Analysis · Semiconductors · August 17, 2026
Micron’s Revenue Rose 346%. Its Cost of Goods Sold Rose 10%.
Micron’s fiscal third quarter, in millions of U.S. dollars. Revenue multiplied by 4.5× year over year. The cost of producing that revenue barely moved — which is the entire story of the quarter, and of the gross margin it produced.
Fiscal Q3 2026 versus fiscal Q2 2026. Micron discloses these as qualitative ranges rather than point estimates, so they are drawn here as ranges — the solid bar is the low end, the lighter extension the high end of the stated band.
Each line runs from last year’s operating margin to this year’s. Mobile & Client — phones and PCs, not data centres — travelled the furthest, from 14.8% to 85.7%.
Micron reported $41.5 billion of revenue in its fiscal third quarter, against $9.3 billion a year earlier. Cost of goods sold over the same span went from $5.79 billion to $6.40 billion. A memory manufacturer — historically the most brutally cyclical, most commoditised business in semiconductors — posted an 84.6% gross margin and an 80.4% operating margin. Those are software margins, produced by a company that runs fabs.
The obvious question is whether this is a demand story or a price story. The filing answers it directly, and the answer is almost entirely price.
1. Price did the work, and the company says so
The most useful disclosure in the filing is the sequential comparison — fiscal Q3 against fiscal Q2, only three months apart, which strips out most of the base-effect distortion in the year-over-year numbers. Micron attributes the change in its own words:
“Sales of DRAM products increased 67%, primarily due to a low-60% range increase in average selling prices and a low-single-digit percentage range increase in bit shipments. Sales of NAND products increased 99%, primarily due to a mid-80% range increase in average selling prices and a mid-single-digit percentage range increase in bit shipments.”
Read that carefully. Micron shipped a few percent more bits than it did three months earlier. It charged somewhere between 60% and 87% more for them. Essentially the entire revenue increase is price.
The balance sheet corroborates it. Inventories were $8.57 billion at quarter end against $8.36 billion at the August 2025 fiscal year end — up 2.5% while quarterly revenue more than quadrupled. Micron is not building stock and is not drawing it down; it is selling roughly what it makes, at dramatically higher prices.
2. Every segment repriced, but not equally
Micron reports four business units. All four saw operating margins expand violently, but the starting points differed enormously — and the segment that moved most is the one furthest from the AI data centre.
Mobile & Client is the important one. If this were purely an AI-datacentre memory story, the expansion would concentrate in Cloud Memory and Core Data Center. Instead the segment serving phones and PCs went from a 14.8% operating margin to 85.7% — the largest move of the four. Shortage pricing is reaching every end market that buys the same bits, not just the ones with AI in the name.
| Measure | FQ3 2026 | FQ3 2025 | Change |
|---|---|---|---|
| Revenue | 41,456 | 9,301 | +346% |
| — DRAM | 31,328 | 7,071 | +343% |
| — NAND | 9,943 | 2,155 | +361% |
| Cost of goods sold | 6,400 | 5,793 | +10.5% |
| Gross margin | 35,056 | 3,508 | +899% |
| Gross margin % | 84.6% | 37.7% | +46.9pt |
| Operating income | 33,318 | 2,169 | +1,436% |
| Operating margin % | 80.4% | 23.3% | +57.1pt |
| Inventories (vs. FY-end Aug 2025) | 8,567 | 8,355 | +2.5% |
| Receivables (vs. FY-end Aug 2025) | 31,025 | 9,265 | +235% |
3. The contracts are the interesting part
Buried in the revenue note is a disclosure about how Micron is trying to hold on to this. The company says it recently executed “strategic customer agreements” structured as take-or-pay arrangements, with binding commitments for specific volumes over multi-year terms and contractually enforceable volumes. Pricing on most is either fixed or bounded by minimum and maximum price bands. Management adds that it expects gross margins from those agreements, “even at floor pricing levels, to yield gross margins well above our peak quarterly margins in any past cycle.”
That is a claim about the shape of the next downcycle, not this upcycle. If it holds, it changes what a memory company is worth, because the historical discount applied to memory equities exists precisely because peak margins have always been temporary.
There is a caveat attached, and Micron states it plainly. The company’s disclosed remaining performance obligations were approximately $5 billion — about one-eighth of a single quarter’s current revenue — and Micron says that figure “is based on minimum committed volumes and minimum pricing and is not expected to be indicative of future revenue under these contracts.” Agreements without fixed pricing or price bands are excluded entirely.
Two RPO disclosures, opposite conservatism
This is worth putting next to the CoreWeave and Nebius backlog I looked at earlier today. CoreWeave discloses $103.7 billion of RPO stated net of estimated variable consideration — a number that embeds management’s estimates of credits and delivery delays, and could be revised down. Micron discloses $5 billion stated at minimum committed volumes and minimum pricing — a number designed to be a floor, which the company explicitly says will understate actual revenue. Both are RPO. Both follow the same standard. They are pointing in opposite directions, and reading either headline number without the accompanying sentence gets you the wrong answer.
4. The case for the other side
Price cycles mean-revert — always have
An 84.6% gross margin in a commodity manufacturing business is, on every historical precedent, a peak rather than a plateau. Supply responds to prices like these. The bear case does not require demand to fall; it only requires competitors to finish building.
Flat inventories cut both ways
Selling everything you make is a sign of genuine shortage. It is also the condition under which added supply has the fastest effect on price, because there is no buffer absorbing it. The same fact supports both readings.
But the contract structure is new
Multi-year take-or-pay agreements with price floors are not how memory has historically been sold. If a meaningful share of volume sits under those terms, the next downcycle’s trough margin genuinely could be higher than past troughs — which is Micron’s explicit claim and the thing most worth verifying.
5. What this means for the book
This is the largest single-theme exposure in the portfolio, and it is worth being blunt about the size.
| Position | Weight | Return | Conviction |
|---|---|---|---|
| Sandisk (SNDK) — NAND | 7.94% | +151.41% | Medium · Under Review |
| Micron (MU) — DRAM and NAND | 5.38% | +152.59% | Medium |
| Combined | 13.32% | — | — |
Both positions have roughly tripled from cost, and this filing explains why: the underlying commodity repriced by 60–87% in a single quarter. That is the honest attribution. Neither return is primarily the result of my having identified a great business — it is the result of holding exposure to a commodity during a violent shortage.
Micron’s NAND ASP disclosure is also the most direct read-through I have to SanDisk, a pure-play NAND holding at nearly 8% of the book and my largest position. A mid-80% range NAND price increase in a single quarter is the mechanism behind that position’s gain. It is equally the mechanism that would work in reverse.
What I take from this: two positions totalling 13.3% of the book are levered to the same variable, and that variable has moved almost vertically. That is a concentration observation, not a sell signal — the sort of thing The Concentration Premium argues HHI alone will not show you, since Micron and SanDisk are different tickers in different weights that happen to be the same bet.
What I am not claiming
I am not forecasting the memory cycle, and I have not modelled industry supply additions, wafer capacity coming online, HBM allocation, or the specific terms of Micron’s strategic agreements — all of which determine what happens next and none of which I have done the work on. I am also not claiming these margins are unsustainable; I am claiming they are historically unprecedented for this industry, which is a statement about the past rather than a prediction. The narrow, checkable claim is that this quarter’s result was produced by price rather than volume, and Micron’s own sequential disclosure says so.
What would change my read
WatchThree things, in order of how much they would move me
1. Bit shipments versus ASP in the next sequential comparison. This is the single most informative line in the filing. If bits accelerate while ASPs flatten, the cycle is maturing healthily. If ASPs fall while bits stay flat, the peak has passed. 2. Inventories. Flat inventories with rising prices is a shortage. Rising inventories with flat prices is the turn, and it usually shows up on the balance sheet before it shows up in the margin. 3. Growth in disclosed remaining performance obligations. Because Micron states RPO at minimum volumes and minimum pricing, growth in that figure is a relatively clean signal of how much revenue is being moved under contractual floors — which is the whole question of whether the next trough is different.
The memory industry’s defining characteristic has always been that its best quarters carry no information about its next ones. Micron just posted a quarter with an 84.6% gross margin and told shareholders, in the revenue note, that it is signing multi-year contracts with price floors designed to make the next downturn less severe than every previous one. Whether that is true is the entire investment question. What is not in doubt is where this quarter’s money came from: not from selling more, but from charging more.