Micron

IThe business

Micron makes memory. DRAM is the fast memory a processor works from, and NAND is the flash that stores data when the power is off. Micron designs them, builds them in its own fabs in Taiwan, Singapore, Japan and the United States, and sells them to the makers of servers, phones, PCs and cars.² It does not use a foundry for its wafers. That one fact shapes everything on this page: Micron owns the factories, so when prices fall it still has the factories.

Fiscal years end on the Thursday closest to 31 August. FY2026 ended on 3 September 2026 and had 53 weeks.² In that year revenue was $133.2B, up 256%. DRAM was $100.7B, NAND $31.8B and other products, mostly NOR, $0.7B.² Net income was $85.0B, or 64% of revenue.²

The filing says why, in one line each. DRAM sales rose 252% on an average selling price up about 180% and bit shipments up in the mid-20s percent. NAND sales rose 274% on prices up about 200% and bits up in the mid-20s.² A bit is the unit of memory sold. So volume grew at a normal pace, and price did nearly all the work. Cost of goods sold rose from $22.5B to $25.7B while revenue rose by $95.8B C (calculated).² Almost all of the extra revenue fell through to profit. That is the most important sentence about this company today, because it also works in reverse.

Why memory became scarce

AI servers carry far more memory than ordinary servers, and the most prized type, high-bandwidth memory or HBM, stacks DRAM dies and uses more wafers for each bit. The 10-K says HBM needs more wafers and more cleanroom space than conventional DRAM for the same bits.² So every HBM bit removes more than one ordinary bit from the market. Management calls this the trade ratio, and on the call it said the ratio rises with each HBM generation.¹⁵

A cleanroom is the sealed building a fab sits in, and it takes years to build. Micron’s first new Idaho fab broke ground in September 2022 and first wafers are due in mid-calendar 2027.² The company says output becomes meaningful only a few quarters after that.⁴ In the meantime the industry grows bits by moving to finer nodes, and management says that source is giving less with each step.¹⁵ It expects industry DRAM bit shipments to grow in the low 20s percent in calendar 2027 and 2028, NAND in the mid-20s, and supply to stay short in both years. It says it has no line of sight to when supply and demand balance.⁴ Those are company claims. SK hynix and Samsung say much the same, which is reassuring and also what one would expect to hear from every seller near a peak.⁷⁸

What it sells, and to whom

Micron reports four business units. Cloud Memory sells to the hyperscale clouds and sells HBM to everyone. Core Data Center sells SSDs and memory to server makers, enterprises and the newer AI clouds. Mobile and Client sells into phones and PCs. Automotive and Embedded sells into cars, factories and consumer devices.²

In the fourth quarter Core Data Center became the largest unit at $18.0B, and data-center SSD revenue was nearly $10B, more than ten times the year-ago quarter and over two thirds of NAND revenue.⁴³ Micron does not give HBM revenue in dollars for the quarter. It says HBM grew faster than the company, and that agreements cover the vast majority of its calendar 2027 HBM bits at much higher prices.⁴

Phones and PCs are where the shortage shows. Mobile and Client revenue tripled while its bit shipments fell, because Micron sent supply elsewhere.² Management expects possible double-digit unit declines in both markets this calendar year, and an analyst on the call pointed out two quarters in a row of lower bits in that unit.⁴¹⁵ Customers there are paying more for less. That is fine for a year. Over several years buyers redesign, as the 10-K warns: high prices may push customers to cut memory content or seek other technologies.²

No customer reached 10% of revenue in FY2026, after one at 17% in FY2025. Over half of revenue came from the top ten, and $91.3B came from customers based in the United States.² China has fallen: management expects single-digit exposure in FY2027, following the May 2023 decision that barred critical infrastructure operators there from buying Micron.¹⁵²

Who pays for all this

A simple check. WSTS’s spring forecast put 2026 memory sales above $800B out of about $1.51T for all semiconductors.¹⁶ The four largest cloud buyers spent $496.7B on capital goods in the first nine months of 2026, and $201.9B of that in the third quarter, about $808B a year, on everything they buy C (calculated).¹⁷ Those are different buyers and different years, since phones, PCs and cars buy memory too. Even so, the world’s memory bill is now about as large as the combined capital budgets of the four customers most responsible for the shortage. I can’t tell how much further that can go. I can say that a cost line this large gets attention, from engineers as well as from buyers.

IIThe moat

A durable business keeps competitors from taking its profits. For memory the test is the record, and the record is mixed. From FY2016 to FY2025 Micron’s gross margin averaged 36.6%, weighted by revenue, and it swung from 58.9% in FY2018 to −9.1% in FY2023 C (calculated).¹⁴⁶ A company with real pricing power does not sell below cost. Micron has. The 10-K says that in some periods selling prices have been below manufacturing costs, and that annual changes in DRAM and NAND prices over five years ranged from about +200% to about −50%.²

What does exist is a high wall around a small group. TrendForce put Samsung at 39.4% of DRAM revenue in the second calendar quarter of 2026, SK hynix at 24.9% and Micron at 23.3%, or 87.6% between them C (calculated).⁵ Entering means years of process work and tens of billions of dollars. Three firms is why prices can stay high at all. It doesn’t stop them from all building at the same time, which is how downturns here start.

Evidence it is widening

The strongest new fact is the contract. Micron has signed 26 strategic customer agreements, multi-year take-or-pay deals with committed volumes. Customers have put up $32B of commitments, mostly cash deposits. Three quarters of the estimated revenue has a defined price framework, most of it with a floor and a ceiling.⁴ A year earlier Micron’s remaining obligations under contract were not material. The 10-K says floor prices will still give, it expects, gross margins meaningfully above any past peak quarter, which was 61.0% in the quarter to August 2018.²¹⁸ That is a claim about the future, and it has not yet met a downturn. But customers paying cash to secure supply is a new kind of evidence.

The second fact is that memory has become harder to make well. Scott DeBoer, the technology chief, said on the call that the process differs more between HBM, server and mobile products on one node than at any time in DRAM’s history.¹⁵ If true, the old description of DRAM as one interchangeable commodity is less accurate.

Evidence it is narrowing

China. CXMT’s share of DRAM revenue rose to 9.5% in the second quarter of 2026 from 7.6%, on revenue that nearly doubled in a quarter to $14.6B, and press reports put its capacity moving from about 320,000 wafers a month to 420,000 by 2027.¹⁹ High prices are the best funding a challenger could ask for. The 10-K itself says sustained high prices may attract new entrants.²

Margins this high also teach customers to look for ways out. Management said its customers are shipping as much compute as their memory allows, and that memory content growth this year is modestly lower than expected because supply is tight.⁴ The first half of that is a sign of scarcity. The second half is a sign that designers can, when pushed, get by with less.

My judgement: the industry’s structure is better than it was, and some of the improvement may last. Whether Micron has a lasting advantage of its own, I can’t yet say. One cycle of take-or-pay contracts will tell us more than any argument can, and that cycle hasn’t happened.

IIIThe competitive advantage

Micron is the smallest of three. In the second calendar quarter of 2026 TrendForce had it $2.6B of DRAM revenue behind SK hynix, and it lifted its share slightly by putting supply into high-priced server DRAM.⁵ The table compares the latest reported quarter of each maker. The quarters end a month apart, and prices were rising, so the later reporters have an edge.

The last column surprised me. On one quarter’s operating income, annualized, Micron trades at more than twice the multiple of SK hynix, which is larger, has more net cash, and made the same margin. Part of that gap is the old discount on Korean shares. Part may be Micron’s American fabs. I can’t split it. A buyer of memory exposure has a cheaper option on this measure, and that belongs in any honest comparison.

Where Micron wins

Margin first. Its 80.7% GAAP operating margin led Samsung’s division by about eight points, in an earlier quarter, helped by having no large foundry or phone business to carry. Geography second: it is the one large memory maker building leading DRAM in the United States, which matters to governments and to customers planning for a Taiwan or Korea problem. Data-center SSDs third: management says calendar 2026 will be the fifth straight year of record share, and that line has gone from a small business to near $10B a quarter.⁴

Where it loses

Scale. Samsung spent KRW 27.2T, about $20B, on chips in its third quarter alone, against Micron’s $10.8B net in its fourth.⁸⁴ Samsung can outspend Micron and carry losses longer. HBM is the second. Micron says its share is about the same as its DRAM share, while TrendForce says HBM is a larger part of SK hynix’s shipments than at the other two, and that Samsung moved early on HBM4 and had the strongest bit growth of the three in the quarter.¹⁵⁵ NAND is the third. Management says Micron’s NAND supply grew less than the industry this year, partly because Singapore cleanroom went to R&D and HBM packaging.¹⁵⁴ It has no special claim on any of these. It has good positions, and it has to keep earning them node by node.

IVManagement

Sanjay Mehrotra, 68, became chief executive in May 2017 and chairman in January 2025. Before that he co-founded SanDisk in 1988 and ran it until its sale in 2016.² He is a memory engineer with more than 70 patents.¹¹ On my reading, Micron moved under him from a follower to a peer on process technology. He also took the company through FY2023, when it lost $5.8B.⁶

In August 2026 Manish Bhatia took the operating P&L and Scott DeBoer the technology roadmap, and both became presidents.²⁴⁴ Micron has not named a successor. With a chief executive of 68, I read the move as the start of that question, but that is my reading and nobody has said it.

How they are paid, and what that pays them to do

Follow the incentive first. The latest proxy covers FY2025. Mehrotra’s reported pay was $30.9M, 529 times the median employee: a salary of $1.45M, a target bonus of 220% of salary, and $25.4M of stock, about two thirds of it tied to performance.¹¹ The bonus is half profit goals, adjusted net income or operating margin, and half five strategic goals the company does not publish.¹¹ The performance stock is half relative share return against the SOX index, with target at the 55th percentile, a quarter HBM share or bits, and a quarter data-center SSD share or bits.¹¹

So what does that pay them to do? Win share in the two hottest products, and beat other chip stocks over about three years. Both are sensible at this point in the cycle. Neither pays for what decides this business over a decade: return on the capital put into fabs across a full cycle. I found no such measure in the plan. A profit bonus in a year when price does the work, and a share-return test over a window that can sit wholly inside a boom, reward being in the right place. The FY2026 proxy, due in about a month, will show whether that changed.

It already shows up in the numbers. In the fourth quarter Micron raised FY2026 incentive pay for every employee, which with a $300M community gift took operating expenses up $1.1B in a quarter, and management expects higher incentive pay again in FY2027.⁴ Sharing a windfall with staff is easy to defend. It is also a cost that is easier to raise than to lower.

How much of their own money is in

His latest Form 4, after vesting on 13 October 2026, shows 355,146 shares held directly and 607,075 through a grantor annuity trust. That is at least 962,221, or about $851M at the close C (calculated).⁹ The filing may not list every family holding, so I treat it as a floor. That is far above the guideline of six times salary, and about 27 years of his FY2025 pay C (calculated).¹¹ He sold 332,500 shares for about $210M between October 2025 and August 2026, at an average of about $631 C (calculated).¹⁰ All insiders sold about $418M in that time. One director bought: T. Mark Liu, the former TSMC chairman, paid about $7.8M for 23,200 shares in January 2026 at about $337 C (calculated).¹⁰¹¹ DeBoer adopted a plan in July to sell up to 85,632 shares from 16 October.² Selling after a sevenfold rise is not a verdict on the future. But the people closest to the cycle have been net sellers, and the size of the holding that remains, not the sales, is what lines them up with an owner.

What they did with capital over a full cycle

This is the fairest test, because it covers good years and bad. From FY2016 to FY2025 Micron earned $46.0B of net income and spent $91.4B on plant, against $105.5B of operating cash flow C (calculated).^6131425 It paid out $9.5B: $7.5B of buybacks and $2.0B of dividends. Shares outstanding still rose from about 1,042M to 1,122M.¹⁴²⁵ Most of the money a decade of memory earned went back into memory, which is the nature of the trade.

Management says it will equip the new cleanrooms only as demand warrants, and that over time it expects to return 100% of excess cash.¹⁵⁴ I like both statements. The first has not been tested yet. A finished cleanroom with no tools is a hard thing to leave empty when a rival is filling its own.

The board has changed too. Richard Beyer and Mary Pat McCarthy retired in January 2026 and Alexis Black Björlin, a former NVIDIA and Meta infrastructure executive, joined in June.²⁷¹¹ The chairman and chief executive are one person, with Lynn Dugle as lead independent director.¹¹

VFinancials

Ten years of Micron contain two full cycles and part of a third. Revenue went up 50% in FY2018, fell 23% and then 8%, recovered, fell 49% in FY2023, and has since gone up 62%, 49% and 256% C (calculated).¹⁴¹³⁶² Any single year is close to useless as a guide. Averages over a cycle are better.

One date trap. FY2026’s extra week fell in the fourth quarter, which ran 14 weeks from 29 May to 3 September. So $54.2B is $3.87B a week, and the $61.5B first-quarter guide is $4.73B a week over 13 weeks, up 22% against a headline 13% C (calculated).¹⁸³ That guide’s 85.95% GAAP gross margin is below the fourth quarter’s 86.8%, because extra incentive pay held in inventory falls mainly on that quarter. Management calls it the floor for FY2027.³⁴

What an owner could really take out

Reported earnings flatter this business, because depreciation runs far below what it takes to stand still. Here is the simple version. In FY2026 net income was $85.0B. Operating cash flow less capital spending less stock pay was $57.6B, or 68% of net income C (calculated).² That figure has two large distortions in it. Receivables took $25.2B as revenue surged, and $9.6B of other noncurrent liabilities, mostly income tax payable later under the new 15% minimum tax, flattered the cash.² The tax line is the one to watch. Micron charged $14.8B of income tax in FY2026 and paid $1.3B in cash, while accrued income and other taxes rose to $5.6B and long-term tax payable to $9.82B.² Management’s adjusted free cash flow of $62.3B also adds $3.3B of government incentives.³

The better question is what the fabs need just to keep Micron’s place. Depreciation was $9.5B. Capital spending, net of incentives, was $27.4B, and that was held down by a shortage of cleanroom space. Gross spending fell to 23.1% of revenue, the lowest in the table. The last time it was that low, in FY2017, it was 41.8% two years later, because revenue fell while the fabs kept coming. Over FY2016 to FY2025 Micron spent 37% of revenue on plant C (calculated).^1413625 The industry needs about 20% more bits a year, so holding a 23% share means growing with it, and new nodes give less each time. I put the cost of standing still at about $42B a year in today’s dollars, with a range of $30B to $60B, and I can’t narrow that honestly E (estimate). On that basis FY2026 owner earnings were about $52B, roughly 60% of reported net income, with a range of about $35B to $65B E (estimate). The longer record says the gap can be much wider: over the ten years to FY2025, $46.0B of net income became $9.5B of owner free cash flow C (calculated). In its fourth quarter alone Micron generated more than three times what it had left for owners in that decade.³

Quality of earnings

Stock pay is small, $1.3B or 1.0% of revenue, and the non-GAAP gap is modest: $75.52 a share against $74.33 GAAP.³ Inventory is $10.4B at 129 days, up nine days, and management says that includes a build ahead of older nodes ending.⁴ Other current liabilities carry $4.32B for amounts payable to customers, mostly price adjustments, up from $1.19B.² That one is worth watching, because money owed back to customers in a rising market is unusual.

The tax rate was 14.8%, with 15.5% guided for FY2027.²⁴ Pillar Two, the global minimum tax, has largely cancelled the Singapore incentive that kept the rate lower.² I use 16%.

The balance sheet

Micron ended the year with $73.45B of cash and investments and $5.2B of debt, $68.3B of net cash.²⁴ I don’t count all of that. $12.9B of noncurrent contract liabilities are customer deposits, nearly all due back between 2029 and 2031, and $9.82B is long-term income tax payable.³² Taking both off leaves $45.6B C (calculated). Customers’ commitments have since risen to $32B, and the extra cash arrives as a financing inflow that is also owed back, so I add neither side.⁴ Those deposits have become a real source of funding. Customers paid in $12.7B in FY2026, nearly half of net capital spending, and in the fourth quarter their $12.3B was more than the $10.8B Micron spent on plant C (calculated).²⁴ Customers are now paying for much of the build, and that money comes due in the same years the new supply arrives.

Ratings are now BBB+ or equivalent from all three agencies, and Micron says capital returns will rise from 9 December 2026, the second anniversary of its CHIPS agreements.⁴² If a downturn comes, this balance sheet is the best protection the shares have ever had. In FY2023, Micron went into its worst year with far less.

VIValuation

First, how this goes badly wrong

Before any numbers, I asked what would most likely ruin this investment, and worked back. Three things can.

• Everyone’s cleanrooms arrive at once. Micron’s Idaho and Tongluo sites start in mid-2027, and Idaho 2, Hiroshima and Singapore NAND in the second half of 2028.²⁴ SK hynix expects investment to rise as Yongin starts, Samsung’s chip capital spending has reached about $20B a quarter, and CXMT is adding capacity.⁷⁸¹⁹ Each is rational alone. Together they are how memory downturns start. With costs mostly fixed, a price fall goes straight to profit, as FY2023 showed: revenue down 49%, a $5.8B loss.⁶
• Demand was partly a fear of running out. The 10-K says customers have overstated demand in the past, possibly to get more supply.² Buyers who are short order double. If AI spending pauses, those orders vanish at the same time as the new supply comes on. Nearly two thirds of revenue through 2030 has no take-or-pay contract at all.⁴
• The money goes in at the top. Even if the business does well, an owner can lose. On my base Micron puts more than $130B into construction and tools over two years, and it has authorized $35B for its own shares, at the highest margins it has ever had. Deposits of up to $32B fall due from 2029 to 2031, which could be the worst years.⁴²

None of these needs a surprise. All three have happened in this industry, and two of them are in the company’s own risk factors. They set my bear case and the reason I use a through-cycle margin at all. Taiwan is the fourth, and it sits outside the model: $27.6B of $64.0B of long-lived assets are there.²

What the price implies

At $884.74 the market values Micron at about $1.02T on 1.15B diluted shares, 7.4 times book equity C (calculated).¹³ On the face of it that is cheap: 11.9 times FY2026 GAAP earnings, 5.8 times the first-quarter guide annualized, and 5.0 times Nasdaq’s FY2027 consensus C (calculated).³¹² That consensus runs $177 for FY2027, $208 for FY2028, $183 for FY2029 and $150 for FY2030, though only three analysts reach the third year and one the fourth.¹² So the Street itself draws a peak. At these multiples the market is plainly saying the earnings won’t last. Low price-earnings multiples at the top are how cyclical stocks usually look.

A single growth rate from a peak year means nothing here, so I turn the reverse DCF around. I take my base cash through FY2030 and ask what the business must earn after that. Those four years of cash, discounted, are $397B. The price pays for $575B more. That needs $78B of owner cash a year from FY2031, growing 3%, for good C (calculated). For comparison, the most Micron ever produced in a year before FY2026 was $8.3B.

Put another way, the price needs gross margin to settle near 61% on my base costs and volumes, with the next low near 58%. Micron’s best full year before FY2026 was 58.9%, and its ten-year average 36.6%. So the market is betting that the low of the next cycle will sit about where the best year of the old one did. That is exactly what management says the contracts will do, on the part they cover. Or the price needs about three more years at the FY2028 peak than I allow, which is a long time for any memory peak to last.

What I can’t judge

I can’t tell you the normal gross margin of a memory maker after this. Nobody can. There has been no downturn since take-or-pay contracts began, and no earlier cycle had AI in it. The useful output is a range, so here it is, and read my $700 with it in mind. Small changes in this one input move value more than everything else on the page.

The discount rate

I use 12.0%. The capital asset pricing model starts with the 4.94% ten-year Treasury yield in Q4 2026 and Damodaran’s 4.20% implied equity risk premium of 1 October.²⁸²⁹ My ten-year beta for Micron, on month-end returns, is 1.75, which gives 12.3%. The five-year beta of 2.33 gives 14.7%, and the three-year beta of 3.48 gives 19.6% and is mostly this year’s run C (calculated).¹³⁰ I take 12.0%, a beta of about 1.7. The downturn is already in my cash flows, so I do not add for it again, which is why I sit just under that range. It is higher than the 10.5% on my Broadcom page, because memory’s earnings are less steady.

Three cases

I keep the model simple. I grow bits, take costs down per bit, roll depreciation from the capital spending, and then set one number a year: gross margin, or revenue for the first two years, where management’s comments give me something to work from. Micron’s FY2027 guide of $61.5B for the first quarter, with revenue up every quarter, gives about $276B in my base.³⁴ I add $4B of one-time cost to FY2027 in all three cases for higher incentive pay, which puts my base gross margin at 87.5% against a guided floor of about 86%. Tax is 16%. Stock pay is a cost. Working capital is 15% of added revenue and comes back when revenue falls, because receivables really do collect. Each year’s cash arrives at the end of the fiscal year, and I deduct the deposits and long-term tax payable from net cash.

Why does the base revenue fall 62% from peak to trough when the worst past fall was 49%? Because the peak is much further above cost than any before. Selling prices in FY2028 would be about five times FY2025’s average on my numbers, and bits double by FY2030. To reach even a 52% gross margin, price has to fall a long way. Revenue of $136B in FY2031 is still as large as FY2026, the best year in Micron’s history until now, and 3.6 times FY2025.

The contracts are the main reason my base doesn’t go lower. RPO, the contracted minimum at floor prices, was $134B at 3 September and about $150B by 30 September.²⁴ The filing says about a quarter of the first figure falls in twelve months and all of it within five years.² Spread evenly, $150B is about $35B a year, and it covers the priced three quarters of the 35% under contract, about 26% of revenue C (calculated). The largest first agreements set their ceilings at second-quarter 2026 market prices, and floors below that.³¹ If those floors applied to everything, company revenue would be about $134B a year E (estimate). That is a rough sum and it flatters the floor, since 74% of the business has none. It does tell me my FY2031 base isn’t absurd, and that the floors protect a level well below today’s run rate of about $201B, the fourth quarter scaled to 52 weeks.

Other uses of the same dollar

Price only means something next to what else the money could do. The ten-year Treasury pays 4.94%. My base here returns 7.1% a year, so about two points over bonds for a business with a 62% fall in the middle of it E (estimate).²⁸ The closest comparison on this site is TSMC, which sells the same AI demand with far steadier prices: 9.4% a year on my base there, 5.3% on my bear. NVIDIA is 9.7% and 5.5%, and Broadcom 7.5% and −4.8%. Those pages use their own prices and rates. Micron’s spread is wider at both ends: −3.4% on the bear and 22.2% on the bull. For AI demand alone, TSMC offers more on my base with less to lose. Micron offers more only if the bull case is right. Within memory, SK hynix trades at less than half Micron’s multiple of current operating income, as the table above shows, though it carries its own Korean and governance questions, which its page sets out.

What past lows looked like

The stock’s own record gives two rough checks. At the last three cycle highs it fell 54% from May to December 2018, 50% from January to September 2022, and 58% from June 2024 to April 2025 C (calculated).¹ In the first two the high came one to three months before the peak in gross margin, 61.0% in August 2018 and 47.2% in March 2022, so the shares turned while the reported numbers were still rising C (calculated).¹¹⁸ The same three falls from June’s $1,213.56 would give $512 to $609. The low so far was $739.00 on 29 July.¹

The second check is book value, the old floor for memory shares. At those three lows Micron sold at about 1.0, 1.1 and 1.5 times its latest book C (calculated).¹¹⁸ It sells at 7.2 times today on the same count of shares outstanding. On my base, book would reach about $511 a share by the end of FY2028 if nothing were paid out E (estimate). One to one and a half times that is $511 to $766. That is no forecast, and contracts may lift the floor. It does show where this stock has found buyers before, and my $700 sits near the top of that range.

The margin of safety

• Price$884.74Close, Q4 2026 [1]
• Base value$700My estimate E (estimate)
• Margin of safety below$533set from my bear case C (calculated)

I set the margin of safety from the bear case instead of as a discount to base value, because one cycle drives this business. Below $533, 24% under my base value, my $400 bear value is at least three quarters of the price paid, and my base case would return about 19% a year and my bear case 5.3% E (estimate). That is 3.2 times my FY2027 earnings, a number that looks silly until one remembers how these stocks trade at peaks. The stock last closed below $533 on 30 April 2026, so this is a price from six months ago.¹ The price sits 66% above it, and there is no margin of safety at $884.74. The cushion between bear and $533 is for what the model doesn’t hold: Taiwan, a change in export rules, and my own error. It doesn’t fully cover the stress case: a repeat of FY2023 at this scale gives $330, which is 62% of $533.

Two different questions

How good is the business, and how good is the price? They have different answers. Today’s Micron is the best it has ever been: more contracted, less indebted, closer to its customers, and less exposed to any one of them. Over a full cycle its record is still that of a capital-hungry, price-taking maker of a near-commodity. So I would call it a much improved business of unproven durability. The price, at 26% above my base, already pays for most of that improvement to last. The low multiple on current earnings does not change that, because current earnings are the thing in question.

Where this could be in 5 and 10 years

For the base to be right at 5 years, the profit spike has to end the way I model it. Revenue peaks at $355B in FY2028, then falls to $136B by FY2031, and gross margin drops from 89% to 52% E (estimate). The base is $515 by then because the richest cash years have been paid out and are gone. For it to be right at 10 years, margin has to settle near 55% and revenue climb to about $181B in FY2036, with capital spending near $49B a year. That gives $589 E (estimate).

Bear and bull differ on how long prices stay high and what it costs to keep up. In the bear, revenue falls to $84B in FY2030 and margin to 38%, so little cash is left and value stays near $290. In the bull, margin stays above 70% and revenue above $320B, but Micron spends $84B to $94B a year on capacity. That still leaves $1,684 in year 10. Growth, margin, capital needed and the 12% discount rate drive all three.

The most likely way my base is wrong is margin. Five points more in every year after FY2028 lifts year 10 to $767, and five points less cuts it to $466 E (estimate).

These are estimates of value, not forecasts of the share price. The market price can sit above or below value for years.

VIIThe case for the price

What most people miss

Three things cut in Micron’s favour and don’t make the headlines. No customer reached 10% of revenue in FY2026, down from one at 17% a year earlier, so the boom has spread Micron’s buyers out, not narrowed them.² NAND has quietly changed too: data-center SSDs were over two thirds of it in the fourth quarter, nearly $10B, more than ten times a year earlier.⁴ And the contract floors are set high: Micron says even floor prices will give margins above any earlier peak quarter, 61.0%.²¹⁸ If that holds, the worst of the contracted part of the business beats the best of the old one.

The bull case

In plain words: memory has stopped being a commodity and become the limit on AI. Management says that DRAM, more than logic or power, is now the main constraint, that supply will be much tighter in 2027 and 2028 than in 2026, and that it can’t see when balance returns.¹⁵⁴ SK hynix, a rival with no reason to help, says demand will exceed supply through 2030, and has signed 14 customers for over half its 2027 and 2028 capacity.⁷

Gavin Baker of Atreides is the best-known investor making the argument. Zacks reported on 6 July that he told the All-In podcast this cycle differs because the chips are paid for from hyperscaler cash flow and put to use at once, unlike the fibre of the late 1990s.³² I could not open the episode, so that is Zacks’ paraphrase. Holdings show who else came in. At 30 June, the latest 13F date, Atreides held 711,364 shares, Altimeter, Brad Gerstner’s firm, a new 209,520, Coatue 3.14M and Situational Awareness 4.83M, while Appaloosa had cut to 975,000 from 1.67M.^3334353637 Those are four months old, and they don’t tell you why.

Against reality

Today’s numbers support the first half. The fourth quarter earned an 87.0% non-GAAP gross margin, and over 75% of FY2027 shipments are committed.⁴¹⁵ If Micron does what it says through FY2028, my base has it making about $312B of owner cash in two years, nearly a third of the market value E (estimate). So near-term delivery is not where the argument is.

The argument is what comes after, and that needs several things at once. If Micron holds its share, if the contracts renew at good prices, and if new supply meets demand instead of swamping it, a 61% normal gross margin is possible, and the price with it. Today Micron has a 26-agreement record of two quarters, and no downturn to test it.

The bull case needs more, and it helps to multiply it out. Say tightness lasts through FY2028, 70%. New cleanrooms don’t cause a glut in 2029 or 2030, 50%. AI spending keeps rising to 2030, 60%. Floors stick and contracts renew at high prices, 70%. Micron keeps its 23% share against CXMT and two larger rivals, 80%. Those are my guesses, and they aren’t independent, since good AI demand helps most of them. Multiplied, they give 12% E (estimate). Allowing for the overlap, perhaps 20% to 25%, which is the weight I give the bull. Each step looks likely. All five together do not.

What a 10x would take. Ten times $884.74 is $8,847, about $10.2T of market value. The largest company I cover, NVIDIA, is about $5T.³⁸³⁹ At 20 times earnings Micron would need about $510B of net income, six times FY2026. At a 50% net margin, far above anything but this year, that is about $1T of revenue, more than the whole memory market WSTS forecast for 2026.¹⁶ Micron’s FY2026 revenue is about a sixth of that market, so at the same share the world would need to buy about $6T of memory a year. By FY2036 that means 26% a year for ten years from a market value of $1T C (calculated). The stock has risen more than 13 times since its April 2025 low of $64.72, but that began at about $70B of value.¹ The best run I know of from this size is NVIDIA’s: about five times from its $40.11 close of May 2023, near $1T, in about three and a half years.³⁸ Micron would need twice that from the same start, and in memory. From here, tenfold is close to impossible, and I would not build a case on it.

Where I come down

The contracts moved me: without them my normal margin would be near 45%, not 55%. The record quarters did not, because one good stretch can’t prove a change in kind. A first downturn in which floors hold and margins stay above 50% would raise my values most.

VIIIRisks, and what would change my view

The three ways this goes badly wrong are at the start of Valuation. These are the rest.

• The cycle itself. Michael Burry disclosed a short position at about $1,051.87 in July, writing, as reported by Invezz, that “Micron defines cyclical like no other.” He counted 34 falls of more than 30% in 42 years, and a median return on equity of 7%.⁴⁰ I could not open his post. The stock has fallen 30% from a high six times in ten years, and it has already fallen 27% from June C (calculated).¹ Even if I’m right about value, the path can be violent in either direction.
• Taiwan and Asia. Long-lived assets of $27.6B are in Taiwan, $12.3B in Singapore and $7.2B in Japan.² Tongluo adds to the Taiwan figure.
• China and trade. Export rules, tariffs and retaliation can cut sales or raise cost. China has already barred Micron from critical infrastructure once.² Chinese rivals are state-supported, and the 10-K says so.²
• Technology. A slip on HBM4E, 1-delta or EUV would cost share at the point prices are highest. Management says 1-delta volume starts in the second half of calendar 2027.⁴
• Contracts in a downturn. Take-or-pay has not been tested. Letters of credit of $7B and deposits are the security, and $7B is small against $150B.² A customer in trouble may renegotiate, and a seller who wants the next contract may agree.
• Legal. The fourth quarter carried a $500M patent license charge. The 10-K reports Netlist and Micron agreed on 5 October 2026 to dismiss all claims.³²
• Power and water. The 10-K warns AI data centers may compete with fabs for utilities.²
• Succession. Mehrotra is 68 and no successor has been named.²