Micron Technology, Inc. (NASDAQ: MU) — Initiating Coverage NEUTRALScenario Fair Value$820

1 · Summary & Verdict

Exceptional HBM-driven execution at a cyclical peak — a disciplined mid-cycle valuation says the supercycle is already in the price; we initiate with a Neutral view and a Scenario Fair Value of $820.

THE THREE RISKS THAT MATTER

2 · Investment Thesis

1. The AI memory supercycle is real, and Micron is a prime beneficiary

Generative-AI infrastructure has structurally re-rated memory demand. AI servers carry multiples of the memory content of traditional servers, and HBM — the stacked-DRAM companion to AI accelerators — has gone from niche to one of the fastest-growing categories in semiconductors. FQ2-26 revenue of $23.9B (+196% YoY) versus $37.4B for all of FY2025; FQ3 guided to $33.5B at ~81% gross margin. HBM is sold out through 2026 under multi-year hyperscaler agreements, and volume HBM4 production for NVIDIA's Vera Rubin platform has begun.

• HBM revenue scaling from ~$4B (FY25) toward ~$28B (FY27E); HBM consumes 2-3x the wafer capacity of standard DRAM per bit, tightening the entire market
• We model FY2026E revenue of ~$105.5B and EPS of ~$53

2. A consolidated oligopoly with genuine technology leadership

DRAM is a three-player market (Samsung, SK Hynix, Micron). This structure has meaningfully dampened — though not eliminated — the historic boom-bust amplitude. Micron has closed its historical technology gap and now holds ~21% of HBM (overtaking Samsung's ~17%, behind SK Hynix's ~62%), with a credible HBM4 design win. As the only U.S.-domiciled producer, Micron enjoys CHIPS Act support (~$6.4B) and is the natural supplier for customers diversifying away from Korean and Chinese sources.

• Rising memory content per device (on-device AI, automotive) provides a secular tailwind beneath the cyclical swings
• ~$200B U.S. manufacturing program positions Micron as the cornerstone of a domestic memory supply chain

3. Valuation already discounts a permanent peak

A disciplined DCF, anchored on mid-cycle free cash flow at an 11.4% WACC, produces ~$191/share — and even the most generous sensitivity (9.5% WACC, 4% terminal growth) reaches only ~$266. Memory peers trade at 6.8–11x forward earnings; MU at ~17x our FY27E EPS sits at a clear premium. Tellingly, the stock trades above the Street's own average price target of ~$674.

• To justify $942 one must capitalize ~$115B+ of unlevered FCF in perpetuity, comfortably above the FY26-27 peak of ~$32B
• Peers' low multiples reflect the market's own expectation that peak memory earnings fade

4. Negatively-skewed risk/reward at current levels

Our scenario framework yields a wide, asymmetric distribution: bull $1,250 (35%), base $700 (45%), bear $350 (20%) — probability-weighted ~$820. Upside to bull is ~+33%; downside to base ~-26% and to bear ~-63%. With the most-likely outcome below the current price, there is no margin of safety at $942.

• Upgrade trigger: ~$700 (cyclical risk priced, bull case becomes a multi-bagger)
• Trim trigger: ~$1,050

3 · Financial Analysis

Historical Performance — The Anatomy of a Memory Cycle

Micron’s financial history is a textbook illustration of memory cyclicality, and understanding it is prerequisite to valuing the equity. Over FY2022–FY2025 the company traversed an entire cycle: from a prior peak of $30.8B revenue in FY2022, into a brutal FY2023 trough of $15.5B (-50%), and back to a record $37.4B in FY2025 (+49% YoY). The trough was not merely a slowdown — it was an earnings collapse. FY2023 gross profit was NEGATIVE at -$1.4B (a -9.1% gross margin, meaning Micron sold memory below cash production cost), operating loss reached -$5.7B, and the company posted a -$5.8B net loss (-$5.34 diluted EPS). This is the single most important data point in the entire analysis: it is what the bottom of a memory cycle actually looks like, and it is why peak earnings cannot be capitalized.

The FY2024 recovery was rapid, with revenue rebounding 62% to $25.1B and the company returning to a modest $778M net profit as DRAM pricing inflected. FY2025 then delivered the AI-driven breakout: revenue of $37.4B at a 39.8% gross margin, $9.8B operating income and $8.5B net income ($7.59 diluted EPS). DRAM revenue grew 62% (driven by a low-40% increase in average selling prices and mid-teens bit growth) while NAND grew 18%. The trajectory accelerated violently into FY2026: first-half revenue of $37.5B already matched all of FY2025, with FQ2-26 gross margin of 74.4% and operating margin of 67.6% — figures with no precedent in the company’s history.

Free cash flow tells the subtler story. Despite record operating cash flow ($17.5B in FY2025, $20.3B in first-half FY2026), free cash flow has been structurally constrained by Micron’s relentless capital intensity: FY2025 FCF was just $1.7B as capex reached $15.9B. As the ~$200B U.S. expansion program ramps (FY2026E capex ~$30B, stepping up further in FY2027E), the gap between earnings and free cash flow widens — a critical consideration for any cash-flow-based valuation.

4 · Projection Assumptions

Our model carries three years of reported actuals (FY2023A–FY2025A) and five years of estimates (FY2026E–FY2030E). The central modeling judgment — and the one that drives valuation — is that FY2026–27 represents a cyclical peak that normalizes thereafter. We detail every major assumption below, by technology, by business unit, by geography, and by line item. All figures tie to the integrated three-statement model, which balances in every year.

Revenue Assumptions by Technology

DRAM (incl. HBM) — the primary value driver

We project DRAM revenue of $84.0B in FY2026E, rising to $94.7B in FY2027E before normalizing to $75.7B (FY28E), $62.7B (FY29E) and recovering to $76.3B (FY30E). The build reflects:

■ FY2026E (+203% YoY to ~$84B): H1-FY26 DRAM strength is already reported; we assume DRAM mix rises to ~80% of revenue as HBM consumes leading-edge wafers. ASPs are the dominant driver, with HBM commanding multiples of commodity DRAM pricing.

■ HBM specifically: we estimate HBM revenue of ~$18B in FY26E rising to ~$28B in FY27E, consistent with management’s ‘sold out through 2026’ commentary and the HBM4 ramp for NVIDIA Vera Rubin. HBM is the single largest swing factor in the model.

■ FY2027E (+13%): continued HBM4 ramp and tight industry supply support another up-year, with DRAM mix peaking near 82%. This is the last year of the assumed up-cycle.

■ FY2028E–FY2029E (-20%, -17%): the core cyclical assumption. As the three suppliers (plus Chinese entrants) add capacity into the demand, we model ASP normalization and a classic down-cycle. Bit shipments keep growing but pricing falls faster.

■ FY2030E (+22% recovery): supply discipline and continued secular AI bit-growth drive a cyclical recovery, though to a level well below the FY27 peak.

Key risk: the depth and timing of the FY28-29 normalization is the model’s most uncertain element. The bull case assumes it never materializes; the bear case assumes it is sharper and earlier.

NAND flash — the secondary line

We project NAND revenue of ~$21.0B in FY2026E, broadly flat-to-down through the forecast ($20.8B FY27E, $18.9B FY28E, $17.7B FY29E, $20.3B FY30E). NAND is structurally less attractive than DRAM — more players, weaker pricing power, harsher cyclicality — but data-center SSD demand tied to AI data pipelines has improved the outlook. We assume NAND mix declines from ~26% historically toward ~18-20% as HBM-led DRAM dominates the revenue base. NAND ASP recovery is assumed more muted than DRAM given the more fragmented competitive structure (Samsung, SK Hynix/Solidigm, Kioxia, Sandisk, plus YMTC).

Revenue Assumptions by Business Unit

■ Cloud Memory (CMBU): the highest-value unit, carrying HBM and hyperscale DRAM. We model it as the largest and fastest-growing BU, ~33-35% of revenue, with the highest gross and operating margins (FQ2-26 CMBU operating margin was ~66%).

■ Core Data Center (CDBU): OEM server DRAM and data-center SSDs; ~18-21% of revenue, growing with the broader server build-out beyond hyperscalers.

■ Mobile & Client (MCBU): smartphone and PC memory/storage; ~32-40% of revenue, benefiting from rising per-device memory content as on-device AI proliferates, but more commoditized.

■ Auto & Embedded (AEBU): automotive, industrial and consumer; ~12-16% of revenue. Smaller but higher-margin, longer-lifecycle and more stable — valuable diversification against the commodity core.

Geographic Revenue Assumptions

Micron reports revenue by ship-to destination; we hold the geographic mix broadly constant in our model given limited disclosure. We assume Mainland China + Hong Kong ~26% (constrained by the CAC purchasing ban on critical-infrastructure operators but still a large assembly/OEM hub), the United States ~18% (rising over time with reshoring), Taiwan ~18% (assembly base), Other Asia-Pacific ~20%, Japan ~8%, Europe ~7%, and Other ~3%. The principal geographic risk is escalation of U.S.–China trade restrictions, which could impair both China end-market access and the Asian manufacturing footprint.

Margin, Operating Expense, Capex & Working-Capital Assumptions

■ Gross margin: the master cyclical driver. We model 74.0% in FY26E (blending reported H1 ~68% with the ~81% FQ3 guide and an assumed FQ4), 70.0% in FY27E, then normalization to 52.0% (FY28E), 38.0% (FY29E) and 48.0% (FY30E). For reference, FY23A gross margin was -9.1% and FY25A was 39.8%.

■ R&D: grows from $3.8B (FY25A) to ~$4.7-5.4B across the forecast, reflecting sustained leading-edge and HBM investment; it falls sharply as a percent of revenue at the peak (~4.5% FY26E) and rises in the trough.

■ SG&A: modeled at ~$1.45-1.55B, broadly flat in dollars and de minimis as a percent of peak revenue.

■ Capital expenditure: ~$30B in FY26E (management guides >$25B), stepping up to ~$40B in FY27E (>$10B more construction-related, per guidance) before tapering to ~$28B. This is the key constraint on free cash flow and the core risk of building into an eventual oversupply.

■ Depreciation & amortization: rises from $8.4B (FY25A) toward ~$15.5B (FY30E) as the ~$200B capacity program comes online.

■ Working capital: modeled via DSO of 80-90 days, inventory of 120-150 days of COGS, and payables/accruals scaling with COGS and revenue. Receivables balloon at the revenue peak (FQ2-26 receivables reached $17.3B).

■ Tax rate: ~10-15% effective, consistent with recent quarters and the mix of U.S. and foreign earnings plus CHIPS-related credits.

5 · Scenario Analysis

Because Micron’s earnings are so cyclical, a single point estimate understates the genuine uncertainty. We frame three scenarios with explicit parameters and assign probabilities. The dispersion in out-year EPS — from $4 (bear, FY29E) to $66+ (bull) — is the essence of a memory-cycle investment.

Bull Case — Durable Supercycle (35% probability) — value ~$1,250

In the bull case, AI memory demand structurally outstrips supply through the decade and the three-player oligopoly maintains pricing discipline. Key parameters: gross margin holds at 66-76% across the forecast (versus a base-case normalization to 38%); FY2030E EPS reaches ~$73 (versus ~$29 base); and HBM4/HBM4E pricing holds as Micron converges toward SK Hynix on share. Catalysts required: (1) HBM sold-out conditions extend into 2027-28 on multi-year contracts; (2) hyperscaler AI capex continues to accelerate rather than digest; (3) no meaningful Chinese encroachment at the leading edge; (4) Micron’s HBM share climbs from ~21% toward ~30%. Under these conditions the market continues to capitalize near-peak earnings at ~14-16x, supporting a value of ~$1,250 (+33% from current). This is effectively the scenario the current share price already reflects.

Base Case — Cycle Normalizes (45% probability) — value ~$700

Our base case assumes the supercycle is real but not permanent. FY2026-27 represent the peak; as capacity additions catch up with demand, FY2028-29 see a classic normalization — gross margin recedes from ~74% to ~38% and EPS falls from ~$53 to ~$18 before a FY2030 recovery. Critically, the market de-rates the multiple as earnings roll over, converging toward the 7-11x forward multiples at which memory peers trade today. The combination of lower earnings and a lower multiple yields a value of ~$700 (-26% from current). We regard this as the single most likely path, which is why a stock trading at $942 carries no margin of safety.

Bear Case — Down-Cycle (20% probability) — value ~$350

The bear case is a repeat of memory history. Triggers: (1) AI capital spending digestion as hyperscalers pause after a multi-year build; (2) HBM and DRAM oversupply as all three incumbents — plus subsidized Chinese entrants CXMT and YMTC — add capacity simultaneously; (3) a loss of pricing discipline that collapses ASPs. Under these conditions earnings approach the FY2023 experience: gross margin compresses toward the 20s-30s, FY2029E EPS falls to ~$4, and the equity de-rates hard on both lower earnings and a trough multiple. We estimate a value of ~$350 (-63% from current). The probability is not negligible: every prior memory peak has been followed by a down-cycle.

Scenario Comparison

The path dependency is severe: the same company is worth ~$1,250 or ~$350 depending almost entirely on where we sit in the memory cycle two years hence. This is why we anchor our target on the probability-weighted blend and on mid-cycle earnings power rather than peak FY2026 results.

6 · Valuation — and the Lab

Methodology. We discount the Task 2 model’s unlevered free cash flow for FY2027E–FY2030E and apply a mid-cycle perpetuity terminal value, valued as of end-FY2026E.

WACC = 11.4%. Risk-free rate 4.45% (10Y UST, May-26), equity risk premium 5.0%, levered beta 1.40 (reflecting deep cyclicality) → cost of equity 11.45%. With ~$1.08T equity against ~$10B debt, WACC ≈ cost of equity.

PV of explicit FCF: $62.7B. Note the FCF dip in FY28–29 — the model’s deliberate assumption that the cycle normalizes. This is precisely why the terminal value must rest on mid-cycle economics.

Terminal value (mid-cycle perpetuity). Normalized unlevered FCF $16.5B (mid-cycle NOPAT on ~$35B EBIT, plus ~$14.5B D&A, less ~$28B normalized capex — memory FCF is structurally capped by relentless capex), g = 3.0%: - TV (end-FY30) = $16.5B × 1.03 / (0.114 − 0.03) = $202.9B - PV of TV = $131.8B (67.8% of EV — passes the <70% sanity check)

Sensitivity (WACC × terminal growth): the per-share value ranges from ~$151 (13.5% WACC, 2% g) to ~$266 (9.5% WACC, 4% g). Across the entire realistic parameter space, the mid-cycle DCF stays a small fraction of $942. The unavoidable conclusion: the DCF is not how this stock is being priced. The market is valuing MU on near-peak earnings power treated as durable — the bull thesis. We give the DCF only 20% weight in the blended target precisely because terminal value is too assumption-sensitive for a hyper-cyclical; but it is an essential discipline anchor and it flashes “expensive.”

Peer set. The cleanest comparables are the other memory makers; we exclude Western Digital from the statistical set as it is now an HDD pure-play (post-Sandisk spin).

¹ EV/EBITDA mixes methodologies: foreign peers (Korea, Japan) reflect analyst-curated forward estimates; US peers (SNDK, WDC) are TTM trailing pulled from yfinance via the kaamos ETL. The qualifier is shown per row; the chart drops “Fwd” from the column label for the same reason.

The cyclical signature. Peers trade at 7–11x forward earnings — low multiples, because the market knows these are peak earnings that will fade. (The cyclical trap: memory looks “cheapest” at the top and “most expensive” at the bottom.) MU at $942 is ~17x our FY27E EPS of $54.62 — a substantial premium to this group.

Premium justification. MU arguably deserves a premium to its Korean peers: it is the fastest HBM share-gainer, the only U.S.-domiciled supplier (CHIPS support, supply-chain-diversification demand), and carries a U.S.-listing liquidity premium. We therefore center comps on a ~12x P/E and ~8.5x EV/EBITDA — giving an HBM-growth premium — which yields ~$650. But even a generous bull multiple (16x P/E) only reaches ~$874, still below the current $942. On comps, the stock is at-to-above fair value even granting a premium.

Not a meaningful method here. At ~$1.08 trillion, Micron is effectively un-acquirable — there is no strategic or financial buyer of that scale, and any memory consolidation would face insurmountable antitrust scrutiny (the deal would reduce DRAM from three players to two). The relevant historical transactions (Micron/Elpida 2013, ~$2.5B; Micron/Inotera 2016) were distressed-asset deals at trough valuations and offer no read-through to today. We assign precedent transactions 0% weight.

Which multiples to actually trust for MU

DCF and forward P/E peer comps tell us what MU is worth. Historical valuation bands (the kaamos 8-year weekly series for every standard multiple) tell us how the market has actually paid for MU through cycles — and that “how” is what dictates short- and medium-term price action for a hyper-cyclical. Not every multiple is equally useful. We pick three and explicitly discard the rest:

• EV / EBITDA — primary. The least-bad single multiple for memory. Capex distorts net income; depreciation schedules distort it more; book value gets impaired in down-cycles. EBITDA at least anchors on the operating cash-engine and reads cleanly across the cycle. MU sits within 0.2σ of its own 8-year mean — the cleanest single signal that, on this metric, the market is paying a roughly mid-cycle price.
• P / Sales — primary (with caveat). Revenue isn’t whipsawed by margin swings, so the P/S band shows where the multiple has been re-rated as a franchise. MU at ~19x sales vs an 8-year mean of ~3.8x is +6σ — a regime-change reading. Interpret as: the market has decisively moved MU out of its “cyclical commodity” valuation regime into an “AI-memory franchise” regime. This is the bull thesis, quantified.
• P / E (TTM) — secondary. Trailing earnings still embed FY23 trough distortions (the 8y mean and SD are dominated by trough quarters where reported P/E spiked into triple digits). The +1σ z-score reading is directionally correct but the absolute SD is too noisy to anchor a trade. Use the forward P/E peer-comp framing above as the primary earnings-multiple anchor.

We exclude P/B (book value impaired in down-cycles → denominator collapse, not real richness), P/CF (mirrors EBITDA), P/FCF (whipsawed by capex timing), and dividend yield (negligible).

The bands chart per ratio below shows the full 8-year history with ±1σ and ±2σ shading; the chip strip at the top of Section 1 carries the snapshot. Net read: on EV/EBITDA (the multiple cyclicals are actually paid on), MU is fairly valued vs its own history — but P/S says the market has priced in a permanent regime shift that the DCF and peer comps don’t support. That’s the entire bull/bear tension, distilled.

The current price sits above the high end of the DCF and both comps ranges and only inside the bull-tilted scenario range. That is the visual statement of an over-extended cyclical.

Scenario Fair Value: $820 (rounding the scenario-weighted blend, which we favor over the lower method-weighted $610 because cyclicals can and do trade on near-term earnings power for extended periods, and we give real weight to the bull regime). The stock at $942 sits 13% above our FV — an expensive read, not a directional 12-month call.

View: Neutral — scenario value $820 (−13%)

We are not bearish on the franchise; we are disciplined on the price. The risk/reward at $942 is negatively skewed: ~+33% to the bull case versus ~−26% to base and ~−63% to bear. Our view would shift to Positive below ~$700 (where the base case offers a margin of safety and the bull case becomes a multi-bagger) and to Cautious above ~$1,050.

Catalysts (could move us more positive)

1. HBM4 ramp / NVIDIA Vera Rubin — share and pricing data points beating expectations would extend the bull case.
2. Continued sold-out HBM and multi-year contract extensions into 2027 — would lengthen earnings visibility and justify a higher normalized FCF (the key DCF lever).
3. Demonstrated supply discipline by all three DRAM players — the single biggest determinant of whether the up-cycle persists.
4. Capex returns — evidence the ~$200B U.S. program earns its cost of capital rather than seeding the next glut.
5. Capital returns — larger buybacks/dividends deploying the peak-cycle cash pile (the model shows cash building toward $100B+ absent returns).

Risks to the target

Downside: (1) Memory down-cycle — the recurring, near-certain-eventually risk; HBM/DRAM oversupply compresses margins fast (high impact). (2) China competition — CXMT/YMTC capacity undercuts commodity pricing and erodes oligopoly discipline (structural). (3) AI capex digestion — a hyperscaler spending pause hits MU’s most profitable, most concentrated revenue (medium probability, high impact). (4) Customer concentration in HBM (NVIDIA/hyperscalers). Upside: (1) Durable supercycle — AI memory demand proves structurally higher for longer; MU re-rates further (the bull case). (2) Beat-and-raise continuation. (3) HBM share gains beyond 21% at premium pricing.

The one-line thesis

Best-in-class execution into the best memory cycle in history — but at $942 the stock already prices the supercycle as permanent. Exceptional company; fully-valued stock. Wait for the cycle, not the headline.

7 · Company

Micron Technology, Inc. is one of the world’s three dominant producers of memory and storage semiconductors, and the only such producer headquartered in the United States. Founded in 1978 and based in Boise, Idaho, Micron designs and manufactures dynamic random-access memory (DRAM) and NAND flash memory — the two product categories that, together, store and move essentially all of the world’s digital data. DRAM is the high-speed “working memory” that sits alongside processors to feed them data in real time; NAND is the non-volatile storage that retains data when power is off, underpinning solid-state drives (SSDs), smartphones, and embedded systems. Micron sells these products both as discrete components and as higher-value modules and subsystems, increasingly under its own Micron and Crucial brands.

The company’s defining characteristic — and the source of both its historical volatility and its current extraordinary profitability — is that memory is a commodity-like, capital-intensive, cyclical industry consolidated into an oligopoly. DRAM is effectively a three-player market (Samsung, SK Hynix, and Micron), and NAND a five-to-six player market. Pricing is set by the balance of industry bit supply against end-market demand, and because fabrication capacity takes years and many billions of dollars to build, supply and demand routinely fall out of balance, producing dramatic boom-bust cycles. Micron’s revenue has historically swung from trough years of operating losses to peak years of extraordinary margins within the span of 24–36 months.

As of fiscal 2025 (the year ended August 28, 2025), Micron generated revenue of $37.4 billion, up 49% year over year, with GAAP net income of $8.5 billion and a gross margin of 40%. The business then accelerated violently into fiscal 2026 on the back of artificial-intelligence (AI) demand: fiscal Q2 2026 (ended February 26, 2026) revenue reached $23.9 billion — a single quarter nearly two-thirds the size of all of FY2025 — up 75% sequentially and 196% year over year, with non-GAAP EPS of $12.20. Management guided fiscal Q3 2026 to record revenue of approximately $33.5 billion at a roughly 81% gross margin and EPS of about $19.15. The stock has responded accordingly, reaching approximately $942 per share in late May 2026 and a market capitalization near $1 trillion, having appreciated roughly 850% over the prior twelve months.

The engine of this surge is high-bandwidth memory (HBM) — a specialized, vertically-stacked DRAM product that is the critical memory companion to AI accelerators (GPUs and custom silicon) from NVIDIA, AMD, and the hyperscale cloud providers. Each AI server consumes vastly more memory content than a traditional server, and HBM in particular is supply-constrained, technically demanding, and sold under multi-year contracts. Micron has stated that its HBM output is effectively sold out through 2026, and it began volume production of next-generation HBM4 for NVIDIA’s Vera Rubin platform in fiscal Q1 2026.

Micron operates a globally integrated manufacturing footprint — fabrication facilities and assembly/test operations across the United States, Taiwan, Japan, Singapore, China, and Malaysia — employing roughly 48,000 people worldwide. The company is in the midst of an enormous capital expansion, having articulated plans to invest on the order of $200 billion in the United States over the long term (approximately $150 billion in manufacturing and $50 billion in R&D), including up to four leading-edge fabs in New York State, two in Idaho, expansion of its Virginia facility, and domestic advanced-packaging capacity for HBM. This expansion is supported in part by approximately $6.4 billion in direct funding under the U.S. CHIPS and Science Act. Fiscal 2026 capital expenditure is guided above $25 billion, with fiscal 2027 stepping up meaningfully further.

In short, Micron is a structurally improved, oligopoly-positioned memory leader experiencing the most favorable demand environment in its history, leveraged to the secular build-out of AI infrastructure — but it remains, at its core, a cyclical commodity manufacturer whose earnings power is a function of an industry supply-demand balance that has historically proven impossible to sustain at peak levels indefinitely. Any investment view on Micron is therefore inseparable from a view on where the company sits in the memory cycle and how durable AI-driven demand proves to be.

Business mix & divisional economics

8 · Risks to the Target

Company-Specific Risks

Cyclical earnings and “peak cycle” risk. Micron’s single greatest risk is that current earnings reflect an unsustainable cyclical peak. With fiscal Q3 2026 guided to ~81% gross margin and EPS near $19, the company is operating at extraordinary, historically anomalous profitability. Memory peaks have always proven temporary; valuing the company on peak earnings is dangerous, and an investor buying near a ~$1 trillion market capitalization is implicitly betting on durability that the industry’s history argues against. The likelihood of eventual normalization is high; the timing and depth are the uncertainty.

Customer concentration in AI/HBM. A growing share of Micron’s most profitable revenue derives from a small number of hyperscalers and AI accelerator vendors (notably NVIDIA). This concentration creates outsized exposure to any single customer’s roadmap changes, in-sourcing efforts, or capital-spending pullback, and amplifies the impact of an AI-spending deceleration on Micron’s highest-margin business.

Execution and yield risk on HBM4 and node transitions. Micron’s HBM share gains and margin profile depend on flawlessly ramping technically demanding products (HBM4, advanced DDR5) at high yield. Stacking, advanced packaging, and leading-edge node transitions are difficult; a yield stumble or a delayed qualification on a flagship platform (as Samsung experienced) could cede share and margin to rivals.

Capital-intensity and over-build risk. Micron is committing to >$25 billion in FY2026 capex and a ~$200 billion long-term U.S. program. If demand softens after this capacity comes online, Micron — and the industry — risks building directly into an oversupply, converting today’s scarcity-driven margins into tomorrow’s losses. The returns on this enormous investment are unproven and depend on sustained demand.

Smaller scale than rivals. Micron is the smallest of the three DRAM makers, with less balance-sheet capacity than Samsung or SK Hynix to sustain a prolonged capacity/technology investment race, leaving it potentially disadvantaged if competitors choose to out-spend it through a downturn.

Industry / Market Risks

Competitive intensity and loss of supply discipline. The favorable economics of memory rest entirely on three rational DRAM players exercising capacity discipline. Should any competitor (or a subsidized Chinese entrant) prioritize share over profitability and flood the market with capacity, pricing would collapse industry-wide. Micron controls none of this; it is exposed to the rationality of others.

Chinese state-backed competition (CXMT, YMTC). Subsidized Chinese memory makers represent the most serious structural long-term threat. Even confined to trailing-edge nodes, their capacity-driven, share-seeking behavior can depress commodity DRAM and NAND pricing globally and erode the oligopoly discipline that underpins current margins, while also displacing Micron within China.

HBM oversupply and demand-normalization risk. The entire industry is racing to add HBM capacity. If AI demand growth moderates or if HBM4/HBM4E capacity additions outpace accelerator demand, the premium pricing and scarcity that define today’s HBM economics could compress sharply — turning Micron’s best business into a margin headwind.

Technology disruption. Long-term, alternative memory architectures, changes in AI compute design that reduce memory intensity, or shifts in the relative value of DRAM vs. NAND could alter the competitive and economic landscape.

Financial Risks

Earnings and cash-flow volatility. Micron’s financials swing violently with the cycle — from operating losses in the 2023 trough to record profits now. This volatility complicates valuation, capital planning, and shareholder returns, and means reported earnings can deteriorate far faster than most investors anticipate.

Leverage and funding through a downturn. Micron carries roughly $14 billion of long-term debt against ~$9.6 billion cash and ~$82.8 billion total assets (FY2025). While manageable at current cash generation, the combination of heavy committed capex and debt service could pressure the balance sheet if the cycle turns sharply while expansion spending remains elevated.

Capex/free-cash-flow tension. Record operating cash flow is being substantially consumed by record capital expenditure; free cash flow, while positive, is a fraction of earnings during the build phase, constraining the pace of buybacks and dividends and leaving less cushion if results weaken.

Macroeconomic Risks

Demand cyclicality and economic sensitivity. Memory demand is highly sensitive to the broader economy — enterprise IT spending, smartphone and PC unit sales, and ultimately AI capital budgets all flex with macro conditions. A recession or a pullback in corporate AI/cloud capex would directly and quickly impair Micron’s volumes and pricing.

Geopolitical and trade risk. Micron is squarely exposed to U.S.–China tensions: the CAC purchasing ban already limits its China access, and any escalation in export controls, tariffs, or retaliatory measures could disrupt its substantial Asian manufacturing footprint and end-market access. As the lone U.S. memory champion, Micron is a natural focal point of trade policy on both sides.

Foreign-exchange and input-cost exposure. With manufacturing and sales spread across the U.S., Taiwan, Japan, Singapore, China, and Malaysia, Micron is exposed to currency fluctuations and to the cost and availability of specialized equipment, materials, and energy required for fabrication.

======================================

Catalysts to watch

• HBM4 / NVIDIA Vera Rubin ramp data points beating expectations
• HBM sold-out conditions and multi-year contracts extending into 2027
• Demonstrated industry supply discipline across all three DRAM players
• Larger capital returns deploying the peak-cycle cash pile

Upcoming events

• 2026-06-25 — FQ3-26 earnings (approx.) (Guide: revenue $33.5B ±$750M, GM ~81%, EPS $18.90 ±$0.40)