Micron: The Floor Under Contracted Supply

Executive Summary

Thesis status: Re-issued after the FY2026 10-K, with five conditions in place of ten (Exhibit 1). The test the thesis turns on, a quarter of falling DRAM prices, is unlikely before FY2029.

1. At the 23 October close of $877.55, the Micron Technology ($MU) share price requires a flat GAAP EBIT margin of 33.7-39.8% from FY2027 to FY2035, about the model's base path, which steps down from 83% in FY2027 to 27% by FY2035. With FY2027 held at 83%, the following eight years need 29.8-36.4%. No eight-year stretch since FY2015 averaged more than 24.7%, and the closest precedent is the cycle from FY2018 to FY2022, which averaged 29.8% through a trough of 14.0% in FY2020. The price is consistent with a downturn that shallow and with neither of the deeper troughs, in FY2016 and FY2023.
2. The strategic customer agreements are built for that trough and reach a minority of the book. Management expects agreements with price bands to yield gross margins "meaningfully above our peak quarterly margins in any past cycle" even at floor prices, yet floors and fixed prices cover only a little over a quarter of revenue through 2030 (derived). On the 10-K's figures, the banded and fixed-price volume due in the next twelve months, valued at minimum prices, comes to under 14% of the FY2027 revenue that 1Q27 guidance implies (derived). The rest of the book reprices with the market, some of it on annual terms.
3. The reads arrive in sequence. 1Q27, expected in mid-December, decides the inventory leg of the early warning after inventory days reached 129 against a threshold of 123. 2Q27 shows whether January's HBM price reset narrows the 4-point gap between the gross margin of the unit that sells all of Micron's HBM, 83% in 4Q26, and the company's 87.0%. The floor itself waits for the first quarter of falling DRAM prices. Management expects calendar 2027 and 2028 to be "much tighter" than 2026, and Micron's next DRAM cleanroom space with a disclosed date is expected in late 2028, which makes FY2029 the earliest likely year.
4. The main uncertainties sit in what Micron does not disclose and in its own forecasts. It publishes no HBM margin, no split of the contract book by customer, unit or product, and no figure for the past-cycle peak its floor statement refers to, so the decisive reads rest on management's wording and a blended unit margin. The supply outlook is Micron's own until SK hynix's and Samsung's results join the read at 1Q27, and the 10-K warns that weaker HBM demand could release conventional supply without any new cleanroom. Customer deposits of $12.75B, nearly a fifth of net cash, fall due in 2029-31, starting in the year the test first looks likely.

I. The Question and the Initial View

Does contracted supply keep Micron's gross margin above its prior-cycle peak through the first quarter of falling DRAM prices, within FY2027-FY2029?

The strategic customer agreements (SCAs) are built to pass this test on the volume they cover. Micron expects the agreements with price bands to yield gross margins "meaningfully above our peak quarterly margins in any past cycle" even at floor prices. They cover a minority of revenue. Management estimates the agreements at over 35% of revenue through 2030 and prices a quarter of that revenue periodically at market, which leaves a little over a quarter of revenue under a floor or a fixed price (derived). In the first falling quarter, company gross margin depends at least as much on how far the rest of the book reprices as on the floors.

The horizon is FY2027 through FY2029, September 2026 to August 2029, set by when the test is likely to arrive. Management expects supply and demand to be "much tighter in calendar 2027 and 2028 than they were in 2026", an outlook that already counts Idaho's first wafers in mid-2027, and gives no date for a return to balance. Micron's next DRAM cleanroom space with a disclosed date, in Idaho and Japan, is expected in late calendar 2028, at the start of FY2029, and output from a new fab becomes meaningful "a few quarters after initial output". FY2027 therefore tests contract coverage and the January 2027 HBM price reset, FY2028 tests whether the industry stays short before the new space arrives, and FY2029 is the earliest year in which a first falling quarter looks likely. If prices hold up through the horizon, the floor goes untested and the question rolls forward, with the contract and supply conditions left to judge the thesis.

The March primer was right about the scarcity and wrong about where its margin would accrue. DRAM prices rose by more than 60% in each of 2Q26 and 3Q26, and 1-gamma became the majority of DRAM bits as targeted. The primer expected HBM4's premium pricing to widen the spread between price and cost, yet the business unit that sells HBM earned the lowest operating margin of Micron's three largest in FY2026. It also placed a 1-gamma base die under both HBM4 and HBM4E. The 10-K puts HBM4 on 1-beta, with base and core dies made in-house, and HBM4E on 1-gamma, while management states that HBM4E's base die is made on a foundry process, which moves part of that generation's yield and timing outside Micron. The March primer stays as published, and this document replaces the question, horizon and ten conditions that the FY2026 memo closed into five.

Signal and Noise

"The contracts end the memory cycle." The agreements assume the cycle continues and set where contracted margin bottoms when it turns. The 10-K describes take-or-pay volumes with pricing "either fixed or periodically negotiated", most of it within minimum and maximum bands, and management's floor statement describes the margin at the bottom of those bands, which binds only if market prices fall that far. The same filing warns that the agreements "could also constrain our available supply and limit our flexibility to respond to changes in market conditions". The floors reach only contracted volume, so supply sold at market prices meets a falling market in full. Condition 1 tests the floor once prices fall, and condition 2 tracks how much of the book it reaches.

"HBM is where the margin is." In FY2026 HBM trailed conventional DRAM on margin, by management's own account. The Cloud Memory Business Unit (CMBU), which sells all of Micron's HBM, earned a 72% operating margin against 79% for Core Data Center and 80% for Mobile and Client. Management states that 2026 HBM prices were negotiated the previous year and that 2027 HBM prices are "much higher than 2026 prices", which is "helping us narrow the margin gap with the non-HBM memory". Micron's reported DRAM prices, which include HBM, rose in every quarter of FY2026. Annual pricing that held HBM back in a rising market would slow its fall in a declining one, though no falling market has yet tested that. The reset first shows in CMBU's 2Q27 margin, which is where condition 4 reads it.

"$150B of locked-in revenue." The disclosed book is smaller and narrower than the headline. The figure of about $150B on the 30 September call counted agreements signed to date, while the 10-K reports $134B at the 3 September year end, a gap consistent with signings after the year end. The 10-K figure counts committed volumes at minimum prices and leaves out agreements without a fixed price or band, and the filing states that it is "not expected to be indicative of future revenue". Management expects revenue to "well exceed" it. About a quarter of it, roughly $33.5B, falls in the next twelve months, under 14% of the more than $246B of FY2027 revenue that 1Q27 guidance implies (derived). Condition 2 reads the book each quarter from the 10-Q.

II. What Must Be True

Five conditions replace the March primer's ten, three load-bearing and two amplifying (Exhibit 2). Four are read on a single metric that Micron reports or states on its quarterly calls, and condition 5 pairs Micron's inventory days with the monthly retail DDR5 price. At adoption two are affirmed and two are developing, while the first, the question itself, cannot yet be read.

Condition 1 has no numeric bar because management has given none. Its statement covers margins on agreements with price bands, and the condition applies the same bar to company gross margin, which is what the question asks. The read in the first falling quarter will set reported gross margin against that statement, with a floor statement withdrawn or qualified in such a quarter counting against the condition. Until then 🟡 marks a condition that cannot yet be read.

Conditions 2 and 3 start affirmed on Micron's own account, which is the limit of both reads. The agreements went from 16 to 26 in a single quarter, and two existing ones were extended to 2031. Conditions 4 and 5 start developing for reasons management expects to unwind. The HBM reset that should narrow CMBU's gap takes effect in January. Management attributes the rise in inventory days to a build-ahead before a node's end of life and to capitalised incentive pay, and expects days "to decline in the coming quarters", but a second quarter above 123 at 1Q27 would fire condition 5 whatever the cause.

The conditions form a chain around one shared trigger, the turn in DRAM prices. Condition 3 sets when condition 1 is tested, so a management date for a return to balance would move the floor from dormant to live inside the horizon. Condition 2 sets how much of the book the floor protects when that happens, and condition 4 decides whether HBM enters the falling quarter above or below the company margin. Condition 5 is built to signal the turn before it reaches Micron's reported prices. The compound failure is a turn that arrives after coverage has stalled, which would test the floor against a smaller contracted share, with conditions 3, 5 and 1 moving within a few quarters of one another.

III. Business Architecture and the Contract Book

Micron sells through four business units, and the two serving data centres earned 61% of FY2026 revenue, with DRAM at 76% of the total. Cloud Memory (CMBU) sells to the largest hyperscale customers and supplies HBM to every data centre customer, Core Data Center (CDBU) sells data centre SSDs and server memory to OEMs, Mobile and Client (MCBU) serves phones and PCs, and Automotive and Embedded (AEBU) serves automotive, industrial and consumer devices at the edge. The 10-K attributes the rise in every unit's operating income primarily to higher average selling prices, and MCBU grew revenue 209% on lower bit shipments as its supply was redirected to other units (Exhibit 3).

Operating margin rose least where prices reset slowest. CMBU gained 27 points to 72%, against 48 to 63 points for the other three units, whose conventional DRAM and NAND repriced through the year. MCBU, selling into phones and PCs on falling bit shipments, ended the year with the highest margin of the four.

The disclosure stops at the business unit: Micron reports no HBM revenue or margin, and CMBU combines HBM, which it sells to every data centre customer, with conventional memory sold to the largest hyperscale customers, so the January 2027 HBM reset will show only in a blended CMBU margin. Micron names no agreement counterparties and does not split the RPO by customer, unit or product, which leaves the contract book measurable only in aggregate.

The agreements span all four units, from small customers to large, and bind them to take-or-pay volumes over terms that now reach 2031 (Exhibit 4). Their pricing cuts both ways. Management stated that the earlier agreements were set at calendar second-quarter 2026 market conditions and that newer ones are negotiated at prevailing prices, so while prices rise, volume under the earlier agreements can sell below the market and the book gives up margin in exchange for its floor. Micron does not disclose how much FY2027 volume ships under those earlier terms.

Customers have secured their volumes with $32B of financial commitments, about $25B in cash deposits and $7B in letters of credit (derived), and $12.75B of the deposits arrived in FY2026. A customer that misses its minimum purchases can lose all or part of its deposit, and the 10-K notes that enforcing commitments "could result in litigation or disputes", so the deposit is the penalty Micron can apply without going to court. The deposits have not yet been tested against a floor that sits above the market price.

IV. Supply, Capacity and the HBM Roadmap

Management's case for a shortage through 2028 rests on supply growing more slowly while demand broadens (Exhibit 5). It names three limits on supply: cleanroom space, which it calls "the principal constraint", node transitions that give "less productivity gain per wafer", and HBM, which takes a rising share of industry output at rising trade ratios as the industry moves from HBM3E to HBM4 and HBM4E. On demand it points to CPU-run agentic workloads alongside accelerators, and states that tight supply has held the growth of memory content per server modestly below its earlier expectations.

The HBM mechanism also runs in reverse, and the 10-K says so. HBM needs more wafers and cleanroom space per bit than conventional DRAM, and the filing warns that if HBM demand weakens and suppliers shift capacity back, "this could result in a significant increase in conventional DRAM supply". A slowdown in accelerator demand could therefore loosen the commodity market without any new cleanroom, which makes HBM demand an input to condition 3 as well as the subject of condition 4.

New space arrives in two stages (Exhibit 6). Idaho's first fab is expected to start wafer output and the acquired Tongluo fab to make meaningful shipments from mid-2027, and management's 2027-28 outlook already allows for Idaho's first wafers. The next DRAM cleanrooms with disclosed dates, in Idaho and Hiroshima, are expected in late calendar 2028, with a NAND fab in Singapore in the second half of that year and New York in 2030, while a second Tongluo cleanroom is under construction without a date. FY2027 capital spending is guided at about $25B in the first half and more in the second, with a majority of the increase over earlier plans going to construction, mostly to bring cleanroom space forward to late 2028 and beyond. Management states that it will equip that space "in line with demand trends at the time", so construction commits the space and tool decisions in FY2028 and FY2029 set how much of it becomes supply.

Micron's outlook is also a forecast of its competitors' discipline. Micron and SK hynix have both raised capital spending by more than 70% in each of their last two reported years. If the industry's new space converts into bits faster than management expects, Micron's own industry bit forecast should rise, and that forecast is the metric condition 3 reads.

HBM4E, whose base die comes from a foundry, is due to ramp in calendar 2027, the same year HBM prices reset and the Singapore packaging plant adds capacity. Management expects its HBM share to track its DRAM share without a fixed target, so HBM's contribution to Micron's margin depends more on the price reset that condition 4 reads than on share.

V. What the Price Requires

This section sets out what the share price requires in cash-flow terms and offers no valuation, price target or view on the shares. At the 23 October close of $877.55 the market values Micron's operations at $938B after $68.3B of net cash. The reverse DCF solves for the flat GAAP EBIT margin from FY2027 to FY2035 that justifies that value, with every other driver on the model's base path. The answer is 33.7% as filed, or 39.8% with capital spending after FY2027 scaled up in line with the re-based revenue (Exhibit 7). As filed, capital spending falls to 10.3% of revenue by FY2035 against depreciation of 15.5%, a gap that fails the model's own coherence check, and the scaled case brings the two into line at 15.8% and 15.5%.

Both readings sit between the FY2015-25 average of 17.5% and FY2018's 49.3%, the best full year before the current boom, and close to the model's own base path. That path holds 83% in FY2027, about the non-GAAP operating margin that 1Q27 guidance implies (derived), then steps down to 40% in FY2028 and 27% by FY2035, and the market's enterprise value is $4.4B, about 0.5%, above its value. The flat requirement sits below the path's 37.3% simple average because about half the value lies in the terminal year, which the path sets at its lowest margin of 27%.

Holding FY2027 at 83% and solving for the following eight years alone gives 29.8-36.4%. Since FY2015 the margin has bottomed three times, at 1.4% in FY2016, 14.0% in FY2020 and a loss of 37.0% of revenue in FY2023, and no eight-year stretch on that record averaged more than 24.7%. The closest precedent is the cycle from FY2018 to FY2022, which averaged 29.8%, about the bottom of the range, through the shallowest of the three troughs. The price is therefore consistent with a downturn as shallow as FY2020's and with neither of the deeper two, and condition 1 tests whether the contract floors keep the next one that shallow.

The requirement has fallen 40.7 points from the 74.4% the 3Q26 model gave on 29 September (Exhibit 8). Moving FY2027 revenue to the $246B implied by 1Q27 guidance accounts for 29.6 points and the 17.6% fall in the share price since 30 September for 8.0, while net cash, the other 4Q26 updates, capital spending, tax and the later valuation date account for a net 3.1. The lower requirement therefore depends on revenue holding near that FY2027 level, from which the model grows it at 10% in FY2028, tapering to 3% from FY2033. If FY2027 proves a peak and revenue falls back, the requirement rises again.

The discount rate moves the requirement about twice as far as terminal growth. Each point of WACC changes it by about 3.8 points and each point of terminal growth by about 2, and every as-filed reading across WACCs from 10.3% to 14.3% stays below FY2018's 49.3% (Exhibit 9). The model's risk-free rate of 4.69%, unchanged from the FY2026 memo, passes almost one for one into WACC, so each 0.1-point rise in it would add about 0.4 points to the requirement.

Peer multiples are withheld again. The FY2026 memo deferred them to this primer so that trailing periods would match, and SK hynix and Samsung report their September quarters on 28 and 29 October, after the 23 October mark, so the matched comparison moves to the 1Q27 memo.

VI. Risks and Capital Allocation

Customers headquartered in mainland China and Hong Kong accounted for 11.0% of FY2026 revenue, against 10.1% in FY2025 and 16.4% in FY2024 (Exhibit 10). The FY2026 rise came from Hong Kong, at 4.7% against 3.0%, while the mainland fell to 6.3%. The 10-K still cites the 2023 decision by China's cybersecurity regulator barring critical information infrastructure operators from buying Micron products. Management expects China exposure in the single digits in FY2027 without defining the measure, and the mainland alone was already below 7%.

Customer concentration fell while Micron's asset base moved further into Taiwan. No customer reached 10% of FY2026 revenue, against one at 17% in FY2025, mainly in CMBU. A majority of Micron's 2026 DRAM output came from its Taiwan fabs, which held 43% of its long-lived assets at the year end, up from 40% (derived), and the 10-K states that any loss of that output could have a material adverse effect.

Net cash of $68.3B at the year end includes the $12.75B of customer deposits, nearly a fifth of the total (derived). Management states that the deposits are unrestricted and will be returned towards the latter half of each agreement's term if minimum purchases are met, which the 10-K places almost entirely in 2029-31. The reverse DCF assumes deposits rise to about $25B and are repaid evenly over FY2029-31, the same years as the earliest likely falling quarter and the equipping of the late-2028 space. A customer that misses its minimum purchases in a falling market can forfeit part of its deposit, which would reduce the repayment where demand fell short.

The board has raised the repurchase authorisation to $35.16B from 9 December 2026, the second anniversary of Micron's CHIPS funding agreements. The figure is the $2.16B left on the 2018 programme plus 4Q26 adjusted free cash flow of $33.2B less dividends, about 3.5% of the market capitalisation at the mark (derived), and compares with $2.66B, the most Micron has repurchased in any year since 2019. Management intends to return all excess cash over time, mainly through buybacks. Until December 2029 the CHIPS agreements subject repurchases to limits based on free cash flow net of grants and dividends, so the cap on repurchases moves with cash flow through the horizon and a falling quarter would lower it.

VII. Watch Conditions

Each card mirrors a condition in Section II, load-bearing first. The first reads come with 1Q27 results, expected in mid-December.

✓ Strengthening

• In the first falling quarter, gross margin stays meaningfully above any past-cycle peak and management attributes the support to agreement floors
• Management repeats the floor statement each quarter and applies it to agreements signed at prevailing prices

✗ Weakening

• Management qualifies or withdraws the floor statement, or narrows it to fewer agreements
• In the first falling quarter, gross margin falls to or near the prior-cycle peak
• Customers renegotiate band floors lower or dispute their minimum purchases

Noise: Gross margin moves while DRAM prices are still rising, since the floor binds only in a falling quarter. Spot or retail price falls that do not show in the DRAM price change Micron reports. :::

✓ Strengthening

• Bit coverage through 2030 rises, or more agreements are signed or extended to 2031 and beyond
• RPO rises from $134B as new agreements outweigh recognised deliveries

✗ Weakening

• Bit coverage through 2030 falls, or management stops disclosing it
• An agreement is terminated, repriced to a lower band or disputed over minimum purchases
• Micron draws on a customer's letter of credit, which the 10-K allows only on default or breach

Noise: Quarterly falls in RPO from recognising deliveries against an ageing book. :::

✓ Strengthening

• Management holds or lowers its 2027-28 industry bit growth view and still gives no date for balance
• Peers delay new cleanroom space or describe their own supply as constrained through 2028

✗ Weakening

• Management raises its 2027 or 2028 industry bit growth forecast, or dates a return to balance
• Peers bring new cleanroom output forward of late 2028
• Weaker accelerator demand moves HBM capacity back to conventional DRAM

Noise: Headline capex totals that mostly fund construction, since tool purchases decide when space becomes bits. :::

✓ Strengthening

• CMBU's gross margin gap to the company, 4 points in 4Q26, narrows in 2Q27 and closes within FY2027
• HBM4E, with its foundry base die, ramps in calendar 2027 as scheduled

✗ Weakening

• The gap fails to narrow in 2Q27, or widens
• HBM4E slips because of the foundry base die
• Management signals that 2028 HBM prices settle below 2027's

Noise: CMBU margin moves in 1Q27, before the reset takes effect. :::

✓ Strengthening

• Inventory days fall to 123 or below in 1Q27, as management expects
• The three-month DDR5 average sets a new peak, which lifts the trigger with it

✗ Weakening

• Inventory days above 123 in 1Q27, which fires the condition whatever the cause
• The three-month DDR5 average falls below the trigger, or declines for three consecutive months

Noise: Single-month retail moves, including promotions, which the three-month average absorbs. ::: ::::

Sources: Micron FY2026 Form 10-K, filed 9 October 2026. Micron prepared remarks and post-earnings analyst call, 30 September 2026. MU reverse DCF model, 4Q26 file marked at the 23 October 2026 close, recalculated by Marvin Labs, with FY2015-22 margins from Alpha Vantage annual income statements. Epoch AI retail memory prices (Keepa) through September 2026. SK hynix annual reports. EBIT margins, including every reverse DCF figure and historical comparison, are GAAP. Gross and business unit operating margins are non-GAAP as Micron reports them. Derived figures are Marvin Labs calculations from Micron disclosures. RPO is remaining performance obligations, the contracted value not yet recognised as revenue. Inventory days are days of inventory outstanding as Micron reports them. Adjusted free cash flow is Micron's measure: operating cash flow less capital spending net of incentives. Fiscal years end on the Thursday nearest 31 August, so FY2026 ended on 3 September 2026 and ran 53 weeks.