to a pure commodity, and commodities are cyclical because high prices are self-correcting: they call forth the supply that destroys them.

Micron's stock has, accordingly, been one of the most violently cyclical large-cap stocks in the market for its entire history — soaring in the upcycles, collapsing in the downturns, a roller-coaster that has buried every investor who mistook a peak for a permanent state. Anyone who has watched the company for more than a few years has seen this movie multiple times, and it ends the same way every time: the euphoria of the shortage gives way to the despair of the glut, and the stock that looked like a one-way bet at the top gives back most or all of its gains on the way down. The current boom is, by the raw numbers, the most spectacular upcycle in the company's history. The question that matters is whether it is also, as the bulls now insist, the last one — whether the cycle that has defined this business for thirty years has genuinely been broken, or whether it has merely, as it always does at the top, become temporarily invisible.

The bust was not in 1997 — it was the day before yesterday

The most powerful rebuttal to "the cycle is dead" is not the crash of 1996 or any other distant episode; it is how recent the last bust was. One does not have to reach back to the 1990s to find Micron bleeding red ink — one has only to look back about two years. In the memory downturn of 2022 and 2023, demand softened, supply that had been built in the prior boom came online, prices collapsed, and Micron's business fell off a cliff: its revenue cratered, it posted a net loss of roughly $5.8 billion for fiscal 2023, and its gross margin — the very metric now celebrated at 75% — turned negative, sinking to around minus 9% for the year and below minus 30% in one brutal quarter, meaning the company was literally selling memory for far less than it cost to make. The stock fell heavily, as it always does in the downturns, and analysts who had been bullish at the prior peak turned bearish at the trough, as they always do. This was not ancient history; it was the most recent chapter of the same book, and it ended, as every chapter does, with the cycle turning.

That is the fact the "death of cyclicality" thesis most needs you to forget: that the same company now valued at a trillion dollars on the conviction that memory has become permanently scarce was, within the span of a normal business cycle, losing money because memory was temporarily abundant. The people declaring the cycle over are, in many cases, the same people who lived through Micron's losses two years ago — and who, two years before that, were similarly convinced of a different narrative. Memory's cycle has not lengthened to the point of disappearing; it has, if anything, continued to turn at roughly its historical frequency, boom to bust to boom, and the current peak is separated from the last trough by an utterly ordinary span of time. To look at a business that posted negative gross margins twenty-odd months ago and conclude that it will sustain 75% margins indefinitely is to ignore not the distant past but the recent one. The bust is not a historical curiosity. It is a recurring event, and the last one is still warm.

The argument for the death of cyclicality

It is worth stating the bull case in its strongest form, because it is not stupid, and a lazy contrarian who dismisses it will get hurt. The "supercycle" thesis runs like this. In past cycles, the bust came because the makers added too much capacity — but this time, the argument goes, AI demand is growing faster than new fabs can possibly be built, so supply cannot catch up to demand within the normal two-year lag; the shortage is structural, not transient. Moreover, the industry has consolidated to just three major suppliers, an oligopoly that has learned, after decades of mutually destructive boom-bust, to exercise discipline — to hold supply tight and prices high rather than flood the market and trigger the next crash; the three have signaled exactly this intention. And HBM specifically, the bulls note, is not a simple commodity like ordinary DRAM: it is technically demanding, hard to manufacture, sold out, allocated to customers under long agreements, and advancing through generations (HBM3E to HBM4) that command premium pricing. Add it up — structural demand outrunning supply, supplier discipline, and a differentiated premium product — and you have, the thesis concludes, a memory business that has escaped its cyclical curse and earned a permanent re-rating. Tight conditions, the bulls project, will persist into 2027 and 2028.

Each of these points has genuine merit. AI demand for memory is enormous and is outrunning near-term supply; the industry has consolidated; HBM is a better, stickier, higher-margin product than commodity DRAM. The supercycle is real, the boom is real, and Micron is a genuinely excellent company executing superbly at the center of it. None of that is in dispute. What is in dispute is whether these factors have abolished the cycle or merely stretched it — whether "this time is different" is, for the first time in the history of cyclical industries, actually true.

Why the seeds of the bust are planted at the peak

Here is the problem the euphoria glosses, and it is the iron logic that has ended every memory cycle: the very conditions that define the peak are the conditions that cause the bust. Why does the memory industry over-build capacity and crash prices? Because at the peak, margins are fat — 75%, in Micron's case — which gives the makers both the cash to fund new fabs and the incentive, since each one wants to capture more of those gorgeous margins, and the fear that if it doesn't add capacity, its rivals will and steal the share. Fat peak margins are not a sign that the cycle is over; they are the fuel for the capex binge that ends it. Micron's 75% gross margins and trillion-dollar valuation are not evidence of a new permanent plateau; they are precisely the signal that has, every prior time, triggered the industry-wide investment surge that floods the market two years later. The boom contains its own ending, and the brighter the boom, the more powerful the incentive to build the capacity that will end it.

And "supplier discipline" — the linchpin of the supercycle thesis — is exactly what every oligopoly promises at the top and abandons under competitive pressure. The three memory makers may say they will hold supply tight, and they may even mean it today. But the temptation to grab share with new capacity, when margins are this fat and a rival might move first, has broken supplier discipline in every prior cycle, because the incentives of an oligopoly are unstable: collective restraint maximizes industry profit, but individual expansion maximizes any single player's profit, and the individual incentive eventually wins. The history of the memory industry is, in large part, the history of supplier discipline being promised at the peak and shattered on the way to the glut. Betting that this time the three suppliers will hold the line, forever, against the temptation of the richest margins in the industry's history, is betting against thirty years of their own behavior. Add to this that the demand side — AI accelerator purchases — is the same circular, capex-driven, forecast-dependent spending questioned throughout this series, which means the "structural" demand that supposedly outruns supply is itself riding the AI capex wave that the grid operators and others have already begun to trim. If that demand merely normalizes while the new HBM capacity the boom is funding comes online, the sold-out market becomes an ample one, and the supercycle becomes, once again, a cycle.

The cyclical's cruelest trick: cheap at the top

There is a specific valuation trap that makes cyclical stocks uniquely dangerous, and Micron is sitting squarely in it. A cyclical company looks cheapest precisely at the top of its cycle — because at the peak, earnings are at their maximum, so the price-to-earnings ratio looks low, luring in value investors who think they are buying a bargain. "Micron at only a handful of times earnings!" the bulls exclaim — but those earnings are peak earnings, inflated by 75% margins and tripled revenue that the cycle's own logic says will mean-revert. When the cycle turns and earnings collapse, the "cheap" stock becomes expensive on the suddenly-shrunken earnings, and the price falls to match. The correct way to value a cyclical is never on peak earnings but on mid-cycle or trough earnings — on what the company makes across the full cycle, including the busts — and on that basis a memory maker at a trillion-dollar valuation, on the most extreme peak earnings in its history, is not cheap at all. It is, in fact, the textbook setup for the cyclical trap: a low headline multiple on earnings that are about to fall, dressed up by a "this time is different" narrative whose entire function is to convince you that the peak earnings are permanent so that you will pay a permanent price for them.

The "death of cyclicality" thesis, in other words, is not an incidental piece of optimism; it is load-bearing for the valuation. The only way to justify a trillion dollars for Micron is to believe that the current earnings are not a peak but a new baseline — that the bust will not come, that the margins will hold, that the supercycle is structural. Strip away that belief and value the company on a normal cyclical basis, across booms and busts, and the trillion-dollar price makes no sense. So the entire valuation rests on the proposition that the most cyclical business in technology has stopped being cyclical — the proposition that, in every prior instance across every cyclical industry, has been the marker of the top rather than the dawn of a new era. The supercycle narrative and the trillion-dollar valuation are the same thing: a bet that the cycle is dead, sold at the price you would only pay if it were.

This too will pass

Perhaps the most fitting last word belongs to a Harvard Business School professor who has tracked semiconductor cycles since the 1980s, and who, watching the 2026 AI memory boom, told Fortune that it "looks like every other memory cycle" he has ever seen — "just bigger." "Anytime people show me these curves that just go to the sky with no end," he said, "that never continues forever." His verdict on the supercycle was three words long, and they are the three words that the entire trillion-dollar edifice is built to deny: this too will pass. He has seen the "this time is different" argument before, in every prior peak, each time dressed in the specific and plausible logic of its moment — and each time it has given way to the same gravity that governs all commodity cycles, the gravity by which high prices summon the supply that destroys them.

None of this is a prediction that Micron crashes tomorrow, or that the supercycle ends this quarter, or that the company is anything other than excellent. The upcycle may well run longer and higher than the skeptics expect; HBM genuinely is a better business than commodity DRAM; the AI demand is real and may stay strong for years; and the bears who called the top too early will, as they always do in the late innings of a boom, look foolish for a while. Being early is, in markets, indistinguishable from being wrong, right up until it isn't. But the forensic point does not depend on timing the turn. It is simply this: that the market has paid a permanent price — a trillion dollars, on peak earnings — for a cyclical business, on the strength of a "this time is different" narrative that the history of every cyclical industry, and of this one in particular, marks as the reliable signature of a top. Memory has buried every investor who believed the cycle was dead, for thirty years, with brutal regularity. The cycle is not dead now. It is merely, as it always is at the peak, temporarily invisible — hidden behind sold-out capacity and 75% margins and an elegant argument, in exactly the way it has hidden at the top of every prior boom before reasserting itself with the force of an economic law. The curves are going to the sky with no end. They never continue forever. This too will pass — and the most dangerous moment to own the most cyclical business in technology is the moment the whole market has agreed that it is no longer cyclical at all.

And so the last of these stories closes where the first ones began, on the single lesson that binds them: that the most dangerous ideas in markets are not the obvious lies but the seductive truths carried one step too far. AI's hunger for memory is real; Micron is excellent; the shortage is genuine; the boom has substance. Every word of the bull case is, as far as it goes, true. The error is not in the facts but in the extrapolation — in taking a real, cyclical, finite boom and pricing it as a permanent, structural, infinite one, because the story is good enough to make the overpayment feel like prudence. That is the trap that runs through every chapter here, from the AI chips to the trillion-dollar IPOs to the crypto towers to the nuclear moonshots, and it finds, in a memory stock at a trillion dollars on peak earnings, perhaps its cleanest expression. Being right about the boom and right about the price were never the same thing. The cycle is the proof, and the cycle always, eventually, comes back to collect.