A commodity is printing 84.9% gross margins. Everything depends on the refresh interval.
A DRAM cell only holds its charge if something refreshes it. Micron’s quarterly revenue went from $9.3B to $41.5B in a year and the stock went up sevenfold — then fell 26% anyway, because the market cannot decide whether AI installed a refresh circuit under memory pricing or whether this bit is already decaying. Six analyst lenses, three scenarios, four time horizons.
Gray line = Micron’s actual price into today, drawn on a log scale because the range is extreme: roughly $85 in early 2024, a $65 trough in April 2025, a $114.25 52-week low in September 2025, then an eightfold run to a $1,255 intraday high on June 25, 2026 and a 26% drawdown since. Path between dated reference points is approximate; the marked points are sourced. Colored paths = synthesized scenario midpoints forward, probability-weighted at bear 33% · base 42% · bull 25% — deliberately bear-tilted because memory has never sustained a peak. The clay circle is the Wall Street 12-month consensus of $1,513 (range $361–$2,200, 48 analysts, as of Sep 1, 2026), which sits above our base case.
Re-weight the scenarios
Those probabilities are a judgment call — and on a memory cyclical they are the judgment call. Drag to set how likely the bear and bull cases are (base takes the remainder); the blended target below, the dotted line on the chart, and the prob-weighted row of every scenario card update live.
Six analyst lenses, six answers
The same quarter — $41.5B of revenue at an 84.9% gross margin — supports targets from $520 to $1,980 depending entirely on what you believe about the durability of memory pricing. Each lens below is a synthesized expert framework with its own 12-month target and the multiple behind it.
The Contracted Commodity
AI did not just raise memory demand; it changed the contract structure. Sixteen multi-year SCAs backed by roughly $22B of upfront customer commitments, some with take-or-pay terms, convert what was a spot-priced commodity into a booked order book. Management says it can fill only half to two-thirds of medium-term demand, and new wafer supply does not arrive meaningfully until fiscal 2028. Revenue went $9.3B → $23.9B → $41.5B in three quarters and the guide is $50B.
The Question Is the E, Not the P
Six times forward earnings with $23.8B of net cash, 66.6% ROE and a 9.4 Altman Z-score is not a hard valuation call — unless the E is a peak. Two flags temper the enthusiasm. Receivables ballooned from $9.3B to $31.0B in nine months, a $21.6B working-capital drag against $47.3B of nine-month net income. And FY27 capex is guided above the mid-$40Bs versus roughly $30B this year, which consumes most of the free cash flow the bull case is built on.
Peak Earnings Wearing a Growth Costume
An 84.9% gross margin on a commodity is not a business model, it is a shortage. Cost of goods sold was $6.4B on $41.5B of revenue — barely above the $5.8B it cost to produce $9.3B a year earlier. That gap is pure price, and price is what capacity kills. Micron, Samsung, SK hynix and CXMT are collectively answering with the largest memory capex wave ever, and management itself expects industry supply to improve “gradually in 2028.” CXMT went 3% → 8% of DRAM in a year; SemiAnalysis models 17% by 2028.
Third Seat at a Three-Chair Table
Only three companies on earth make HBM, capacity takes 12–18 months to add, and lithography export controls cap China’s path into it. That is a genuine oligopoly. But Micron is the third seat: SK hynix holds roughly 58% of HBM, and in Q2 CY26 DRAM revenue share Samsung led at 39% with SK hynix 26% and Micron 25%. Counterpoint expects Micron to take the number-two slot — a real share gain, on a base that is still price-driven rather than structural.
A Violent Consolidation, Not a Top
Up 684% in 52 weeks, then a 26% drawdown that has gone nowhere for six weeks. Price sits essentially on the 50-day ($946) with the 200-day 36% lower ($595) — a steep but intact uptrend. RSI 50.5 is dead neutral. The tell is positioning: short interest is only 2.66% of shares at 0.7 days to cover, so there is no squeeze fuel and no crowded bear. Beta 2.22 means the stock is a levered proxy for the AI trade in both directions.
Memory Became a Sovereign Asset
Micron is now a policy instrument, which cuts both ways. A $100B New York megafab, Idaho output from mid-2027, Singapore, India assembly, and a $1.8B Taiwan fab purchase all sit on a foundation of industrial policy and export controls that keep CXMT out of HBM. Against that: rising yields have been repricing long-duration AI names all summer, CXMT is suing the Pentagon over its Chinese-military designation, and Micron Taiwan faces a union strike threat as workers claim a share of the windfall.
Wall Street 12-month price targets
What the sell-side expects over the next year. Bars are sorted low to high; the dashed line is today’s $933.44. Note the shape of the disagreement — not a debate about degree, but a $1,839 spread between the lowest and highest target on the Street.
Selected sell-side 12-month targets from the 48 firms covering Micron; the clay dotted line is the $1,513 consensus. Rating distribution as of August 2026 was 34 strong buy, 9 buy, 4 hold, 0 sell — a near-unanimous bull book, which is itself a contrarian datapoint. The interesting names are the edges: Goldman Sachs (clay) has raised its target from $400 to $900 to $1,100 across 2026 while never leaving Neutral, arguing earnings peak in fiscal 2027; the $361 Street low implies a 61% decline. Note also the direction of travel — Citi cut from $1,400 to $1,150 on Aug 10 and Mizuho from $1,375 to $1,300 on Aug 25, the first target reductions of the cycle. Targets and ratings as reported through Sep 1, 2026.
How long does the charge hold?
Synthesized scenario midpoints, dated to each September. Returns are versus today’s $933.44. These are illustrative frameworks, not predictions — and on a memory cyclical the dispersion is the honest part of the answer. Note the bear path troughs and recovers: that is what a cycle does.
1 Year
Sep 20272 Years
Sep 20283 Years
Sep 20295 Years
Sep 2031▶ Bull case — show the assumptions & math
FY28E $210 × 12× → $2,520
FY29E $230 × 13× → $2,990
FY31E $280 × 14× → $3,920 · 5-yr price CAGR ≈ +33%/yr
▶ Base case — show the assumptions & math
FY28E $135 × 9.5× → $1,285
FY29E $120 × 12× → $1,440
FY31E $150 × 13× → $1,950 · 5-yr price CAGR ≈ +15.9%/yr
▶ Bear case — show the assumptions & math
FY28E $55 × 8× → $440
FY29E $38 trough × 11× → $420
FY31E $65 recovery × 11× → $715 · 5-yr price CAGR ≈ −5.1%/yr
▶ Why 33 / 42 / 25 rather than 25 / 50 / 25
implied 5-yr CAGR at the blend ≈ +16.9%/yr · drag the sliders above to test your own weighting
Revenue, capex, free cash flow & debt ($B)
Where the money actually goes. Four fiscal years, and the fourth bar group barely fits on the same axis as the first three — that discontinuity is the investment debate.
Three years of a normal memory company, then a fourth that looks like a different industry. FY2023 burned $6.1B of free cash flow on $15.5B of revenue; FY2026 is tracking to roughly $129.7B of revenue and $49.6B of free cash flow — a swing of $55B in three years. Two things to watch. The clay capex bar has nearly quadrupled from $7.7B to about $30B and management has guided fiscal 2027 above the mid-$40Bs, which is the bear’s supply argument and the value analyst’s cash-flow argument in the same number. Meanwhile the slate debt bar collapsed from $15.4B to $6.4B in nine months — Micron used the windfall to repay $10.4B of debt and now sits on $23.8B of net cash, which is exactly the balance sheet you want going into a cycle you cannot time. Revenue and capex FY26 estimated from nine-month actuals plus company guidance; FCF FY26 is the S&P Global consensus. Total debt is gross and includes leases; balance-sheet figures as of May 28, 2026.
The EPS ladder underneath every target ($)
The price targets are not pulled from the air — each is an EPS estimate times an exit multiple. Here is the actual quarterly earnings staircase, every rung from the company’s own filings, and the last one from its own guidance.
Eighteen times, in five quarters. GAAP diluted EPS went from $1.68 in the May 2025 quarter to a guided $30.73 in the August 2026 quarter, and every bar except the last is a filed number rather than an estimate. This chart is why the two sides of the debate can both be right: annualise the guided quarter and Micron earns about $123 a share, which puts the stock at seven and a half times — but the whole reason the multiple is seven and a half is that nobody, including the bulls, believes a memory company sustains $30 of quarterly EPS forever. Every target on this page is one of these bars extended and multiplied. FQ1’26 EPS derived from reported net income of $5,240M over diluted shares; FQ4’25 derived from FY2025 net income less nine-month reported. FQ4’26E is the company’s own GAAP guidance midpoint of $30.73 ± $1.00 issued June 24, 2026; it reports on September 30, 2026.
Every line is a triple digit
Fiscal Q3 2026 versus the year-ago quarter, sorted low to high. Olive is the steady core — phones, cars, the whole revenue line; clay is the AI frontier. Read these against a stock that is down 26% from its June high.
The disconnect in one chart: the slowest-growing line in the business grew 254%. Core Data Center revenue went from $1.53B to $11.52B, and adjusted free cash flow from $1.95B to $18.30B. The bull reads this as a business the market is mispricing at six times forward earnings. The bear reads the exact same chart as the definition of an unsustainable comparison — nothing that grows 653% is at a steady state, and the year-ago base is what a memory trough looks like. Both readings are correct, which is precisely why the stock cannot settle. Segment figures from the FQ3’26 press release; DRAM growth per company disclosure and Counterpoint Research.
Bull vs. Bear
The whole valuation argument compresses into one disagreement: has AI installed a refresh circuit under memory pricing, or is this the largest cyclical peak in semiconductor history?
▲ THE BULL CASE
- The order book is contracted, not spot. Sixteen Strategic Customer Agreements backed by roughly $22B in upfront commitments, some with take-or-pay terms — including a partnership making Micron a primary memory supplier to Anthropic. Fiscal 2026 supply is fully sold out.
- Supply physically cannot answer before 2028. New fabs take 12–18 months minimum and management expects industry supply to improve only “gradually in 2028.” Micron says it can fill just half to two-thirds of medium-term demand.
- The valuation is the thesis. 6.0× FY27 consensus EPS of $155.03, against 21× trailing. The market is refusing to capitalise earnings it can already see in the guidance.
- The balance sheet flipped. Total debt cut from $15.4B to $6.4B in nine months; $23.8B net cash, 66.6% ROE, 67.6% ROIC, Altman Z-score 9.4. A downturn would be survived from a position of strength for the first time in Micron’s history.
- Share is moving the right way. DRAM revenue share reached 25% in Q2 CY26 against SK hynix at 26%; Counterpoint expects Micron to take the number-two slot. HBM4 is in high-volume shipment for its lead customer’s platform.
- Cash conversion is real. $18.3B of adjusted free cash flow in a single quarter — more than the company generated in the four fiscal years 2022 through 2025 combined.
▼ THE BEAR CASE
- 84.9% gross margin is a shortage, not a moat. Cost of goods sold was $6.4B on $41.5B of revenue, versus $5.8B on $9.3B a year earlier. Almost the entire increase in profit is price. Price is exactly what new capacity destroys.
- The industry is already answering. Micron guides fiscal 2027 capex above the mid-$40Bs, up from roughly $30B; Samsung and SK hynix are expanding in parallel. The supply response is funded and under construction.
- China is scaling fast. CXMT went from 3% to 8% of DRAM in a year, IPO’d in Shanghai raising up to $9.8B, and SemiAnalysis models 17% share by 2028. YMTC has already overtaken Micron in global NAND shipment share.
- The receivable is enormous. Receivables went from $9.3B to $31.0B in nine months — a $21.6B working-capital outflow against $47.3B of net income. Revenue recognised is not yet cash collected, and collection quality is tested in downturns, not booms.
- History is unambiguous. Micron earned $8.7B in fiscal 2022 and lost $5.8B in fiscal 2023. Every DRAM peak in four decades has been followed by margin collapse. Goldman Sachs has raised its target three times in 2026 while never leaving Neutral, arguing earnings peak in fiscal 2027.
- The bull book is crowded and the architect left. 43 of 47 ratings are buy or strong buy with a 62% average upside. On Aug 26, 2026 Sumit Sadana — the executive who built the SCA structure the entire bull case rests on — moved to Senior Advisor.
Risk map — likelihood × impact
Where each risk sits over a three-to-five-year horizon, not just how big it is. For most companies the hot corner is empty; for Micron it is occupied by something close to a certainty — an 84.9% gross margin does not stay an 84.9% gross margin. The debate is about when and how far, which is why the rest of the grid matters so much.
- Multiple de-rating
- FY27 capex step-up
- Taiwan labour costs
- Gross-margin reversion
- Receivables / DSO
- Leadership churn
- CXMT supply flood
- AI capex digestion
- SCA renegotiation
- Taiwan rupture
- Memory architecture shift
Gross-margin reversion
What breaks: at $50B of quarterly revenue, every 10 points of gross margin is $5B of gross profit a quarter. A retreat from 86% toward the 50s halves earnings without a single lost unit.
CXMT supply flood
What breaks: China’s CXMT tripled to 8% DRAM share in a year and raised up to $9.8B in its Shanghai IPO. If SemiAnalysis is right about 17% by 2028, the three-player oligopoly the whole pricing thesis rests on becomes four.
AI capex digestion
What breaks: hyperscaler spending pauses for even two quarters. Micron’s demand is now concentrated in a handful of AI buyers, so a digestion phase hits harder and faster than a broad PC cycle ever did.
FY27 capex step-up
What breaks: capex above the mid-$40Bs against roughly $30B this year consumes most of the free cash flow, and lands new capacity into whatever demand exists in 2028 rather than the demand that justified it.
Taiwan labour costs
What breaks: Micron Taiwan unions are threatening to strike for a share of the windfall, with record bonuses already reported. Taiwan is a major part of Micron’s DRAM base — a stoppage hits output during the highest-price quarter in company history.
Taiwan rupture
What breaks: a strait crisis or blockade. Low odds in any given year, but Micron’s Taichung and newly acquired Tongluo fabs make this an existential rather than a financial risk. Genuinely uninsurable.
Memory architecture shift
What breaks: compute-in-memory, 3D DRAM or a custom-HBM base-die architecture that Micron is late to. In a business where qualification takes 12–18 months, being one generation behind on the wrong standard is close to unrecoverable.
Receivables / DSO
What breaks: receivables at $31.0B against $9.3B a year ago. If customers who committed at peak pricing slow payment in a downturn, reported earnings and collected cash diverge sharply.
Leadership churn
What breaks: Sumit Sadana, who architected the Strategic Customer Agreements, moved to Senior Advisor on Aug 26, 2026 alongside a broader reorganisation. The contract structure is the bull case; its author is now advisory.
SCA renegotiation
What breaks: the take-or-pay floor turns out to be legally soft. Customers facing collapsed spot prices renegotiate rather than litigate, and the “contracted” revenue reprices with the market it was supposed to be insulated from.
Multiple de-rating
What breaks: nothing operationally — the market simply keeps refusing to pay more than single digits for peak memory earnings. Low impact only because it has arguably already happened at 6.0× forward.
The jargon, decoded
Hover the dotted terms in the metrics and the panel, or scan the desk’s working definitions here.
- DRAM
- Dynamic random-access memory — the working memory in every server, phone and PC. Each bit is a capacitor that loses its charge unless it is periodically refreshed, which is where this report gets its title.
- HBM
- High Bandwidth Memory. DRAM dies stacked vertically and bonded next to an AI accelerator. Only three companies make it, qualification takes 12–18 months, and it is the highest-margin product Micron sells.
- Gross margin
- Revenue minus the cost of making the product. For a memory maker it is the cycle indicator: 37.7% in the May 2025 quarter, 84.6% a year later, guided to about 86%. It is also the first thing to fall.
- Take-or-pay
- A contract term obliging a customer to pay for agreed volume whether or not they take delivery. It is what turns a spot commodity into contracted revenue — if it holds when the customer is under pressure.
- Strategic Customer Agreement (SCA)
- Micron’s name for its multi-year supply contracts. Sixteen were in place as of July 2026, backed by roughly $22B of upfront customer commitments.
- Forward P/E
- Price divided by next fiscal year’s expected earnings. Micron’s 6.0× on FY27 consensus of $155.03 is not a bargain signal so much as a statement that the market disbelieves the estimate.
- Free cash flow
- Cash from operations minus what is spent building fabs. Micron generated $18.3B in a single quarter — and plans to spend above the mid-$40Bs on capex next fiscal year.
- EV/EBITDA
- Enterprise value (market cap plus debt minus cash) over earnings before interest, tax, depreciation and amortisation. At 15.1× Micron looks expensive on this measure precisely because depreciation on $57B of plant is enormous.
- DSO / receivables
- Days sales outstanding — how long customers take to pay. Micron’s receivables tripled to $31.0B in nine months, so a growing share of reported profit has not yet arrived as cash.
- Exit multiple
- The P/E assumed at the end of the forecast. Multiply it by projected EPS to get a target price. On a cyclical, choosing the exit multiple matters more than forecasting the earnings.
- Prob-weighted
- Each scenario’s price multiplied by its probability, summed into one expected value across bear, base and bull. The clay row in every scenario card, and the number the sliders move.
- Bit shipments vs. ASP
- The two halves of memory revenue: how many bits you sold and what each bit fetched. In FQ3’26 bit shipments rose a low single-digit percentage while prices rose in the low 60s sequentially — this cycle is almost entirely price.