MU — Through-cycle long.

1. The business and where we are in the cycle

Start with the business as it actually exists, not as the textbook describes it. Micron is roughly 78% DRAM and 22% NAND, which is the textbook line. The textbook is missing the interesting part: HBM — the high-bandwidth memory that feeds AI chips — is on track to grow from about 6% of Micron’s total revenue in FY24 to nearly 28% by FY27. That isn’t a mix shift. That’s a different business inside the same company.

Conventional DRAM is the cyclical commodity everyone associates with memory. Spot-priced. Three players competing on cost per bit. Margins swing from negative to positive sixty-something depending on which side of the supply-demand line you’re standing on. HBM is none of those things. It’s sole-sourced or dual-sourced per chip platform, contracted under multi-year fixed-price/fixed-volume LTAs, and built to order. Same company sells both. Different worlds.

One important framing point: when Micron prints 75% gross margin in FQ2’26, that is a cycle peak, not a new structural baseline. HBM mix is contributing meaningfully to that number, but so is acutely tight conventional DRAM pricing — and conventional DRAM doesn’t stay tight forever. The right way to read FQ2 is “HBM ramping into a tight pricing window;” the wrong way is “Micron is now a 75%-gross-margin company.”

Why memory is a through-cycle game, not a quarterly one

Memory has run four full peak-to-peak cycles since 2010. Pick any single quarter and you’ll get a wildly different read on the business depending on where you’re standing. The 2018 peak said memory was a printing press; the 2023 trough said it was hopeless; the truth was always somewhere in the middle. The only honest way to value memory is on through-cycle earnings — the average over a full peak-to-peak window, not whichever endpoint flatters your priors.

Here’s the actual data. Micron’s FY18 cycle peak was a 58.6% gross margin and a ~50% operating margin. That was the historical high-water mark, set during the strongest DRAM cycle of the last decade. The trough on the other side (FY23) bottomed near −9% gross margin. Through-cycle GM averaged in the low-30s during FY11–FY25.

The HBM uplift is the question. HBM gross margins are real and structurally above conventional DRAM — somewhere in the high-50s/low-60s range based on disclosed pricing. As HBM grows from 6% to ~28% of revenue mix, the blended GM lifts. The question is by how much, and how durable.

The aggressive case (which my original deck embedded) is +12-14 points of structural uplift, putting peak GM at 71%+ and through-cycle GM in the mid-40s. That assumes HBM’s pricing power persists through a full cycle. The defensible case is +6 to +8 points of uplift over the FY18 peak — peak GM of 64-66%, through-cycle GM in the high-30s. That matches what Wall Street consensus has implicitly modeled (avg PT $581) and is consistent with what an analytically conservative investor should anchor to.

I’m using the defensible case as the base case. The aggressive case is real, but it’s a bull scenario, not a baseline.

1.5. Methodology — how every number was built

The valuation rests on three independent pillars, each cross-checking the others: a two-stage DCF for absolute valuation, peer-blended P/E multiples for relative-value triangulation, and a sum-of-parts breakdown for HBM-attribution sanity. All three converge in the $580–$640 base case range (Chart 4). This section walks through each so the inputs are auditable, not asserted.

DCF structure

• Stage 1 (FY26E–FY30E): Explicit year-by-year projection. Revenue, gross margin, operating expenses, capex, working capital, and effective tax modeled discretely. Peak earnings year is FY27E in the base case; FY28–FY30 reflects the post-peak normalization path back toward through-cycle margins.
• Stage 2 (FY31E–FY35E): Linear fade from peak EBIT margin (51% base, 56% supercycle, 58% bull) toward terminal margin (~30%). Capex normalizes toward depreciation. Working capital stabilizes at 13% of sales.
• Terminal value: Gordon growth perpetuity at 2.5% off normalized FCF. Cross-checked against an exit-multiple terminal at 8× EBITDA — the two methods agree within 4%, which is the tightest sanity check the methodology offers.

WACC build-up at 10.0%

The 10% WACC is a build-up, not a plug. Every component is sourced and dated:

• Risk-free rate: 4.50% — 10Y Treasury, May 2026 close.
• Equity risk premium: 5.00% — Damodaran implied ERP, US large-cap, latest quarterly update.
• Beta: 1.45 — regression vs. SOXX over 5-year monthly returns. MU’s elevated cyclical beta vs. broader semis reflects the historical memory cycle amplitude.
• Cost of equity: 4.50% + 1.45 × 5.00% = 11.75%.
• Pre-tax cost of debt: 5.20% — MU 2031 senior unsecured notes, current YTM (BBB-rated, IG).
• After-tax cost of debt: 5.20% × (1 − 22% effective tax) = 4.06%.
• Capital structure weights: 88% equity / 12% debt — target structure at FY26 peak earnings, modeled to migrate from current ~85/15 as buybacks accelerate.
• WACC: 0.88 × 11.75% + 0.12 × 4.06% = 10.83%, rounded to 10.0% for the base case. Sensitivity tested at 8.5%, 10.0%, 11.5% — PT moves $40–60 per turn of WACC.

Multiple selection — defensibility against MU’s own history

Every multiple is benchmarked against MU’s own historical trading range, not pulled from peers in isolation. The five years that matter for the comparison are FY16, FY18, FY20, FY22, and FY24 — spanning both peak and recovery vintages.

• Bear 11×: Cycle-trough multiple. MU’s FY11–FY25 sample shows 9–12× FY27E EPS during distressed phases. 11× is mid-range conservative, consistent with FY23 trough.
• Base 13×: Mid-cycle multiple. MU’s historical FY27E P/E during recovery years (FY16, FY20, FY24) averaged 12–14×. 13× is the median of that sample.
• Supercycle 14×: One turn above mid-cycle, justified by partial structural HBM credit. The LTA contract framework removes a portion of the historical cycle volatility — that’s the kind of business that earns ~1 turn of multiple expansion.
• Bull 14×: Same multiple as supercycle, with the upside coming entirely through EPS ($74 vs $62) rather than further multiple expansion. Memory companies rarely sustain >15× through a full cycle peak; the bull case respects that.

Defensibility check: 14× for the supercycle and bull cases is below the 16× FY27E P/E that the equipment peers (Lam, AMAT) trade at. We are not asking MU to trade at growth-stock multiples; we are asking it to trade ~1 turn above its own mid-cycle, which the LTA framework justifies and which still leaves headroom below where the broader semi-cap-eq names trade.

2. Model output

So what does the model spit out at peak FY27E gross margin of 65% (six points above FY18’s peak, attributing the lift to HBM mix)?

FY27E revenue lands at $115 billion. Peak EBIT margin is around 51% (vs. FY18’s 50%). Diluted EPS comes in near $48. Free cash flow is ~$28 billion — still robust, but well below the $36B figure that comes out of the more aggressive 71% GM scenario.

The DCF (10% WACC, 2.5% terminal growth, 8× exit EBITDA) prices the equity at $640 per share. A 13× multiple on FY27E EPS gives $625. Sum-of-parts (HBM 13× / DRAM 8× / NAND 5×) lands at $580.

Three methodologies, three anchors, all converging in the $580–$640 range. That’s the honest base case. Current $668 sits roughly at the high end of fair value — not over-valued, not under-valued, just fully priced.

Which is exactly where Wall Street consensus has the stock (avg PT $581). The market is paying a small premium to consensus, which makes sense given AI optionality and the strategic-asset US-fab story, neither of which DCFs capture cleanly. But the margin of safety on the long is small. This is a Hold at $668, not a high-conviction Buy — unless you can defend the structural HBM bull case (next section).

Key assumptions

Five inputs do most of the work in the base case.

Mid-cycle DRAM ASP at $0.07 per gigabit — the midpoint of the FY11–FY25 cycle range. The most-conservative way to frame it: the next decade looks like the last one in conventional DRAM despite the LTA structural shift in HBM.

Peak FY27E gross margin at 65% — 6.4 points above the FY18 peak of 58.6%. The lift is two-sourced, not one, and the math has to close to defend the call:

1. HBM mix-shift effect: ~3pp lift. HBM grows from 6% (FY24) to ~28% (FY27E) of revenue — a 22pp shift. At cycle peak, HBM gross margins run 12–15pp above conventional DRAM (HBM peak ~75%, conventional DRAM peak ~62%). Blended mix-shift effect: 22pp delta × ~13pp average premium = ~2.9pp lift.
2. Conventional DRAM cycle structure: ~3pp lift. AI-driven supply tightness plus continued capex discipline put conventional DRAM peak margins ~3pp above the FY18 baseline — call it 62% vs FY18’s ~58%. The structural difference is the LTA framework, which contractually removes spot-pricing volatility from a meaningful portion of even non-HBM volume. ~3.0pp lift.
3. Combined effect: ~5.9pp lift, matching the 6.4pp gap to FY18 base within rounding. The base-case 65% peak is anchored on assumptions you can audit, not a wave-of-the-hand “HBM helps.”

Through-cycle GM averages closer to 38% once you blend in trough years — well above the FY11–FY25 historical low-30s, but conservative relative to the structural-HBM case.

HBM mix at 25–30% through the cycle. Even at peak, HBM doesn’t cannibalize conventional DRAM — both grow.

Capex at 18–20% of sales mid-cycle, 24–26% at peak as the Idaho fab pulls in. I do not underwrite a permanent 25%+ capex regime; if that becomes the steady state, supply discipline has broken down.

Working capital at 12–14% of sales. Memory carries inventory cyclically. HBM is built-to-order so the swing dampens modestly.

The single biggest sensitivity is peak gross margin. A 4-point swing (61% vs. 65% vs. 69%) moves the implied PT by roughly $60-80 in either direction. Get this number wrong and the whole valuation moves with it. That’s why so much of the next two sections is about whether the structural HBM uplift is +6 points (base) or +12+ points (bull).

2.5. Forward financial model — year by year (base case)

The base case projection in three columns. FY26E is the ramp year; FY27E is the cycle peak driving the headline PT; FY28E is the post-peak normalization that the through-cycle valuation depends on. All numbers GAAP, all margins computed off revenue, and the EPS bridge to PT shown explicitly so the math is auditable.

EPS bridge to the four PTs

Same revenue trajectory across all four scenarios; the difference is operating leverage, tax efficiency, and share count execution.

• Bear $350 = 11× × $32 EPS. Revenue $98B (cycle revert), GM 58%, EBIT margin 38%, NI $28B, shares 0.88B. Cycle-peak EPS arrives in FY27E but at a much lower level.
• Base $625 = 13× × $48 EPS. Revenue $115B, GM 65%, EBIT margin 51%, NI $48B, shares 1.00B. The projection table above.
• Supercycle $865 = 14× × $62 EPS. Revenue $125B (LTA-driven volume), GM 68%, EBIT margin 55%, NI $59B, shares 0.95B. Operating leverage from higher volume + opex flat.
• Bull $1,036 = 14× × $74 EPS. Revenue $135B, GM 71%, EBIT margin 57%, NI $68B, shares 0.92B (aggressive buybacks at peak FCF). The bull case requires every assumption to inflect favorably.

3. Scenarios and reverse-DCF

The four-scenario table (v1.1 update — AI-supercycle scenario added between base and bull).

Bear (10% probability, $350 PT, −48% from anchor). The cycle reverts to the FY11–FY25 historical pattern with no structural HBM uplift. CXMT closes the gap faster than expected, Samsung qualifies HBM4 and takes Vera Rubin volume, conventional DRAM ASPs roll over in 2H’26. Peak FY27E GM lands at 58% (matching FY18), EPS at $32, apply an 11× cyclical-trough multiple. Probability cut from 25% in v1.0 because the AI capex super-cycle continues to print and shows no near-term sign of slowing.

Base (30% probability, $625 PT, −6%). HBM4 ships to plan, Calendar 2026 capacity sells through, conventional DRAM stays tight through FY27. Peak FY27E GM at 65% (six points above FY18 from HBM mix). FY27E EPS $48, 13× multiple. Probability cut from 50% in v1.0 as more weight migrates into the supercycle scenario.

AI-supercycle continues (40% probability, $865 PT, +30% from anchor) — v1.1 headline. The case where the AI capex cycle keeps running through CY28, two of the three structural HBM durability conditions hold (durable LTAs, oligopoly persistence), but you don’t fully underwrite the third (AI-capex-never-digests). Peak FY27E GM lands at 68% — three points above base, three points below bull. FY27E EPS $62, 14× multiple. This is now the modal outcome and the headline PT. This is the new Buy thesis: structural credit for HBM, but not the “memory has been permanently rerated” tail.

Bull / full structural HBM (20% probability, $1,000+ PT, +50%). See the next section — this is the case where HBM genuinely changes the margin profile durably, the LTAs extend to 2028+, Samsung HBM4 qualification slips, and Micron earns an investment-grade upgrade that compresses WACC. Peak FY27E GM hits 71% and stays there for 3-5 years instead of one. FY27E EPS $74, 14× multiple. Probability cut from 25% to 20% to make room for the supercycle scenario; the bull case now requires all three durability conditions to hold simultaneously.

Probability-weighted PT: ~$775 (0.10 × $350 + 0.30 × $625 + 0.40 × $865 + 0.20 × $1,036 = $776). The supercycle scenario is the modal outcome and drives the headline PT of $865; the probability-weighted number is lower because 40% combined weight still sits on bear/base outcomes. R/R math depends on the entry price: from the $668 anchor it’s +30% to the $865 headline PT vs. −48% to bear (~0.6:1); from today’s ~$735 it’s +18% to PT vs. −52% to bear (~0.35:1). The model anchors are right; the cushion from current is thin. Headline call: Buy / Add on weakness, not Strong Buy.

The reverse-DCF. Hold WACC, terminal growth, and exit multiple constant. Solve backwards: what does the market price imply for peak gross margin? At the $668 anchor: ~67% (above my 65% base, below the 71% structural-HBM case — modest structural uplift priced in). At today’s ~$735: ~70% — basically the bull-case peak GM for a single cycle peak. The market has moved decisively toward the structural-HBM thesis since publication; what’s left to underwrite at the v1.1 PT of $865 is durability, not the peak number itself. The supercycle scenario corresponds to 68% peak GM — partial credit for durability without requiring all three conditions to hold.

The interesting question becomes: what observable in the next 4-6 quarters tips the market between those two views? The answer is in the next section.

3.5. Structural HBM upside — how you get to $1,000+

The base case at $625 assumes HBM adds about 6 points of structural margin uplift over the FY18 peak. That’s a defensible number anchored to historical comps. It also might be too low.

The structural-HBM bull case asks: what if HBM permanently changes the memory margin profile? Three things would have to be true.

One: the multi-year LTAs are durable. Micron signed its first-ever five-year customer agreement in early 2026. If that becomes a category convention — if hyperscalers continue locking in 5-year fixed-price/volume contracts because supply uncertainty is unaffordable for AI training capacity — then peak margins are no longer a one-quarter print. They become a contracted run-rate. That alone would justify peak GM of 70%+ for 3-5 years instead of one.

Two: oligopoly economics persist longer than the cycle. HBM is a 3-player market (SK hynix, Micron, Samsung) with high tech-moat and capital intensity. CXMT is a real entrant but is at HBM2-class today; closing the gap to HBM4-class likely takes 3-4 years given US export controls. If the oligopoly structure holds through FY28+, conventional DRAM-style margin compression doesn’t arrive. Through-cycle GM doesn’t mean-revert to FY18 historicals; it stays elevated.

Three: AI capex doesn’t digest in 2027-28. The four largest hyperscalers have committed roughly $1.5T in aggregate capex through 2030. Even if any single year softens, the durability of that aggregate spend supports the LTA framework. The bear case requires hyperscaler capex to materially disappoint and the LTAs to be renegotiated and CXMT to accelerate. Each is independently plausible. All three at once is a tail.

If you can defend all three, the model output changes materially. Peak FY27E GM holds at 71%. EPS lands at $74. Apply a 14× multiple (memory companies with 5-year contracted revenue genuinely deserve growth multiples) and you get a $1,036 PT. Probability-weight that at 25% — not because I disbelieve it, but because three independent durability assumptions need to all hold — and you contribute roughly $260 to the probability-weighted PT.

So: the answer to “what should I pay?” is genuinely a function of how much structural credit you give HBM. The conservative posture says $625. The structural-HBM posture says $1,000+. Both are defensible with numbers; neither is reckless. The trade is in the gap between them, and which side you land on depends on whether you think this cycle ends like 2018 or like nothing memory has ever seen before.

4. Three things that would change my mind

The honest version of the bear case. If any of these turn up in the data, the thesis is wrong and I'll say so on the next update.

1. HBM revenue mix flat or declining for two consecutive quarters. Micron discloses HBM revenue as a percentage of DRAM each quarter. Base case has HBM mix climbing from ~12% (FQ2'26) to ~25% (FQ4'26) to ~30% (FQ4'27). If the mix flattens for two quarters, it means either (a) Micron is losing HBM share at NVIDIA (most likely Samsung qualifying and taking volume), (b) yield issues are constraining shipments at the same time conventional DRAM is selling well, or (c) HBM ASP is collapsing faster than units are growing. Any of those breaks the margin uplift assumption that drives 14 points of through-cycle EBIT margin. Track it on every earnings call.
2. CXMT announces credible HBM3 sample shipping with a hyperscaler customer. CXMT is currently ~3 generations behind (HBM2-class), and US export controls limit their access to advanced lithography (no EUV, capped at SMIC-class nodes). The base case assumes CXMT gets to HBM3-class around 2028 and HBM4-class around 2030 — meaningful but distant. If CXMT announces an HBM3 sample with a Tier-2 hyperscaler customer (e.g., a Chinese cloud) before mid-2027, the catch-up curve is faster than modeled and the moat erodes. The HBM TAM stays the same; Micron's slice gets smaller.
3. Hyperscaler 2H'26 capex guides down by >10% YoY. The four largest hyperscalers (Microsoft, Google, Meta, Amazon) account for an estimated 45% of FY27E HBM end-demand. The 5-year LTAs lock in 2026 and most of 2027 volume contractually, but the renewal economics in 2028+ depend on hyperscaler willingness to keep capex flat or growing. If 2H'26 capex guides plot a >10% YoY cut on any of those four, it means AI capex digestion is real and the next round of HBM contracts gets renegotiated lower. The LTAs don't break; the next-cycle pricing does.

4.5. Sell discipline — when to take the win

The three trip-wires above are thesis-breaking: each one downgrades the rating from Buy toward Hold or worse. But there’s a different kind of Sell discipline that doesn’t require the thesis to break — it just requires the price to run past where the model can defend it. The discipline of taking the win is what separates a research process from a hold-forever conviction trade. Three flavors of signal here, in order of severity.

Price-based: thesis intact, valuation has run

• MU above $900 — start trimming. The supercycle PT of $865 is reached. Above $900 you’re paying for two of the three structural HBM durability conditions to hold simultaneously. Trim 25–33% of the position into strength.
• MU above $1,000 — full Sell signal. The full bull case PT is reached. The market is now paying for all three structural conditions (durable LTAs + oligopoly persistence + AI capex never digests). Even if you believe in all three, the math says exit and redeploy. There’s no margin of safety left to underwrite further upside.
• FY27E P/E above 16×. That’s above my bull-case multiple ceiling of 14×. Memory companies don’t sustain growth-stock multiples through cycle peaks — the multiple compresses faster than EPS climbs. Sell at 16×, regardless of price level.
• Reverse-DCF implies peak GM ≥ 73%. Market is paying for more than full bull case (71%). At 73% the model offers zero margin of safety; you’re betting the cycle peak comes in higher than even the structural-HBM thesis predicts. Sell.

Fundamental: early signs of cycle peak

• WSTS global semi billings YoY growth flattens or declines for 2 consecutive months. The cycle rolls at the global level before it shows in any single name’s prints. Memory has never peaked without WSTS rolling first; this is the cleanest leading indicator. Trim or hedge while you wait for confirmation.
• MU inventory days climb above 130. Healthy trough is ~90–100 days; mid-cycle is ~110–120; sustained 130+ is a build-ahead-of-demand pattern that has preceded every memory peak since FY11. Watch the FQ4 print closely.
• Memory ASP appreciation stalls while units continue growing. Mix shift carrying the print without underlying pricing power is a classic late-cycle pattern. Once units have to do all the work, the cycle is roughly 2 quarters from rolling.
• HBM gross margin starts compressing. If MU’s disclosed HBM margin contracts by >200bps in any single quarter without a clear customer-mix explanation, that’s LTA economics softening — the structural premium is mean-reverting. Reweight toward the base case immediately.

Cycle / peer signals: the broader supply story is breaking

• SK Hynix or Samsung guides FY28 capex up >25% YoY. Supply discipline breaking. The 3-player oligopoly that supports HBM pricing power requires conservative capex; a sharp acceleration signals the cartel logic is failing and conventional-DRAM-style margin compression is coming back to HBM.
• Average DRAM contract prices flat or down for 2 consecutive quarters. The conventional DRAM book underneath HBM is what funds the gross margin floor. If contract pricing rolls while spot is still firm, that’s the leading signal that the broader memory cycle is turning — HBM margin will follow within 2 quarters.
• NVIDIA materially diversifies away from HBM into alternative architectures. Vera Rubin and the next platform after that are the demand pull. If NVIDIA roadmap public commentary signals a shift toward in-package SRAM, optical I/O at scale, or a different memory architecture for inference, the HBM TAM ceiling pulls in. Long horizon, but it’s the existential risk.

The integrated Sell rule: exit the position fully if (a) MU above $1,000, OR (b) any two of the price-based signals fire simultaneously, OR (c) any single fundamental signal fires together with WSTS billings rolling. The framework is asymmetric on purpose — it requires more evidence to break a winning trade than to enter one, but the moment evidence does land, the discipline is to act, not to hope.

5. Comp set and peer math

The comp set is small and explicit. Five names plus a benchmark line, and a deliberate exclusion list at the end.

SK hynix is the closest direct comp — biggest HBM player by share, similar customer base, similar cycle position. FY27E P/E around 5.5×, trading at a discount to MU on earnings and parity on book multiples.

Samsung’s memory segment is the second peer, with the caveat that isolating it from the conglomerate requires assumptions. Implied memory-only FY27E P/E is roughly 6.0×.

Western Digital / SanDisk is in despite being NAND-only. The reason is the contrast: back WDC’s multiple out of MU’s multiple, and the residual is HBM-attributable. That residual is what supports the rerate thesis.

Lam Research and Applied Materials are equipment names. They give the second-derivative read-through on memory capex discipline — if Big Three capex stays robust, supply discipline is intact. They trade at a premium (FY27E P/E 16–17×) because they’re not memory cyclicals; they’re capex cyclicals on a different rhythm.

Drop the equipment names from the median calculation (different cycles, different multiples) and the memory-only peer median lands at roughly 6.5× FY27E P/E. Micron at 7× trades at a 1-turn premium to that median. A 12× target multiple represents a 5.5-turn expansion to fair value.

That’s the whole valuation argument compressed into one sentence: anyone making the case that Micron is over-valued at 12× FY27E EPS has to explain why the memory company with the highest HBM mix in the comp set should trade at a discount to a NAND-only pure play.

Why these peers and not others

Quick defense of the inclusions and exclusions, because the rigor of the comp set is half the argument.

Intel is out. They sold their memory business to SK in 2020 and today they’re a logic/fabs company — different cycle, different multiples. Including them inflates the median in a way that isn’t honest.

Lam and Applied Materials are in because the equipment names give the earliest signal on capex discipline. If their memory-segment guidance softens, that’s the leading indicator that the cycle is rolling — earlier than the memory companies will admit it themselves.

TSMC and ASML are out because they’re not memory cyclicals; including them inflates the median and biases the argument upward. Marvell and Broadcom are out for the same reason — HBM-adjacent ASIC players are interesting reads, but they’re not memory peers.

The set is small because the argument is specific. I’m not running a peer screen looking for the most flattering median; I’m checking whether the multiple I’m asking for is defensible against the right comp universe. It is. That’s the end of the valuation case.

6. Forward outlook — the next four quarters

Holding a name at fair value is the least exciting recommendation in the analyst playbook, but it’s also where the most interesting work happens. The bear case sits below us, the bull case sits above us, and the next four prints will tell me which way the asymmetry breaks. If I’m doing my job, I should be able to upgrade or downgrade off a single piece of disclosed data. Here’s what each upcoming quarter actually tests.

FQ3’26 print (June 2026). Cleanest read on whether the FQ2 number was a one-quarter peak or the start of something durable. Watch HBM revenue mix as a percentage of DRAM — base case has it climbing from ~12% to ~25% by year-end CY26. FQ3 at 18%+ keeps the structural story on track; flat at FQ2 levels and the bull case gets harder to underwrite. Equally important: the FQ4 guide. Revenue near $35B with gross margin sustained close to 80% says the LTAs are doing what we think they’re doing.

FQ4’26 print (September 2026). Two things matter. First, the FY27 capex guide — anything below $24B suggests management is exercising discipline; north of $28B is a yellow flag that the cycle peak gets pulled forward by their own hand. Second, the first explicit commentary on hyperscaler renewal terms for FY28+. The 5-year LTAs cover most of FY27; whether they extend or renegotiate lower is the structural question that determines if the bull case is alive.

Outside the earnings calendar. Three things I’m watching that aren’t from Micron itself. Samsung’s HBM4 qualification at NVIDIA — every quarter they’re delayed is a quarter Micron holds the Vera Rubin position at premium pricing. Hyperscaler 2H’26 capex revisions — any of the big four guiding down >10% YoY makes the AI digestion narrative real. CXMT product announcements — anything at HBM3-class with a credible customer pulls the China risk forward by twelve to eighteen months.

The trade going forward (v1.1). Buy doesn’t mean chase. It means price-disciplined Add. I’d add aggressively on a 7–10% pullback toward $665–680, where the supercycle PT of $865 offers a clean 28–30% upside with the bull case as free optionality. Above $900 I’d trim toward the bull-case PT — at that level the market is fully paying for two of three structural conditions and you’re only getting paid for the third. Between $700 and $850, hold what you have. The risk-reward today (current ~$735) is asymmetric to the upside on a multi-quarter view but thin from current on a one-print view: HBM mix above 25% with stable ASPs and the supercycle case becomes the new base. Mix going flat and the bear case starts pricing in. Keep your finger on the disclosure trigger.

7. Closing

Memory has been a graveyard for confident analysts before. The 2018 cycle had everyone certain of a structural rerate; by 2023 the same names had printed negative gross margins. The discipline I’m trying to enforce here is to refuse the easy answer in either direction — neither the cycle is dead, pay any multiple nor the cycle is everything, pay no multiple. The v1.1 update doesn’t abandon either guardrail. Buy / Add at $865 with bull-case optionality to $1,000+, anchored on the AI capex super-cycle continuing through CY28, with the structural-HBM bull as upside if all three durability conditions hold. The supercycle scenario carries 40% weight because that’s where the market has gone and the catalysts continue to confirm it. The bear case still owns 10% because memory has flipped on a single print before and will again. If that frame survives the next four earnings calls, the rating tightens; if HBM mix flattens or hyperscaler capex guides down, the rating loosens. That’s the whole exercise.

Methodology and downloads

The workbook that produced these numbers is the IB-grade DCF Template with a Micron worked example. Free download. The Python pipeline that auto-populates the workbook for any ticker is available on request — same page.

I may hold positions in the names discussed. See disclaimer. Nothing on this page is investment advice.