Micron: an 84.6% gross margin and the contracts that promise to end the cycle

A deep review of Micron Technology (Nasdaq: MU), the only American maker of DRAM and one of just three companies producing it at industrial scale anywhere. Micron has just reported a quarter with no precedent in its history: a gross margin of 84.6%, against a previous record of 61.0% set in 2018. The stock is up elevenfold in two years. What follows is what actually happened, what drove it, how durable the company's competitive position is, and what has to be true for the shares not to be a textbook trap at the top of a cycle.

The full financial history is on our company page.

A quarter without precedent

Micron's third fiscal quarter of 2026 ended on 28 May. Revenue came in at $41.5bn against $9.3bn a year earlier, a 4.5-fold increase. Gross profit was $35.1bn, operating profit $33.3bn (80.4% of revenue) and net profit $28.2bn, or $24.67 per diluted share versus $1.68 a year ago.

For the first nine months of the year revenue reached $79.0bn and net profit $47.3bn.

Guidance for the fourth quarter, which closed at the end of August: revenue of $50bn plus or minus $1bn, gross margin around 86%, earnings of $30.73 per share. The report lands on 30 September. That is the nearest point at which every assumption in this review gets tested.

For scale. Gross margin is the share of revenue left after the direct cost of production. For a memory chipmaker it has historically swung from minus 30% at the bottom of a cycle to 55-60% at the top. Micron has visited both edges in three years.

All of the growth came from price, not from volume

This is the single most important observation in the report, and the company does not hide it. The 10-Q breaks down where the year-on-year sales growth came from:

– DRAM (the main memory that accounts for the bulk of the business) - revenue up 343%, of which average selling prices up roughly 260% while bit shipments rose only about 20%

– NAND (flash memory used in storage) - revenue up 361%, with prices up around 310% and volumes up by a low double-digit percentage.

Cost of goods sold rose from $5.8bn to $6.4bn over the year, or 10%, while revenue multiplied by 4.5. Micron is selling roughly the same bits it sold a year ago, at 3.5 times the price. That is where the 84.6% comes from: every increment of price drops into gross profit whole.

The practical conclusion. This is not a story about operational improvement or a new product that displaced competitors. It is a price shock in an industry where Micron happens to be one of three recipients of the rent. So the thing to assess is not execution quality but the durability of the shortage itself.

The breadth of the move shows up in the segments. All four business units now run operating margins above 75%, including automotive and embedded, which has nothing to do with artificial intelligence. When server memory and memory for household appliances earn equally well, that is a sign of shortage, not of product advantage.

The shortage is real, and there is no quick fix

An independent read on the industry from research house TrendForce (July-August 2026): the DRAM sufficiency ratio, which measures how far supply covers demand, stands at minus 1-2% in 2026, and the gap is expected to widen in 2027. The DRAM market is put at $618.7bn in 2026, a fourfold increase, and $903.3bn in 2027, up 46%. The whole memory market is forecast to exceed $1.28tn in 2027.

The reason the shortage cannot clear inside a year is arithmetic. HBM is high-bandwidth memory, the kind stacked next to AI accelerators. In 2026 it will account for only 9% of DRAM bits but consume 22% of the industry's silicon wafers; by the end of 2027 that becomes 13% of bits and 30% of wafers. HBM is assembled in multi-layer stacks, and each usable bit takes several times more silicon area than a conventional memory module. Every HBM bit a manufacturer sells is subtracted from the supply of ordinary memory. That is why prices have risen for server and consumer modules that have no direct connection to AI at all.

New plants do not help inside the forecast horizon: on TrendForce's estimate, meaningful output from new capacity will not arrive before 2028. Micron's own record shows the timescale. Ground was broken on the Boise, Idaho fab in September 2022, construction started in October 2023, and the first DRAM wafers are expected in mid-2027. Five years from shovel to product.

The real news is not the margin, it is the contracts with a price floor

In the third and fourth quarters Micron began signing what it calls strategic customer agreements. The terms are set out in the 10-Q, and they change the shape of the business more than the record margin does:

– they are take-or-pay, meaning the customer either lifts the agreed volume or pays for it anyway; volume commitments are locked in for several years

– for most agreements the price is either fixed or bounded by a floor and a ceiling

– for the largest agreements the ceiling approximates the market price of the second calendar quarter of 2026, and the floor holds for the whole term

– in the company's own words, gross margins under these contracts even at floor pricing will run well above the peak quarterly margin of any past cycle, which is to say above 61%

– a minority of the agreements carry no price bands and follow the market

– Micron expects $22bn of deposits and related financial commitments from agreements concluded so far, of which roughly $18bn arrives as cash.

What this means in substance. A memory maker has always traded cheaply on peak earnings precisely because peak earnings said nothing about next year. If the contracts work as described, a significant part of revenue for several years stops being a function of the spot price and starts to resemble contracted revenue with a guaranteed minimum margin. The $18bn of deposits also amounts to free financing for the build-out: customers are prepaying for the very capacity they have signed up for.

The other side sits in the same paragraph. A ceiling near the second-quarter 2026 level means further increases in spot prices will largely bypass Micron. The company has traded the upper tail of the distribution for the lower one.

That is the most plausible explanation for the share price. From a high of $1,213 on 25 June the stock has drifted back to $1,017, down 16%, even though the reported numbers have only improved since. The market has worked out that the option on higher prices is now closed.

Strong in DRAM, a distant third in HBM

Two markets are often conflated here and need separating. Conventional DRAM is a three-way oligopoly in which Micron is firmly established. HBM is a narrower and far more profitable segment, and the line-up there is different.

On Counterpoint Research data for the second quarter of 2026, in DRAM Samsung holds 38%, SK hynix 25% and Micron 24%, with China's CXMT at 10% and Taiwan's Nanya at 2%. Micron sits third, close behind SK hynix, and Counterpoint notes separately that Micron's DRAM revenue has risen fivefold in a year and that the company could take second place.

In HBM the picture is worse: SK hynix 50%, Samsung 33%, Micron 18%. SK hynix has lost 14 percentage points in that segment over the year, down from 64%, and Samsung was first into volume production of HBM4 in February 2026. In the most valuable part of the market Micron is the one catching up, and the competition there is real rather than nominal.

What is genuinely solid:

– Technology. HBM4 on the 1-beta process is already shipping in volume to the lead customer, with qualification samples out to several end customers; HBM4E on the next 1-gamma process is scheduled for volume production in 2027. Alongside that: 256GB DDR5 modules, LP5X SOCAMM2 in high-volume production, PCIe Gen6 drives and a 245TB SSD.

– The barrier to entry is money and time. Five years from groundbreaking to first wafer, and $27bn of capital expenditure in fiscal 2026 alone. That explains why the three incumbents are in no hurry to fight for share: Micron states plainly in its filing that its objective is to maintain stable bit share, not to grow it.

– Less dependence on a single buyer. The largest customer accounted for 10% of revenue in the first nine months of 2026, down from 16% a year earlier.

– Government support. Up to $6.4bn in grants under the US CHIPS Act, a 35% investment tax credit on US manufacturing investment, and a non-binding term sheet with New York State worth up to $5.5bn. In a world where memory has become a strategic commodity, the only American DRAM producer gets an extra layer of protection.

What could break the story

China. The main threat to the structure of the market is CXMT. On Counterpoint data its share of DRAM revenue has gone from 4% to 10% in a year, with revenue up 716%, and for the first time the combined share of the top three has fallen below 90%. The technology gap, once put at five years or more, has narrowed to two or three. In July 2026 the company raised $8.6bn in an IPO and is weighing a sixth fab; capacity of roughly 300,000 wafers a month is planned to reach 550,000-600,000 by 2028, comparable to SK hynix. For reference, Micron's own capacity at the end of 2026 is estimated at around 385,000 wafers a month (SemiAnalysis). CXMT plans volume production of HBM3E as soon as this year.

NAND is diverging from DRAM. TrendForce expects flash supply to turn loose in the second half of 2027 as new capacity comes online while consumer electronics demand stays weak. Micron's NAND prices are up 310% over the year, and that is the part of the business where the turn arrives first.

Micron's own capacity wave. The company is simultaneously building a second Idaho fab (output late 2028), a site in Clay, New York (supply from 2030), HBM packaging in Singapore from the first half of 2027 plus an additional fab in the second half of 2028, and in Taiwan the Tongluo fab bought from Powerchip for $1.8bn with shipments from mid-2027. The competitors are doing the same. Discipline in the memory industry has always held right up to the point where everyone starts building at once.

A competitor's inventory. Samsung's inventory has risen from 87 to 124 days of cost of sales in three quarters. Micron and SK hynix are stable around 122-123 days. On its own this is not a signal that the cycle has turned, since Samsung may be building for contracted shipments. But it is usually the first metric to break before prices do.

Tax. The effective rate has risen from 11.1% to 15.0% because of the Pillar Two global minimum, which in Singapore has offset the incentives Micron used to enjoy.

Our cycle indicator is pinned at its maximum, which historically is a bad sign a year out

We maintain our own composite indicator of the memory cycle. It pulls in monthly revenue from Taiwan's Nanya and Winbond, DRAM spot prices, and the trend in margins, inventory and capital spending at Micron and SK hynix; each component scores plus one, zero or minus one. The composite currently reads plus 5 out of 5 active signals, the maximum. Nanya's three-month revenue is up 50% on the prior three months and sevenfold year on year, Micron's margin gained 10 percentage points in a quarter, capex to revenue sits below its three-year median, and SK hynix's inventory days fell by 11.

A backtest over 2006-2026 produces an uncomfortable result. With a positive composite, the following six months are genuinely better than average for Micron shares: a median of plus 11.6% against plus 5.3% after negative readings. But over twelve months the sign flips: a median of plus 3.8% after strong readings against plus 18.7% after weak ones.

Put plainly, strong fundamental readings confirm a turn that has already happened rather than predict the next one. A maximum composite describes the present; it is not a forecast.

Valuation: 8 or 23 times earnings, and the gap rests on a single assumption

At $1,016.59 a share Micron is capitalised at roughly $1.15tn. For the first time in its history the company holds a net cash position: $30.1bn of cash and investments against $5.7bn of debt, cut from $14.6bn over nine months. Enterprise value is about $1.12tn.

From there everything depends on which earnings you treat as normal:

– on the last twelve months of profit ($50.5bn) the stock trades at 23.1 times earnings and 16.4 times EBITDA; adjusting for stock-based compensation ($954m over nine months), EV/EBITDAC is about 16.8

– on full fiscal 2026 earnings including fourth-quarter guidance ($72.13 per share) the multiple is 14.1

– on annualised fourth-quarter guided earnings it is 8.3 times, with EV/EBITDA on the annualised latest reported quarter at about 7.9.

Twenty-three times earnings is expensive for a cyclical company. Eight is cheap even for a stable one. The distance between those two numbers is the whole argument about Micron, and it reduces to one question: how many years can a margin above 60% last. If the floor-price contracts work as the filings describe, then three or four years of earnings on this order are already contracted and the right anchor is closer to eight. If the contracts cover a smaller share of volume than the wording implies, the story reverts to an ordinary cycle.

Capital returned to shareholders is meanwhile token. The dividend is $0.15 a quarter, a yield of 0.06%, and $7.84bn of the authorised $10bn buyback has been used over the life of the programme. The money is going into construction: $27bn of capex in fiscal 2026. An investor in Micron is paying not for distributions but for a claim on future capacity.

An earnings-quality check: receivables tripled, but not in days

With growth like this, the first thing to check is whether the profit has been manufactured out of shipments that will never be paid for. Micron's receivables tripled, from $9.3bn to $31.0bn. But measured in days of revenue that is 68 days against 75 a year earlier, so collection has improved, not deteriorated. Inventory is steady at $8.6bn, or 122 days of cost of sales.

Cash flow corroborates the profit. Over nine months operating cash flow was $45.7bn against net profit of $47.3bn, capital expenditure was $19.6bn and free cash flow roughly $26bn. On the numbers there is nothing to fault in earnings quality.

Conclusion

Micron today is not a bet on the memory cycle, nor a pure bet on artificial intelligence. It is a bet that multi-year contracts with a price floor will survive the next downturn. Everything else in the story has already happened: prices rose, margins hit a record, the stock went up elevenfold in two years, and the contractual ceiling has been set.

For: the DRAM shortage is independently confirmed and structurally unfixable before 2028; the contract floor is, on the company's own account, above any previous peak; net cash for the first time in years; a valuation of 8 to 14 times earnings; and earnings quality that survives inspection.

Against: all of the growth came from price rather than volume; Micron is third in HBM with 18%; CXMT has multiplied its share by 2.5 in a year and closed the technology gap to two or three years; NAND is forecast to turn as early as the second half of 2027; and our cycle indicator is at its maximum, after which twelve month returns have historically been weak.

What to watch. The main event is the fourth-quarter report on 30 September, and with it any disclosure of how much 2027-2028 volume is already under contract. That is the single number that settles the central argument. After that come Samsung's and SK hynix's inventories in the next quarterly filings, and DRAM spot prices. Until those start falling, talk of a turn is premature.

The full Micron page with reporting history and multiples is on the Enhanced Investments portal. Company deep dives, weekly reviews and investment ideas are published in our Telegram channel Enhanced Investments.