The MoatNarrow moat

The most important number about Micron is a strange one: at a market value of about $1.24 trillion it trades at about 25 times trailing earnings and 7.5 times the earnings expected over the next year¹. A low multiple on record profits looks like a bargain the market overlooked. It is the opposite. Micron makes memory — the DRAM and NAND chips that store and feed data in every phone, car, PC, and above all every AI server — and memory is the most violently cyclical business in all of technology. Here, a low price-to-earnings ratio is not cheapness; it is the market's warning that these are peak-cycle earnings it expects to fall. Everything about Micron flows from that single tension: a business with a real but genuinely narrow moat, earning the greatest profits in its history, priced by a market that has seen this movie before and knows how it usually ends.

That Micron has a moat to discuss at all is the achievement, because for most of its history memory was the worst business in semiconductors — dozens of makers producing near-identical chips, competing on price alone, racing one another into ruinous busts that destroyed capital across the cycle. What changed was consolidation. Decades of attrition culled the DRAM industry to three survivors — Micron, Samsung, and SK Hynix — who together took 89.7% of DRAM revenue in the first quarter of 2026² and, scarred by the old price wars, have mostly learned to compete on technology and discipline rather than suicidal capacity races. That is the moat: not a cartel, but a commodity market finally rational enough to earn money across a full cycle instead of only at its peaks. The caveat is stitched into the same fact — NAND flash, the other half of Micron's business, has five or six players and far weaker economics, so even the structural moat is only partial.

The reason the stock re-rated so violently is a single product: high-bandwidth memory. HBM is DRAM stacked into dense towers that sit beside Nvidia's and AMD's AI accelerators and feed them data, and because an AI chip's speed is throttled by how fast memory can supply it, HBM has become one of the scarcest, most sought-after products in technology. What makes it matter for the moat is that it behaves unlike ordinary memory: it is far harder to manufacture, it is sold on long-term contracts rather than volatile spot pricing, and it carries margins from another planet. For this cycle at least, HBM has broken Micron partway out of the commodity trap — and it is the entire engine behind the historic results.

Underneath both the oligopoly and HBM sits the barrier that makes them durable: memory is now as hard and as expensive to make as leading-edge logic. A single frontier fab costs well over ten billion dollars³, the process pushes the same physics as the most advanced processors, and the manufacturing know-how is a learning curve accumulated over decades that money alone cannot buy. That wall is why the field cannot grow past three, and why a well-funded newcomer cannot simply appear. But the same wall is a weight Micron itself must carry — tens of billions in capex every year, committed into booms and exposed in busts — and Micron is the smallest of the three, the most resource-constrained, spending the most relative to its size just to stay at the frontier. Its one genuinely distinctive asset is its address: it is the only major memory maker headquartered in the United States, which in an age of chip sovereignty makes it America's memory champion and the beneficiary of CHIPS Act subsidies for megafabs in New York and Idaho — a favor that cuts both ways, since the same status costs it access to China, the world's largest memory market, where it has been targeted for retaliation.

The scale of the boom demands to be stated plainly, because it is almost hard to believe. After fiscal 2025 revenue of $37 billion, Micron's quarterly revenue roughly doubled twice in a row, reaching a record $41 billion in the May 2026 quarter⁴, as gross margins exploded from around 45% toward an almost unheard-of 85%. Trailing revenue reached some $90 billion⁵ and net income around $50 billion. As recently as fiscal 2023, in a down-cycle, this same company lost nearly six billion dollars⁶.

So the verdict is narrow, and honest about it. Micron is a genuine oligopolist with real technology, capital, and now HBM advantages, welded to the most cyclical business in technology, at the very top of the greatest boom it has ever seen. The whole question — the one the nineteen-times multiple is asking — is whether AI has permanently raised memory's floor or whether this is simply the grandest cyclical peak yet before the inevitable reversion. The number to watch is not revenue or even margin at the peak, both of which the boom guarantees for now; it is gross margin through the next downturn. If HBM and oligopoly discipline hold margins far above the near-zero or negative lows memory has always collapsed to in a bust, the moat was real and the floor did rise; if they crater the way memory margins always have, this was a cycle and not a transformation — and the low multiple will have been right all along.