Micron closed fiscal 2026 with $133bn of revenue, an 87% gross margin in its final quarter and $73.5bn of cash and investments. Seven times guided earnings looks cheap, but that is what memory peaks look like. At $1,065.11 the market is not paying for the peak: the price already discounts a memory downturn on roughly our timetable, with through-cycle cash flow a little below the sell side’s own estimate. The contracts and the cash pile are in the price; so is the bust. Our band is $968–1,334, and we rate the shares fairly valued.
The trough is partly contracted, and the peak is partly capped. Micron has signed 26 strategic customer agreements: multi-year, take-or-pay commitments that management expects to cover more than 35% of revenue through 2030, with about $150bn of remaining performance obligations valued at committed volumes and minimum prices, and $32bn of customer financial commitments, $12.7bn of it already received in cash. Management says even floor pricing yields margins well above any prior cycle peak. The same agreements cap the upside: the largest set ceilings for existing products at the calendar second-quarter 2026 market price. Our base case credits the floors at face value; our bear case assumes they are renegotiated at the margin, because no floor in this industry has yet been tested in a bust.
The price already embeds a downturn; the analyst targets do not. Run backwards on our modal path — a peak plateau through fiscal 2028, a downturn from fiscal 2029 and a trough in 2030 — $1,065.11 requires steady-state free cash flow of about $79bn a year, against $86bn in our base case and $62bn actually generated in fiscal 2026. Equivalently, it requires a 10.4% discount rate instead of 10.0%, or a revenue path 4.5% below ours. The disagreement between our model and the market is small. The disagreement with the average analyst target, near $1,350, is large, because those targets capitalise peak-cycle earnings without modelling the normalisation.
Two years of cash pay for a fifth of the value; the rest depends on what “normal” means. The present value of fiscal 2027–28 free cash flow is about $264bn, 21% of our equity value; the terminal value is 54% of enterprise value. The single most important input is therefore through-cycle cash flow, which our base case sets at the sell side’s own through-cycle level. Holding our cycle path, $65bn a year gives $972, still inside the band, and $50bn gives $871; the full bear path, with an earlier and deeper downturn, gives $702. Nothing in Micron’s history supports margins this high through a downturn — which is precisely why the contracted floors matter.
At $1,065.11 Micron’s market value is about $1,203bn and its enterprise value about $1,135bn after $68.3bn of headline net cash. That is 7.0 times the first-quarter fiscal 2027 GAAP earnings guide annualised, and 14.3 times trailing GAAP earnings. A seven-times multiple on run-rate earnings is a statement that run-rate earnings will not last. Our through-cycle model says the same thing explicitly: revenue grows through fiscal 2027 as guided, plateaus in fiscal 2028, falls 25% and then 14% in fiscal 2029–30 as the industry’s new clean rooms ramp, and settles from fiscal 2032 at about $86bn of free cash flow a year. That path is worth $1,113 a share, 4.5% above the price.
Run the model in Section 02 backwards at our 10.0% discount rate and the price needs one of four things, each on its own with everything else held at the base case. Steady-state free cash flow of about $79bn a year from fiscal 2033, instead of $86bn — about 8% below the sell side’s own through-cycle estimate. Or a discount rate of 10.4%. Or a revenue path 4.5% below ours in every year. Or the same downturn arriving a year earlier, in fiscal 2028, which gives $1,069. None of these is a stretch, and that is the point: the burden of disagreement with the market is light in both directions. The price sits inside our fair-value band of $968–1,334, 10% above its lower edge and 20% below its upper one.
The large gap is elsewhere. The average analyst price target is about $1,350 (MarketBeat, late September 2026), with a high of $2,000. Our base case would justify $1,350 only if the downturn were pushed beyond fiscal 2030 or never came. Those targets capitalise peak-cycle earnings; ours models the normalisation. Micron has lived through this twice in the last eight years: its operating margin peaked at 49% in fiscal 2018, and five years later, in fiscal 2023, it lost money for a full year.
The signature measure for Micron is the share of future revenue that no longer depends on the spot market. Against our base-case revenue for fiscal 2027–30 of $1,008bn, drawn to scale:
Derived. The $150bn of remaining performance obligations is committed volume times minimum price across the agreements with a defined pricing framework, and part of it runs beyond fiscal 2030, so the floor-protected share of this window is, if anything, overstated. Revenue above the floor applies management’s “more than 35% of revenue through 2030” to our base path. Three quarters of expected agreement revenue has a pricing framework, most of it with ceilings as well as floors. About 15% of the next four years is floor-protected, about 35% sits under the agreements, and the remaining 65% is where the cycle lives.
Whether a hyperscaler pays a floor price far above spot in a glut is a question no memory contract has yet answered. Until it is, the bear case keeps a 25% weight. The earliest readings arrive before the reported numbers do: the agreements’ remaining-obligation schedule in the fiscal 2026 Form 10-K in October, calendar first-quarter 2027 contract prices in January, and the capex plans of Samsung and SK hynix. At $1,065.11 the price asks for a downturn about one year earlier than ours, or a slightly lower normal — not for the floors to fail.
On the twelve-month multiples the market uses, the price is unremarkable. Memory shares near past peaks have traded at roughly five to eight times next-twelve-months earnings; on our base fiscal 2027 GAAP EPS of about $171 that range spans about $853–1,365 and brackets the price. The rating follows the cash flows, not that range. The shares closed 15% below their 52-week high of $1,255.
FY2027 revenue $276B · lowest, FY2028–33 $198B · FY2033 free cash flow on the path $86.0B · 1,133m shares · terminal value is 54% of enterprise value.
This is an unlevered discounted cash flow over seven fiscal years, 2027 to 2033, long enough to contain one full memory cycle in every scenario. Each scenario is a whole published path: revenue, GAAP operating margin, cash tax rate, depreciation and amortisation, net capex after government incentives, and the change in working capital, year by year. Free cash flow is operating profit after cash tax, plus depreciation, less net capex and working capital. Stock-based compensation is charged as a cost through the GAAP margin, so unvested awards are kept out of the share count. Cash tax is 15.5% in fiscal 2027, the guided rate, 16% in fiscal 2028 and rising to 17% from fiscal 2030 as global minimum-tax rules and incentive expiries bite. Flows are valued as at 1 October 2026 with a mid-year convention, timed from the 3 September 2026 balance sheet, so the first sits at early March 2027. The terminal value grows the steady-state free cash flow at 3.0% from the end of fiscal 2033. Micron’s fiscal year ends on the Thursday nearest 31 August.
The controls bend whichever path is loaded; they do not replace it. The revenue control scales revenue, D&A, capex and working capital together, so capital intensity is held while the size of the cycle moves. The normal-margin control sets the fiscal 2033 operating margin and shifts every year from fiscal 2029 by the same number of points, leaving the two peak years, which sit inside management’s guidance window, untouched. The steady-state control is the free cash flow the terminal value grows from: it follows the loaded path’s fiscal 2033 free cash flow, including any change you make to the first two controls, until you move it yourself; it then holds your number until a scenario card is pressed. One discount rate and one terminal growth rate apply to every scenario.
Why these inputs. Fiscal 2027, $276bn (assumed from guidance). The $61.5bn first-quarter guide plus sequential growth in each later quarter at a slowing rate, about +9%, +7% and +5%, consistent with management’s “more moderate” price increases. Our derived GAAP EPS of about $171 sits above pre-results consensus near $155 and Wells Fargo’s $166, as the guide itself does. Fiscal 2028, $305bn (assumed). Bits up about 18% and prices easing about 6% as Idaho ID1, Tongluo and peer capacity arrive; management expects the year to be tighter than 2026, which we read as a plateau rather than further price increases. Fiscal 2029–30, the downturn (assumed). Revenue falls 25% and then 14% as new clean rooms built for peak prices ramp: average prices fall about 35% and then 25% while bits grow in the mid-teens. Operating margin troughs at 52% — far above the 2019 and 2023 troughs, because roughly a third of revenue sits under floors that management says yield above-peak margins, and because depreciation, not cash cost, is the main cost growth. Fiscal 2032–33, normal (anchored to the sell side). Free cash flow of about $86bn, the level consistent with the sell side’s through-cycle EPS of $75–90 (Wells Fargo, September 2026) and with BofA’s estimate that EPS stays near $100 even in a conventional downturn. Capex settles at about 20% of revenue, against a historical mid-30s and management’s description of today’s intensity as low.
The 10.0% discount rate is the market-risk rate we use for large, net-cash semiconductor companies, and it is the same in every scenario. The memory cycle is priced in the cash flows and the scenario weights, not by adding a premium to the rate, so it is charged once. Buybacks are not modelled: repurchases at fair value do not change value per share.
Share count and bridge. 1,133m shares: 1,129.4m on the cover of the Form 10-Q for the quarter ended 28 May 2026 (17 June 2026), plus an estimated 1.6m of net vesting in the fourth quarter and about 2m vested options. Cash, marketable investments and restricted cash of $73.48bn at 3 September 2026, less $0.49bn of current and $4.69bn of long-term debt, give headline net cash of $68.3bn. The $12.7bn of customer cash deposits received under the strategic customer agreements is refundable over the agreements’ terms and is deducted as debt-like, giving $55.6bn. Unvested restricted stock units are not added, because stock-based compensation is charged as a cost. The fiscal 2026 Form 10-K had not been filed at publication; the share count is trued up when it is.
The number above is not a fact about Micron. It is what these assumptions produce, and the one that decides it is not visible in any filing: the level of free cash flow Micron earns once the cycle has turned and recovered. Each $1bn a year of steady-state free cash flow is worth about $6.70 a share. The rating holds for steady-state free cash flow between about $57bn and $105bn a year, or a normal operating margin between about 40% and 57%, with the rest of our cycle path unchanged. Below that range the band’s upper edge falls under the price and the reading turns overvalued; above it, undervalued. On the full bear path the band would be $620–828, with the shares overvalued — which is why the behaviour of the contracted floors in the first downturn is the variable to watch.
Each card loads its published path into the model above and prints that path’s value: $702, $1,113, $1,620 and $2,247. All four use the same 10.0% rate and 3.0% terminal growth, so the differences between them are operating paths only; the memory cycle, floor failure and industry overbuild sit in the paths and the weights, not in the discount rate. Weighting the four at 25 / 45 / 25 / 5 gives $1,193.
| Input | Bear (25%) | Base (45%) | Bull (25%) | Super-bull (5%) |
|---|---|---|---|---|
| Cycle shape | FY27 peak; pricing cracks FY28; trough FY30 | Plateau FY28; downturn from FY29; trough FY30; normal from FY32 | Tight through FY29; shallow FY30–31 correction | No conventional downturn |
| Revenue FY27–33, $bn | 262, 215, 150, 140, 158, 170, 176 | 276, 305, 229, 198, 232, 240, 248 | 284, 330, 345, 290, 300, 315, 325 | 290, 345, 375, 385, 395, 410, 425 |
| GAAP operating margin | 81, 70, 48, 40, 43, 41, 41% | 83, 80, 64, 52, 55, 52, 51% | 84, 83, 80, 67, 64, 62, 61% | 85, 84, 82, 78, 74, 72, 70% |
| Contracted floors | Renegotiated at the margin | Honoured at face value | Not binding: prices above floors | Coverage ~50%, renewed past 2030 |
| Net capex FY27–33, $bn | 55, 52, 36, 30, 32, 35, 37 | 58, 65, 52, 42, 45, 48, 51 | 60, 70, 72, 55, 58, 61, 64 | 62, 74, 78, 76, 76, 78, 80 |
| Free cash flow, FY2033 | $49bn | $86bn | $134bn | $202bn |
| Common to all | WACC 10.0%, terminal growth 3.0%; cash tax 15.5% in FY2027 (guidance), 16% in FY2028, 16.5% in FY2029 and 17% from FY2030; stock-based compensation charged inside the GAAP margin; net cash after customer deposits $55.6bn; 1,133m shares | |||
| Value per share | $702 | $1,113 | $1,620 | $2,247 |
| $bn unless stated, FY to early September | FY27 | FY28 | FY29 | FY30 | FY31 | FY32 | FY33 |
|---|---|---|---|---|---|---|---|
| Revenue | 276 | 305 | 229 | 198 | 232 | 240 | 248 |
| Growth | +107% | +11% | −25% | −14% | +17% | +3% | +3% |
| GAAP operating margin | 83% | 80% | 64% | 52% | 55% | 52% | 51% |
| EBIT | 229 | 244 | 147 | 103 | 128 | 125 | 126 |
| Cash tax rate | 15.5% | 16.0% | 16.5% | 17.0% | 17.0% | 17.0% | 17.0% |
| NOPAT | 194 | 205 | 122 | 85 | 106 | 104 | 105 |
| + D&A | 12 | 16 | 21 | 25 | 28 | 31 | 33 |
| − Net capex | (58) | (65) | (52) | (42) | (45) | (48) | (51) |
| − Increase in working capital | (11) | (4) | 10 | 6 | (5) | (1) | (1) |
| Free cash flow | 137 | 152 | 101 | 74 | 84 | 86 | 86 |
| Capex / revenue | 21% | 21% | 23% | 21% | 19% | 20% | 21% |
| GAAP EPS derived | $171 | $181 | $109 | $76 | $94 | $92 | $93 |
Sonde estimates. EBIT is GAAP, with stock-based compensation charged. EPS adds interest income and uses 1,147m diluted shares; it is shown for comparison with consensus and is not a valuation input. Fiscal 2026 contained 53 weeks, so fiscal 2027 growth is understated by about 2%.
| Base case | $bn | $ per share |
|---|---|---|
| Present value of free cash flow, FY2027–28 (the peak plateau) | 264 | 233 |
| Present value of free cash flow, FY2029–33 (downturn and recovery) | 287 | 253 |
| Present value of terminal value (from FY2033 free cash flow of $86bn, 3% growth) | 654 | 577 |
| Enterprise value | 1,205 | 1,064 |
| + Cash, marketable investments and restricted cash, 3 September 2026 | 73.5 | 65 |
| − Debt, current and long-term | −5.2 | −5 |
| − Customer deposits under the agreements, refundable, treated as debt-like | −12.7 | −11 |
| Equity value | 1,261 | $1,113 |
Per-share figures on 1,133m shares. Rows may not add exactly because of rounding.
The grid holds the base operating path fixed and varies the discount rate against terminal growth. The whole grid is the published fair-value band, $968–1,334: WACC 9–11% by terminal growth 2.5–3.5%.
| WACC ↓ / terminal growth → | 2.5% | 3.0% | 3.5% |
|---|---|---|---|
| 9.0% | $1,206 | $1,265 | $1,334 |
| 9.5% | $1,134 | $1,183 | $1,240 |
| 10.0% base | $1,072 | $1,113 | $1,160 |
| 10.5% | $1,017 | $1,052 | $1,092 |
| 11.0% | $968 | $999 | $1,033 |
The $1,065.11 price lies inside the band, 10% above its lowest cell and 20% below its highest: fairly valued. It sits between the 10.0% and 10.5% rows at 3% growth. Run across the same grid, the probability-weighted value spans $1,035–1,436. Two sensitivities matter more than the grid. Timing: the same downturn arriving a year earlier, in fiscal 2028, gives $1,069; a year later, in fiscal 2030, $1,168. Through-cycle cash flow: each $1bn a year of steady-state free cash flow is worth about $6.70 a share. The grid is reproducible from the model above by moving the last two controls.
Modal base case $1,113, 4.5% above the price. Fair-value band $968–1,334, the WACC 9–11% by growth 2.5–3.5% grid, which anchors the rating; the price is inside it: fairly valued. Probability-weighted value $1,193 across the four scenarios at our own weights, 12% above the price and 7% above the base case. The skew is to the upside: the bull and super-bull outcomes sit further above the base than the bear sits below it, because the bear’s floor is a balance sheet with $68bn of net cash and contracted minimum prices.
| Single input changed, all else at base | Price implies | Our base |
|---|---|---|
| Steady-state free cash flow from FY2033 | $79bn | $86bn |
| Discount rate | 10.39% | 10.00% |
| Revenue path, every year scaled | 0.955× base | 1.000× |
| Timing of the downturn | FY2028 ($1,069) | FY2029 |
The market and the model are close. The price is consistent with a downturn one year earlier than ours, or with through-cycle cash flow about 8% below the sell side’s own through-cycle estimate, or with a discount rate under half a point higher. None of these is a stretch, and the burden of disagreement is light in both directions. The large gap is between the price and the average analyst target of about $1,350, which our base would justify only if the downturn were pushed out beyond fiscal 2030 or never came.
| Scenario | Value | Against price | Weight | Contribution |
|---|---|---|---|---|
| Bear | $702 | −34% | 25% | $175 |
| Base | $1,113 | +4% | 45% | $501 |
| Bull | $1,620 | +52% | 25% | $405 |
| Super-bull | $2,247 | +111% | 5% | $112 |
| Probability-weighted | $1,193 | +12% | 100% | $1,193 |
We weight the bear at 25% rather than higher because the contracts and the deposits are evidence a normal memory bust has never had; we weight it at 25% rather than lower because those contracts have never been tested.
| Risk | Priced in | Not also in |
|---|---|---|
| Memory price cycle | Cash flows: the FY2029–30 downturn in the base case, earlier and deeper in the bear | WACC; no exit multiple is used |
| Contracted floors failing | The bear case (floors renegotiated) and its 25% weight | The base case, which credits floors in full |
| Industry overbuild | Capex paths and FY2029–30 pricing in the base; FY2028 pricing in the bear | Terminal growth |
| Customer concentration | The bear revenue path | No separate haircut |
| HBM share loss to Samsung | Bear and base revenue; the base assumes share broadly held | — |
| China access and IP litigation | Bear revenue; litigation costs inside operating margins | No discrete reserve |
| Tax and trade policy | Cash tax rising to 17% in every scenario | No tariff scenario |
| Market risk | The 10.0% WACC | No company-specific premium |
| Taiwan supply disruption | Not modelled in any scenario: an unpriced tail, disclosed here | — |
Every listed memory maker trades at single-digit multiples of current earnings. The useful question is not which is cheapest, but whether the sector’s discount already pays for the downturn the multiples imply.
| Company | Mkt cap | EV | Growth y/y, latest qtr | Op. margin | EV / run-rate revenue | EV / run-rate op. profit | P/E, guided or run-rate |
|---|---|---|---|---|---|---|---|
| Micron $1,065.11 | $1,203bn | $1,135bn | +379% | 80.7% | 5.6× | 7.0× | 7.0× FQ1 guide |
| SK hynix ADS | $1,342bn | $1,265bn | +257% | 76% | 5.4× | 7.1× | 9.9× Q2 run-rate |
| Samsung Electronics ≈ | $1,157bn | $1,033bn | +130% | 52% | 2.0× | 3.9× | 5.8× Q2 run-rate |
| SanDisk | $254bn | $249bn | +372% | 78.5% | 7.0× | 8.9× | 9.7× FQ1 guide |
| Kioxia ≈ | $189bn | $188bn | +416% | 72% | 4.2× | 5.8× | 5.8× Q2 guide |
Prices are 30 September 2026 closes (MU, the SK hynix ADS and SNDK on Nasdaq; Samsung on the KRX), except Kioxia, at its 24 September 2026 Tokyo close before its 3-for-1 split (the 30 September close was not verified). FX on 30 September: ₩1,356.65 and ¥157.56 per dollar. Shares: Micron 1,129.4m (Form 10-Q cover, 17 June 2026); SK hynix 728.9m (711.1m per its prospectus plus 17.79m issued in the July ADS offering); Samsung 5.85bn common shares, derived from its KRX market value, preferred shares excluded; SanDisk 146m (Form 10-K, 3 July 2026); Kioxia 545.5m, derived from its buyback notice. Net cash: Micron $68.3bn at 3 September 2026, before the customer deposits we treat as debt-like in the model; SK hynix $77.7bn at 30 June 2026 plus $26.5bn of gross ADS proceeds received on 14 July; Samsung $123.5bn and Kioxia $1.2bn at 30 June 2026 as reported; SanDisk $4.8bn of cash at 3 July 2026 and no debt. Run-rate is the latest reported quarter annualised, with Micron’s 14-week quarter scaled to 52 weeks. P/E bases: guided first-quarter GAAP EPS times four (Micron, SanDisk); second-quarter operating profit times four at a 24% tax rate (SK hynix, Samsung); guided second-quarter net income times four (Kioxia). Rows marked ≈ carry the stated approximations.
Cheap, fair or expensive? In line with SK hynix, below SanDisk, above Samsung and Kioxia. On enterprise value to run-rate operating profit, Micron (7.0×) and SK hynix (7.1×) are essentially level. SanDisk carries a premium for its pure exposure to data-centre storage and its own contract book; Samsung’s discount reflects a conglomerate whose phone and appliance businesses roughly broke even in the second quarter; Kioxia sold off after its September-quarter guidance disappointed. On forward guidance Micron is cheaper still: enterprise value is 4.6 times the annualised first-quarter revenue guide. Price to free cash flow is 19.3 times trailing fiscal 2026 adjusted free cash flow but under 10 times the fourth quarter annualised. The PEG ratio is about 0.14 for Micron and near zero for every peer, because earnings tripled or more in a year; for a cyclical at a peak it carries no information, so we do not rank on it.
What the multiple is assuming. Seven times run-rate earnings, capitalised at a normal market multiple, implies the market expects average earnings over the next decade to be well under half of today’s. That is the downturn our model draws explicitly. Because the sector discount is the market’s price for the cycle, we apply no further multiple haircut: the cycle is charged once, in the cash flows. What would expand it: a downturn that arrives and leaves margins above prior peaks with floors honoured, the only real proof; buybacks shrinking the share count at today’s prices; HBM pricing holding into 2028. What would contract it: the first publicly renegotiated agreement; spot DRAM prices turning before contract prices; cuts to hyperscaler capex guidance; CXMT or Samsung pricing for share.
Intrinsic value is not a price target and this report does not publish one. But the market prices memory shares on next-twelve-months earnings, and the multiple regime it applies near a peak is worth being explicit about. Memory shares near past peaks have traded at roughly five to eight times next-twelve-months earnings. On our base fiscal 2027 GAAP EPS of about $171, that range spans about $853–1,365 and brackets the current price. This is a market-regime estimate, not a fair-value estimate, and it is not the anchor of the rating.
Through-cycle cash flow is set at the sell side’s level, not at history’s. The base case normalises free cash flow at about $86bn, consistent with sell-side through-cycle EPS of $75–90. Micron has never earned anything close to that through a downturn: its prior peak EPS was $11.51, in fiscal 2018, and it lost money in fiscal 2023. On our cycle path, $65bn a year gives $972, $50bn gives $871, and a return to pre-AI economics scaled for volume, about $35bn, gives $770. The first two remain inside or near the band; the third does not. The floors are untested. Agreement terms, counterparties and enforcement mechanics are not public; remaining performance obligations are stated at minimum price, and some extend beyond 2030. Whether a hyperscaler pays a floor price far above spot in a glut is a question no memory contract has yet answered. The fiscal 2026 Form 10-K is not yet filed. The share count, the obligation schedule, fiscal 2026 customer concentration, the $500m patent charge and the fourth-quarter jump in other non-current liabilities will be checked against it. Industry inputs are assumptions. Bit growth and price paths are ours; management sees NAND supply-constrained through 2028, while TrendForce has forecast NAND moving towards balance in 2027. Peer data carry approximations. Samsung excludes preferred shares; Kioxia’s price predates its split and our data date; SK hynix’s net cash adds its July ADS proceeds gross of fees; Korean and Japanese figures are converted at 30 September rates. Unmodelled: buybacks, which are value-neutral at fair value; the timing of deposit refunds; a Taiwan supply disruption; and any tariff regime.
Micron reported its fourth quarter and fiscal 2026 on 30 September, after the close. The quarter ran 14 weeks to 3 September 2026; the first quarter of fiscal 2027 runs 13 weeks to early December. Revenue of $54.23bn cleared the top of the $49–51bn guidance range and a consensus of about $51.1bn, and non-GAAP EPS of $33.42 beat the $31 midpoint.
| Metric | Q4 FY26 | Q3 FY26 | Change | Why it matters |
|---|---|---|---|---|
| Revenue | $54.23bn | $41.46bn | +31% | +21% per week; a 14-week quarter |
| DRAM revenue | $39.8bn | $31.3bn | +27% | Prices up in the high teens, bits up in the mid-single digits |
| NAND revenue | $14.1bn | $9.9bn | +42% | Prices up about 30%, bits about 10%; data-centre SSDs nearly $10bn |
| Gross margin, GAAP | 86.8% | 84.6% | +2.2 pts | Pricing ahead of costs; Q1 guided lower at 85.95% |
| Operating margin, GAAP | 80.7% | 80.4% | +0.3 pts | Operating expenses nearly doubled: bonuses and a $500m patent charge |
| Diluted EPS, GAAP | $32.87 | $24.67 | +33% | Non-GAAP $33.42 against a $31 guide midpoint |
| Operating cash flow | $43.97bn | $25.39bn | +73% | 81% of revenue |
| Net capex | $10.77bn | $7.08bn | +52% | Rising towards about $12.5bn a quarter in H1 FY27 |
| Adjusted free cash flow | $33.20bn | $18.30bn | +81% | Excludes $12.3bn of customer deposits received |
| Days of inventory | 129 | 120 | +9 days | Bonus costs capitalised; end-of-life build-ahead |
| Cash and investments | $73.48bn | ≈$30.1bn | +$43bn | Includes deposits; net cash $68.3bn |
| Signed agreements / obligations | 26 / ≈$150bn | 16 / ≈$100bn | +10 / +50% | Obligations at committed volumes and minimum prices |
Sources: fiscal Q4 2026 results release and call (30 September 2026); fiscal Q3 2026 release and call (24 June 2026). Q3 cash and investments derived from the 28 May 2026 balance sheet. Consensus as reported after the print (LSEG) and not independently verified.
One: another beat-and-raise of unusual size. First-quarter guidance of $61.5bn of revenue and $38.15 of non-GAAP EPS sits well above the $35.40 EPS consensus. Adjusted for the extra week, revenue per week rose about 21% in the fourth quarter, and the first-quarter guide implies another 22% (derived).
Two: the contract book grew by half in a quarter. Signed strategic customer agreements rose from 16 to 26; remaining performance obligations from about $100bn to about $150bn; customer financial commitments from $22bn to $32bn, mostly cash. Three quarters of expected agreement revenue has a defined pricing framework, mostly floor-and-ceiling bands, and some agreements now run into 2031. Management still targets around half of revenue under the agreements.
Three: management says the shortage gets worse, not better. It expects supply and demand to be much tighter in calendar 2027 and 2028 than in 2026, has no line of sight to balance, says more than 75% of fiscal 2027 output is committed, and has agreed most of its 2027 HBM volume at significantly higher prices.
Four: capex is going up, and it is going into concrete. First-quarter capex of about $11.5bn, first half about $25bn and the second half higher; construction spending is growing faster than equipment, aimed at clean rooms from late 2028. No full-year figure was given.
Five: the cash is coming back. From 9 December 2026, the second anniversary of its CHIPS agreements, Micron intends to increase capital returns and, over time, return all excess cash, mainly by buyback. It expects to reach its target cash level by the end of the first quarter and to seek a larger authorisation than the $2.2bn now in place.
Guidance for the first quarter of fiscal 2027: revenue of $61.5bn ± $1.5bn; GAAP gross margin about 85.95% (non-GAAP 86.25%); GAAP operating expenses about $2.31bn; GAAP EPS of $37.84 ± $1.00 on about 1.15bn diluted shares; a tax rate of about 15.5%. Management called the first quarter the floor for gross margin in fiscal 2027, attributing the dip to roughly $1bn of higher costs — mostly incentive compensation absorbed into inventory in the fourth quarter — plus start-up costs. Full-year operating expenses are expected to rise by about $2.5bn, and revenue to grow sequentially in every quarter.
The tone was confident to the point of expansive: AI framed as becoming “super intelligence”, the outlook described as stronger than last quarter. On the hard questions — buyback size, capex, a large customer trimming HBM content — the answers were firm on direction and silent on numbers, a familiar register at the top of a memory cycle, and so far one the numbers have supported. UBS and BofA pushed on the size of capital returns and the minimum cash level; UBS on long-run capital intensity; TD Cowen on a large customer reducing HBM content, on freed wafers flooding conventional DRAM, and on equipment availability; BofA on whether 2028 is the earnings peak; JPMorgan on how far agreement coverage can go.
| Positive surprises | Negative surprises |
|---|---|
| The size of the beat and of the first-quarter guide | Gross margin guided down in the first quarter |
| Remaining obligations up about $50bn in a quarter; some agreements extended to 2031 | Operating expenses up $1.1bn in a quarter, including $300m of community investments and a $500m patent-licence charge |
| First-quarter free cash flow guided “significantly higher” than the fourth quarter’s $33bn | Server memory content growth “modestly lower” than expected |
| NAND revenue up 42% on the quarter | Capex raised without a full-year number |
| Two rating-agency upgrades to BBB+ | Days of inventory up nine |
What to watch next. The fiscal 2026 Form 10-K in October, for the obligation schedule, customer concentration and the $500m charge; the buyback authorisation after 9 December; the first-quarter results in December, where the test is a gross margin stepping back up; a full-year capex figure; and calendar first-quarter 2027 contract prices.
Micron designs and manufactures memory and storage chips: DRAM, the working memory every processor needs; NAND flash, the storage inside solid-state drives and phones; and a small amount of NOR. It is one of three companies that make DRAM at scale, alongside SK hynix and Samsung, and the only one headquartered in the United States. The model is vertically integrated and capital-heavy: Micron runs its own wafer fabs in Idaho, Virginia, Taiwan, Japan and Singapore, packages and tests chips in Asia, and spends tens of billions of dollars a year on equipment and clean rooms. Revenue is product sales; there is no recurring software or service stream. In fiscal 2026 DRAM was $100.7bn, 76% of revenue, and NAND $31.8bn, 24%.
Since fiscal 2025 Micron has reported four business units organised by customer rather than by product:
| Business unit, Q4 FY2026 | Revenue | Share | Gross margin | Op. margin | Main products |
|---|---|---|---|---|---|
| Cloud Memory | $16.28bn | 30% | 83% | 76% | HBM and high-capacity server DRAM for hyperscalers |
| Core Data Center | $18.00bn | 33% | 90% | 85% | Server DRAM and data-centre SSDs for OEMs |
| Mobile and Client | $13.11bn | 24% | 90% | 88% | Phone and PC DRAM and NAND |
| Automotive and Embedded | $6.82bn | 13% | 84% | 79% | Automotive, industrial and consumer |
Source: fiscal Q4 2026 results release, 30 September 2026. Quarter of 14 weeks ended 3 September 2026.
Fiscal 2026 revenue rose 256% to $133.2bn, and the quarterly path shows how: $13.6bn in the first quarter, $23.9bn in the second, $41.5bn in the third and $54.2bn in the fourth. Almost all of it was price. Data-centre SSD revenue reached nearly $10bn in the fourth quarter, more than two thirds of NAND, and HBM grew faster than the company total. The two data-centre units together were 63% of fourth-quarter revenue. That composition is the central fact of this report: revenue that rises on price falls on price. Volumes grow roughly 20% a year across cycles; prices are what swing.
AI has turned memory into a bottleneck. Accelerators need high-bandwidth memory stacked beside the processor; inference at long context needs large server DRAM pools and fast SSDs for key-value caches. IDC estimated in July 2026 that the DRAM market would reach roughly $500bn in 2026 and approach $1tn in 2027; Micron’s fiscal 2026 DRAM revenue implies a share of about a fifth (derived; calendar and fiscal years differ).
Supply is slow to respond. HBM consumes about three times the wafer area of standard DRAM per bit, node transitions now add less output per wafer, and new clean rooms take two to three years to build and longer to ramp. Micron expects industry DRAM bits to grow in the mid-20s percent in calendar 2026 and in the low 20s in 2027–28, NAND in the low 20s and then the mid-20s, with both markets supply-constrained throughout. TrendForce broadly agrees on DRAM — it estimated a 1–2% shortfall for 2026, widening in 2027 — but has forecast NAND moving towards balance in 2027, the clearest independent dissent from management’s outlook.
The industry is learning to contract. Long-term agreements with floors are spreading: SanDisk reported $91.1bn of remaining obligations including deals signed after its June quarter, Kioxia targets half of its 2028 volume under such agreements, and SK hynix has signed multi-year contracts with price ranges. If they hold in a downturn, troughs become shallower for everyone.
Demand is being rationed at the edge. Management expects PC and smartphone units to fall by double digits even as revenue in those markets grows; Samsung’s phone business booked its first-ever operating loss in the second quarter of 2026 on component costs, and customers are trimming memory per server to ship more servers. Rationing is a sign of shortage, but it is also how demand is lost when prices stay high.
Capacity is coming. Micron’s Idaho ID1 fab begins wafer output in mid-2027, ID2 and a Japan expansion in late 2028, a Singapore NAND fab in the second half of 2028 and New York in 2030. SK hynix raised $26.5bn in its July 2026 US listing, earmarked for its Yongin fab and advanced packaging; Kioxia plans about ¥470bn a year of capex through fiscal 2028; China’s CXMT keeps adding commodity DRAM. Every one of these fabs was justified at today’s prices.
| Company | Position | Latest quarter | What it means for Micron |
|---|---|---|---|
| SK hynix | DRAM and HBM leader; Nasdaq ADSs since July 2026 | Q2 2026 revenue ₩79.3tn (+257%), operating margin 76% | Sets the HBM benchmark; well funded to add capacity |
| Samsung Electronics | Largest memory capacity; catching up in HBM | Q2 2026 chip-division revenue ₩127.5tn, operating profit ₩89.2tn | Its capacity decisions decide when the cycle turns |
| CXMT (China) | Fast-growing commodity DRAM supplier | Not separately reported here | Limited AI threat today; a price-setter for commodity DRAM in a downturn |
| SanDisk and Kioxia | NAND joint-venture partners; data-centre SSD rivals | SanDisk fiscal Q4 2026 revenue $8.97bn, gross margin 84.6% | Compete with Micron’s number-two position in data-centre SSDs |
| YMTC (China) | Chinese NAND maker | Not separately reported here | Litigant against Micron; NAND supply outside the oligopoly |
Sources: SK hynix 2Q26 release (29 July 2026); Samsung Electronics Q2 2026 results (30 July 2026); SanDisk fiscal Q4 2026 release (5 August 2026).
| Fiscal year | FY18 | FY19 | FY23 | FY24 | FY25 | FY26 | FY27 base |
|---|---|---|---|---|---|---|---|
| Revenue, $bn | 30.4 | 23.4 | 15.5 | 25.1 | 37.4 | 133.2 | 276 |
| Gross margin | 58.9% | 45.7% | (9.1)% | 22.4% | 39.8% | 80.7% | ≈87% |
| Operating margin | 49.3% | 31.5% | (37.0)% | 5.2% | 26.1% | 74.6% | 83% |
| Diluted EPS | $11.51 | $5.51 | $(5.34) | $0.70 | $7.59 | $74.33 | $171 |
GAAP. Sources: Micron Form 10-K filings, fiscal 2018–2025; fiscal Q4 2026 release. The fiscal 2027 column is a Sonde estimate.
Fiscal 2026 operating cash flow was $89.7bn; net capex, after $3.3bn of government incentives, was $27.4bn, and adjusted free cash flow $62.3bn. Customer deposits of $12.7bn were booked as financing inflows and are excluded from that free cash flow — a conservative presentation. Debt fell from $14.6bn to $5.2bn, and cash and investments rose to $73.5bn. Fiscal 2026 contained 53 weeks and its fourth quarter 14. The 2018 and 2023 columns are the reminder: the previous peak operating margin was 49%, and five years later the company lost money for a full year.
Disclosed drivers: HBM repricing in 2027, with most calendar 2027 HBM supply agreed at significantly higher prices, narrowing HBM’s margin gap with conventional DRAM; more than 75% of fiscal 2027 output committed and talks moved on to 2028; data-centre SSDs at nearly $10bn a quarter and a fifth consecutive year of share gains expected in 2026; 1-gamma DRAM and G9 NAND now the largest nodes, the next nodes reaching volume in the second half of 2027, and a custom HBM4E developed with NVIDIA; level-4 autonomous vehicles carrying more than 200GB of DRAM, an order of magnitude above today’s driver-assistance systems; and, from 9 December 2026, a step-up in capital returns. Assumed in our base case: fiscal 2027 revenue of $276bn, a plateau in fiscal 2028, and a downturn from fiscal 2029 cushioned by floors honoured at face value.
Risks, in order of weight: the cycle itself, since pricing drove the entire upswing and would drive the downswing; the industry’s supply response, sized at peak prices; contract quality, with terms not public, floors untested in a downturn and ceilings capping the upside; customer concentration, with one customer 17% of fiscal 2025 revenue and the top ten about half; China and geopolitics, with Chinese critical-infrastructure operators unable to buy Micron products since a 2023 cybersecurity review and much of DRAM output in Taiwan; intellectual property, with YMTC suits continuing and a new Netlist complaint before the US International Trade Commission; and demand, with hyperscaler capex growth widely forecast to slow sharply by 2028 and high memory prices already shrinking PC and phone units. Section 07 scores each.
The question is not whether Micron has an advantage over its peers, but whether the three DRAM makers together now have one over their customers.
| Source | Micron | SK hynix | Samsung | Assessment |
|---|---|---|---|---|
| Brand | 2 | 2 | 3 | Engineering reputation matters in qualification; no pricing power from brand |
| Network effects | 1 | 1 | 1 | None |
| Switching costs | 3 | 3 | 3 | Qualification takes months; HBM is co-designed with the accelerator; the agreements lock in volume |
| Cost advantage | 3 | 4 | 3 | SK hynix leads on HBM yields; Micron’s node lead offsets its smaller scale |
| Scale | 3 | 4 | 5 | Samsung has the most capacity and the deepest pockets |
| IP and technology | 4 | 4 | 3 | Micron’s 1-gamma lead; SK hynix’s HBM packaging; Samsung behind in HBM |
| Distribution | 3 | 3 | 4 | All three sell direct to the same few hundred buyers |
| Regulation and geography | 4 | 2 | 2 | US supply and CHIPS support favour Micron; China access cuts against it |
| Data | 1 | 1 | 1 | Not relevant |
| Customer loyalty and contracts | 4 | 4 | 3 | Micron’s 26 agreements; SK hynix’s long HBM partnership with NVIDIA |
| Total | 28 / 50 | 28 / 50 | 28 / 50 | Equal totals, different shapes |
The totals come out level, and that is the finding. No member of the oligopoly has a durable advantage the others lack: Micron’s edges — node leadership, US supply, the most aggressive contracting — are matched by SK hynix’s HBM lead and Samsung’s scale. Whatever moat exists is the industry’s: three DRAM makers with enough concentration, and now enough contract structure, to avoid the price wars of 2019 and 2023.
Strongest advantages. Technology timing: Micron reached 1-gamma first and has its next nodes on track for the second half of 2027. The US manufacturing footprint, at a time when customers price geopolitical risk. And a contract book that, if honoured, turns part of a commodity into something closer to a take-or-pay supply arrangement.
Biggest threats. Samsung qualifying HBM4 at scale and using its capacity to buy share; CXMT pricing commodity DRAM aggressively once its capacity matures; the industry’s own clean-room programme landing in 2028–29; and customers designing around memory cost, which management already acknowledges in lower-than-expected content growth per server.
Direction: expanding, from a low base. Historically memory had no moat: margins went to zero in every bust. The new contract structure and HBM co-design widen it to narrow. The proof will arrive only in a downturn, which is why the market will not yet pay a full multiple for it.
| Concern | Severity | Evidence and why it matters |
|---|---|---|
| Revenue quality | HIGH | Growth is price, not volume: fourth-quarter DRAM prices rose in the high teens on bits up in the mid-single digits, and NAND prices about 30%. Revenue built on price reverses on price (Q4 FY26 call) |
| Capex into the peak | MED-HIGH | Net capex of $27.4bn in fiscal 2026; the first half of fiscal 2027 alone about $25bn, and the second half higher. Construction for late-2028 clean rooms is the classic seed of the next glut, across the whole industry |
| Customer concentration | MEDIUM | One customer was 17% of fiscal 2025 revenue, mainly in Cloud Memory; the top ten about half (FY25 10-K). The fiscal 2026 figure is not yet disclosed |
| Management language | MEDIUM | “No line of sight” to supply-demand balance and “super intelligence” framing: the vocabulary of every memory peak, though so far matched by results |
| Margin pull-through | MEDIUM | First-quarter gross margin guided down about 0.8 points on about $1bn of bonus and start-up costs; fiscal 2027 operating expenses up about $2.5bn; $300m of discretionary community spending in the fourth quarter |
| Insider activity | MEDIUM | The chief executive sold about 177,000 shares between May and August 2026 at $536–1,192 under a 10b5-1 plan for up to 200,000 (derived from Forms 4); a director also sold; no open-market purchases found |
| Legal and regulatory | MEDIUM | China restricts purchases by critical-infrastructure operators (2023 review); YMTC patent suits in the US and China; a Netlist complaint now before the ITC (September 2026), though five Netlist patents were invalidated on appeal the same month |
| Accounting | LOW | A $500m patent-licence charge excluded from non-GAAP results, counterparty not named; other non-current liabilities up $4.5bn and other non-current assets up $4.6bn in the fourth quarter, unexplained until the 10-K |
| Calendar effects | LOW | Fiscal 2026 had 53 weeks and its fourth quarter 14; the +379% year-on-year comparison is flattered by about 8%, and fiscal 2027 comparisons get slightly harder |
| Cash-flow quality | LOW | Clean. Receivable days fell to about 65 from about 75 a year earlier (derived); deposits sit in financing, not free cash flow; the inventory build is explained |
| Dilution | LOW | Basic shares up 0.7% in a year; stock-based compensation $1.33bn, about 1% of revenue |
| Debt | LOW | $5.2bn of debt against $73.5bn of cash and investments; rated BBB+ by all three agencies |
Overall red-flag score: 4 out of 10, moderate. The accounting and the balance sheet are clean; the flags cluster in cyclicality and the industry’s capacity response.
| Strength | Score / 10 | Evidence and why it matters |
|---|---|---|
| Customers prepaying | 8 | $12.7bn of customer cash on the balance sheet and $32bn of commitments. Buyers do not prefund supply they expect to be plentiful |
| Balance sheet and returns | 8 | Net cash of $68.3bn; first-quarter free cash flow guided above $33bn; an intention to return all excess cash from December. At more than $120bn a year of free cash flow, buybacks could retire several percent of shares a year |
| Contracted floors | 7 | About $150bn of obligations at minimum prices; management says the floors imply margins above any prior peak. No memory downturn has had this cushion |
| Technology position | 6 | 1-gamma DRAM and G9 NAND are the largest nodes; next nodes in volume in the second half of 2027; custom HBM4E with NVIDIA; data-centre SSD share up five years running |
| US-based supply | 6 | The only US-headquartered DRAM maker, building in Idaho, Virginia and New York; supply-chain security is now a purchasing criterion |
| Conservative disclosure | 5 | Obligations stated at minimum price; deposits kept out of free cash flow; the first-quarter margin dip flagged in advance |
| Optionality | 4 | Physical AI, with vehicles carrying more than 200GB of DRAM; key-value-cache SSDs; LPDDR in servers. Valued at roughly zero in the base case |
Overall green-flag score: 7 out of 10, strong. The contract book and the balance sheet are new to this industry, and both are untested.
| Dimension | Score / 5 | Assessment |
|---|---|---|
| CEO track record | 4 | Sanjay Mehrotra, chief executive since 2017 and a SanDisk co-founder. Carried the company through the 2019 and 2023 downturns without a balance-sheet crisis, closed the HBM gap and pioneered the strategic-agreement structure. Not yet tested by a bust with this capacity programme |
| CFO credibility | 4 | Mark Murphy, chief financial officer since 2022. Conservative presentation: deposits kept out of free cash flow, obligations at minimum price, the first-quarter margin dip flagged in advance |
| Guidance accuracy | 4 | Fourth-quarter revenue and EPS above the top of the range; third-quarter revenue beat guidance by about a quarter. Beating guidance in an upcycle is easier than in a downturn |
| Transparency | 3 | Good on the agreement framework; thin on terms, counterparties, deposit-refund schedules, full-year capex and the $500m patent charge |
| Capital allocation | 4 | Debt cut from $14.6bn to $5.2bn and cash built during the peak. The test is the capacity programme: construction capex for late-2028 clean rooms is the industry’s perennial error, though the agreements’ visibility mitigates it |
| Acquisitions | 3 | No material M&A; about $1bn of non-marketable equity investments in fiscal 2026. Neutral |
| Buybacks | 3 | Only $0.65bn in fiscal 2026 while CHIPS-related constraints applied. The commitment to return all excess cash from December is promising; execution is unproven |
| Dilution | 5 | Basic shares up 0.7% in a year; stock-based compensation about 1% of revenue |
| Insider ownership | 2 | The chief executive holds about 287,000 shares directly and 607,000 indirectly, about $950m at $1,065, but has sold steadily under a 10b5-1 plan; no open-market purchases found |
| Compensation | 3 | Fiscal 2026 incentive pay raised for every employee; executive pay not reviewed here pending the 2026 proxy |
| Board quality | 3 | Chair and chief executive roles combined. Not reviewed in depth |
| Communication style | 4 | Consistent, data-rich and responsive on direction; reluctant on numbers; promotional framing at the top of the cycle |
| Average | 3.5 | Good operators with a clean balance-sheet record |
Do they act like long-term owners? Mostly, on the balance sheet; not yet proven on capacity. Repaying debt and building cash at the top, keeping customer deposits out of free cash flow and contracting forward are what owners of a cyclical should do. Raising capex while seeing no line of sight to balance is what every memory management has said at every peak; the agreements give this one a better reason than its predecessors had. Steady 10b5-1 selling by the chief executive reveals little, but no insider is buying at these prices.
| Window | Event | Upside | Downside | Confidence · source |
|---|---|---|---|---|
| Oct 2026 | Fiscal 2026 Form 10-K | Obligation schedule front-loaded; concentration lower | Unfavourable agreement terms; detail on the $500m charge | High · required filing |
| Oct 2026 | Samsung and SK hynix third-quarter results (SK hynix 27 October) | Peers confirm pricing and capex discipline | Capex raised; HBM share shifts | High / medium · company calendars |
| 29 Oct 2026 | Quarterly dividend of $0.15 | — | — | High · Q4 FY26 release |
| Mid-Nov 2026 | NVIDIA quarterly results | Accelerator demand and HBM content up | Platform delays; content cuts | Medium · company calendar |
| 9 Dec 2026 | Capital-return step-up and a new buyback authorisation | A large authorisation, executed quickly | A token programme; cash kept for capex | High · Q4 FY26 call |
| Mid–late Dec 2026 | Q1 FY27 results and the Q2 guide | Gross margin steps back up; capex framed for the year | The margin dip persists; content cuts spread | High · timing typical |
| Jan 2027 | Calendar Q1 2027 contract-price settlements | Further DRAM increases | The first flat or down quarter | Medium · industry trackers |
| Early 2027 | Singapore HBM packaging output begins | An on-time ramp adds HBM supply | Delay; yields | Medium · Q4 FY26 call |
| Jan–Feb 2027 | Hyperscaler capex guidance for 2027 | Raised again | Growth slowing faster than expected | Medium · company calendars |
| Late Mar 2027 | Q2 FY27 results | Gross margin above 88%; more agreements | Price moderation faster than guided | High · timing typical |
| Mar 2027 | NVIDIA GTC: next platforms and custom HBM4E | Content per GPU up | Memory de-specification | Medium · industry calendar |
| Mid-2027 | Idaho ID1 wafer output; Tongluo shipments | Supply for the agreements on time | Adds to industry supply into a peak | Medium · Q4 FY26 call |
| Late Jun 2027 | Q3 FY27 results and 2028 contract talks | 2028 HBM and agreement pricing above 2027 | Prices reset for 2028 | High · timing typical |
| H2 2027 | Next-generation DRAM and NAND nodes to volume | The cost lead widens | Slips | Medium · Q4 FY26 call |
| 2027 | Netlist ITC investigation; YMTC suits | Further invalidations | Exclusion-order risk on modules | Low · 10-Q; press |
| Late Sep 2027 | FY27 results and the FY28 outlook | A tighter 2028 confirmed | A plateau or decline signalled | High · timing typical |
| Through 2027 | Peer capacity: SK hynix Yongin, Samsung, CXMT, Kioxia’s ¥470bn a year | Delays and discipline | Accelerations | Medium · peer disclosures |
Dated where the company or a counterparty has given a date; “timing typical” where we infer from past practice. Micron’s fiscal quarters end on the Thursday nearest the end of November, February, May and August. Macro items to watch throughout: long-dated US Treasury yields, reported at multi-decade highs on 1 October 2026; US trade policy on semiconductors; and AI-infrastructure financing conditions. Confidence reflects the timing of the event, not the direction of its outcome.
BULLGrowth. Revenue went from $11bn a quarter to $54bn in a year and is guided to $61.5bn next quarter, with growth every quarter after that. More than 75% of fiscal 2027 output is committed, 2027 HBM is sold at higher prices, and AI inference keeps expanding the memory each user needs. This is demand outrunning an industry that cannot build fabs faster.
BEARNearly all of that growth was price; bits grew in the mid-single digits last quarter. Prices that rose fourfold can halve, and the industry is now building clean rooms sized for today’s prices. PC and phone units are already falling by double digits and customers are trimming memory per server. Growth in fiscal 2027 is real; growth in 2029 is something nobody can see.
BULLValuation. Seven times guided run-rate earnings, with $68bn of net cash and free cash flow above $120bn a year at the current run-rate. Even BofA’s bear case keeps EPS near $100. Pay 11 times that and the shares are fine; add buybacks and they compound.
BEARSeven times peak earnings has been the price of every memory peak. Micron’s prior peak EPS was under $12 and it lost money three years ago. Sell-side “trough” estimates are always made at the peak. Discount a normal bust and the stock is worth several hundred dollars less than today.
BULLBusiness quality. Twenty-six take-or-pay agreements, $32bn of customer commitments and floors above prior-peak margins. Memory has become a strategic asset that customers prepay for, a structurally better business than the one the old multiples priced.
BEARThe same agreements cap prices at mid-2026 levels for the largest customers. The floors have never been tested; in 2023 buyers cancelled orders, and no hyperscaler will happily pay double the spot price for long. A contract is only as strong as the customer’s need for the next one.
BULLRisks and financials. The strongest balance sheet in the company’s history and $5bn of debt. The downturn, when it comes, starts from margins so high that halving them still leaves them above any prior peak.
BEARThe balance sheet protects the company, not the share price. Capex is doubling into the peak; one customer was 17% of revenue; China access is restricted; much of DRAM output is in Taiwan; and the Netlist and YMTC litigation continues.
BULLManagement and catalysts. Management has beaten its guidance and is about to return all excess cash. December brings the buyback authorisation and another record quarter.
BEARThe chief executive has been selling steadily, no insider is buying, and the language on the call — “super intelligence”, “no line of sight” to balance — is how peaks sound. The next catalysts are calendar 2027 contract prices and peers’ capex plans, and both can turn first.
On the next eighteen months, the bull has the better evidence: guidance, committed output, deposits and pricing all point the same way, and the bear offers base rates rather than data. On the five years after that, the bear has the better evidence: nothing in the contracts has been tested, and the industry is doing what it has always done at the top. The price sits between the two: it already discounts a downturn on roughly the bear’s timetable, but not its depth. The ruling is a draw on valuation and a narrow win for the bull on fundamentals, consistent with a fairly valued rating. Most uncertain: whether the floors can be enforced, and the 2028 supply path. Verify next: the obligation schedule in the 10-K, calendar first-quarter 2027 contract prices, the size of the December buyback, and the capex plans of Samsung and SK hynix.
Micron makes memory chips. DRAM is the short-term memory a computer uses while it works; NAND flash is the long-term storage inside phones and solid-state drives. Artificial-intelligence systems need enormous amounts of both, and a special kind of very fast DRAM, called HBM, sits right next to the AI chips that companies such as NVIDIA make.
By selling chips. The price of memory chips goes up and down a lot, like the price of oil or wheat, because every maker sells nearly the same product. When there is a shortage, prices and profits soar; when the factories catch up, prices crash.
Because there is a shortage of historic size. Micron earned more in one quarter of 2026 than in many whole years before, and it says the shortage will get worse in 2027 and 2028.
Today’s profits are not expected to last. All three big memory makers are building new factories, and when those open, around 2028–29, prices could fall sharply, as they have in every past cycle. What is new this time is that Micron’s biggest customers have signed long contracts with minimum prices, and some have paid cash up front. If those minimum prices hold when the shortage ends, the next downturn will be much milder than the last ones; if they do not, it could look like the old ones. Nobody knows yet, because it has never been tested.
The shares look cheap against today’s profits, but those profits are a peak. Against our estimate of what Micron earns across a whole cycle, the price is about right: we put the business at $968–1,334 a share, most likely about $1,113, and at $1,065 the shares sit inside that range, so we call them fairly valued.
Right: the shortage lasts longer than expected, the long-term contracts protect prices when it ends, and Micron uses its huge cash pile to buy back shares.
Wrong: the new factories open at the same time and prices fall as they always have, the customers push back against the minimum prices, and customers keep using less memory per device because it has become so expensive.
The annual report in October, which will show how much of the contracted revenue falls in each year. The size of the share buyback announced after 9 December. And the prices memory makers agree with their customers for the first months of 2027, which are set in January.
| Status | What it covers |
|---|---|
| Verified from Micron filings and releases | Revenue by product and business unit, margins, EPS, operating cash flow, capex and government incentives, adjusted free cash flow, cash and investments, debt, customer deposits, remaining performance obligations, the number of strategic customer agreements and their pricing framework, customer financial commitments, guidance, capital-return intentions and the dividend — from the fiscal Q4 2026 results release and earnings call (30 September 2026), the fiscal Q3 2026 release and call (24 June 2026), the Form 10-Q for the quarter ended 28 May 2026 and its cover share count (17 June 2026), the Form 10-Q for the quarter ended 26 February 2026, the fiscal 2025 Form 10-K and earlier 10-K filings for fiscal 2018–2024, and Forms 4 filed May–August 2026. |
| Verified from market data and peer filings | The $1,065.11 close of 30 September 2026 and the 52-week range of about $165–1,255; peer closes on 30 September 2026 (Kioxia 24 September); FX on 30 September 2026. Peer figures from SK hynix’s 2Q26 results (29 July 2026) and ADS prospectus (July 2026), Samsung Electronics’ Q2 2026 results (30 July 2026), SanDisk’s fiscal Q4 2026 release and Form 10-K (August 2026) and Kioxia’s Q1 fiscal 2026 results (31 July 2026). |
| Third-party, attributed | Industry estimates from TrendForce (2026 releases) and IDC (July 2026). Sell-side views as reported in the financial press: Wells Fargo, BofA and UBS (September 2026); consensus estimates as reported after the print (LSEG); average and high price targets from MarketBeat (29 September 2026). |
| Our estimate or judgement | Every scenario path in Section 03; fiscal 2027–33 revenue, margins, tax, capex and working capital; the share of revenue under contract in fiscal 2027–30; derived EPS; the 1,133m share count above the filing cover; the treatment of customer deposits as debt-like; peer share counts marked as derived; the 10.0% rate; the 25 / 45 / 25 / 5 weights; the sensitivity grid, the reverse DCF, the $968–1,334 band and the $1,113 central estimate; and all scores in Sections 06, 07 and 08. The model on this page is the model that produced those figures — the scenario cards load the published paths, and the sensitivity grid and the steady-state cash-flow readings are reproducible from the controls. The two timing readings, a downturn one year earlier or later, are separate runs of the base path shifted by a year. |
| Still open | The fiscal 2026 Form 10-K had not been filed at publication: the share count, the year-by-year schedule of remaining performance obligations, fiscal 2026 customer concentration, the counterparty to the $500m patent charge, the fourth-quarter jump in other non-current liabilities and legal reserves will be checked against it. Agreement terms and counterparties are not public. |
Method. Through-cycle discounted cash flow for cyclicals, methodology version 1.3: seven explicit fiscal years, 2027–2033, spanning one full memory cycle in every scenario, valued as at 1 October 2026 with a mid-year convention timed from the 3 September 2026 balance sheet, and a Gordon terminal value growing a normalised fiscal 2033 free cash flow at 3.0%. Free cash flow to the firm is GAAP operating profit, with stock-based compensation charged, after cash tax, plus D&A, less net capex and the increase in working capital. Refundable customer deposits are deducted as debt-like. One discount rate and one terminal growth rate are used across all scenarios, and each material risk is priced once, in the cash flows or the scenario weights. Scenario weights are owned by the analyst, the fair-value band is read from the sensitivity grid, and the reverse DCF is run against the market price. The rating changes to undervalued if the price falls below $968 with no change in the evidence, and to overvalued above $1,334; the conclusion is most sensitive to through-cycle free cash flow, the timing of the downturn and the discount rate, in that order. Ratings are undervalued, fairly valued or overvalued according to where the price sits relative to the band; Sonde does not issue buy, sell or hold recommendations or price targets. The reference price used throughout is the $1,065.11 close of 30 September 2026, and every multiple in this report scales directly with it.