Micron has been the single greatest beneficiary of the memory shortage, and the numbers are genuinely difficult to argue with. The May quarter produced $41.5 billion of revenue at an 84.9% gross margin, with the current quarter guided to $50 billion and roughly 86%, which is a level of profitability no memory manufacturer has ever printed and one that would be considered aggressive for a software company. Management has signed sixteen strategic customer agreements carrying around $100 billion of remaining performance obligations, some of them take or pay, and the company can only fill somewhere between half and two thirds of what its core customers are asking for. On the surface this is the strongest fundamental setup in the history of the business.
The market is being told one story about what this means, and I want to lay out both cases so members can judge for themselves rather than accept the consensus framing.
The first case, the one currently embedded in sell side models and in the price, is that the bottleneck persists well past 2029 and that memory has structurally re rated from a commodity into a scarce input. The argument rests on physical constraints rather than sentiment, and it is not a weak argument. Micron's Idaho fab does not produce DRAM until the middle of 2027, the New York fabs only poured first concrete this month and will not contribute meaningfully until 2028, SK Hynix's M15X arrives mid 2027 and Samsung's P5 in 2028. High bandwidth memory consumes roughly three wafers of capacity for every wafer of conventional DDR5 it displaces, so every generational step toward HBM4E and HBM5 tightens the commodity pool rather than loosening it. Total DRAM wafer starts are growing at six to eight percent a year against AI demand growing multiples of that, and industry capacity is forecast to compound at under five percent through 2030.
The second case, the one I hold, is that none of this repeals the cycle, it only postpones it and makes the eventual reversal more violent. Every memory upcycle in history has ended the same way, with the industry earning returns so far above cost of capital that capital floods in, and the capital has already been committed. Micron has lifted its own US programme past $250 billion through 2035, Samsung and SK Hynix have jointly pledged around $518 billion to four new Korean sites and an HBM packaging hub, and CXMT continues to scale in China outside the pricing discipline that has held the western oligopoly together. That supply does not arrive in time to break 2026 or most of 2027, which is precisely why the stock can stay elevated in the near term, but it arrives in a wall across 2028 and 2029 into demand that will by then be growing off a vastly larger base. When it does, margins do not glide lower, they fall off a cliff, because 86% gross margins in a capital intensive business are almost entirely price, not cost. A 20% decline in average selling prices removes the majority of the earnings.
The tell that the market half believes this already is the multiple. Micron is trading below seven times its current quarterly run rate of earnings, which is not the multiple of a business the market believes has permanently escaped its cycle. It is the multiple of a business the market believes is printing peak earnings and will give a large portion of them back. That framing is important, because it means the risk to holders is not that the shortage ends tomorrow. It is that the shortage ends roughly on schedule, earnings normalise from a hundred and twenty dollars of annualised power toward something far lower, and the stock spends two years going nowhere or worse while the multiple that looked cheap turns out to have been correctly cheap all along.
My position is therefore not that Micron is a short today. Near term the contracted revenue, the take or pay structures and the sold out capacity make the next several quarters very difficult to bet against, and the pain of shorting a name in a genuine shortage while it beats every quarter is well documented. My position is that this is a business whose peak is identifiable in advance, that the identifiable trigger is the first quarter in which DRAM average selling prices stop rising, and that retail is being sold a permanence story at exactly the point in the cycle when permanence stories are most dangerous.
Good luck to you all.