Position to be managed in the supercycle of memory but too expensive for long-term hold

Target Entry Point: Under $350 - ride 12 month supercycle (if you have appetite)

Micron represents a cycle-aware position in the $200B+ global memory semiconductor market, entering the most profitable period in its 47-year history driven by AI-fueled demand for High Bandwidth Memory (HBM), data center DRAM, and enterprise SSDs. FY2026 is a confirmed supercycle: Q1 delivered $13.64B in record revenue (+57% YoY), Q2 is guided at $18.7B with 68% gross margins, and full-year consensus is ~$76B in revenue with ~$34 EPS. The core thesis — that AI structurally elevates memory from commodity component to strategic infrastructure — is validated by Micron's entire 2026 HBM supply being sold out under finalized contracts. However, two material headwinds have emerged since the thesis was formed: (1) Micron has been excluded from Nvidia's flagship Vera Rubin HBM4 slots, with Samsung and SK Hynix taking 100% of the allocation, and (2) insider selling has been aggressive, with CEO Mehrotra executing 60 consecutive sales and zero purchases over the past five years. These realities, combined with memory's deep cyclical history — revenue fell 50% peak-to-trough in FY2022-23 — require active position management rather than passive conviction. We recommend a 5–7% allocation on pullbacks below $350, with a 12-month target of $480–520 and a probability-weighted 2031 target of $28–35 per pre-split equivalent, or $280–350 at current share count.

Revenue 2031: $80B

Starting from a ~$76B FY2026 consensus run-rate (Q1 actual $13.64B + Q2 guided $18.7B + H2 acceleration). The bridge has three components:

Cycle Path (the unavoidable reality): Memory has never delivered monotonic growth over any 5-year period in its history. FY2018 peaked at $30.4B; by FY2023, revenue had halved to $15.5B. FY2026's ~$76B peak will be followed by a downcycle as $20B+ of industry capex (Micron Idaho fabs, Samsung P5, SK Hynix M15X) brings new capacity online in 2027–2028. We model a 25–30% peak-to-trough revenue decline to ~$55B in FY2028–29 — shallower than the historical 50% because contracted HBM volumes and structural AI demand create a floor that did not exist in prior cycles. Recovery to $75–85B by FY2031 as AI inference proliferates to edge, automotive, and enterprise.

Industry TAM Growth: The total memory market surpassed $200B in 2025 and is projected to reach $310–350B by 2031 (8–9% CAGR per Yole Group, Fortune Business Insights, and Research & Markets). DRAM alone is projected at $224B by 2034 at 7.2% CAGR. HBM specifically is projected to grow from $35B in 2025 to $100B by 2028 at ~40% CAGR, with Yole Group projecting ~$98B by 2030. This expanding TAM supports recovery beyond prior peaks even after a downcycle.

Market Share: Micron holds ~25% of DRAM and ~10–12% of NAND, for a blended ~20–22% of total memory. The HBM4 Vera Rubin exclusion is a real setback — SemiAnalysis reduced Micron's Nvidia HBM4 share to zero for the initial Rubin build, with SK Hynix taking ~70% and Samsung ~30%. However, each new HBM generation is a fresh competitive opportunity (HBM4E in 2027–28, HBM5 beyond), and Micron retains HBM3E supply for current Blackwell platforms and HBM4 for mid-tier Rubin CPX. We model blended share stabilizing at 21–23% through 2031, reflecting modest HBM share loss offset by gains in data center SSDs (where Q1 revenue exceeded $1B for the first time) and SOCAMM (a market Micron pioneered).

Calculation: FY2031 memory TAM of ~$330B × 22% blended share = ~$72B memory revenue + ~$8B in SSD solutions/premium mix = ~$80B total. This implies that FY2031 revenue approximately matches FY2026's peak — which sounds conservative, but reflects the reality that FY2026 is an anomalous supercycle year (103% YoY growth) and FY2031 is modeled as a solid mid-cycle recovery year, not another peak. From the FY2028–29 trough of ~$55B, $80B represents ~45% recovery growth, which is consistent with historical patterns (FY2023 trough of $15.5B to FY2025 of $37.4B = 141% recovery).

Net Margin 2031: 23%

Current peak is extraordinary: Q1 FY2026 delivered 38% net margin, with Q2 guided to approach 42% given 68% gross margins and relatively fixed opex. But Micron's margin history is the single most important thing any investor must internalize:

Historical Margin Reality: The 10-year average gross margin is 33.06%. The 5-year average is only 27.34%. The 10-year average net margin is approximately 15%. Peak-cycle gross margins (FY2018: 58.9%, FY2026E: ~65%) have always compressed violently — FY2023 saw negative gross margins. This is not a bug; it is the structural feature of a high-fixed-cost, capital-intensive business where ASPs are set by supply/demand, not cost-plus.

What AI Changes (and Doesn't): AI structurally elevates the margin floor through three mechanisms: (1) HBM carries 2–5x the ASP of commodity DRAM with higher margins, and by 2031 should represent ~25–30% of Micron's revenue; (2) multiyear contracted volumes (a new feature of this cycle) reduce spot-market exposure and revenue volatility; (3) data center SSD and SOCAMM products carry higher margins than consumer NAND. We estimate these structural shifts lift the through-cycle gross margin floor from the historical ~33% to ~40–45%, and the through-cycle net margin from ~15% to ~20–25%.

Why 23% and Not Higher: A FY2031 mid-recovery year will have gross margins in the 42–48% range (above historical average, below current peaks). R&D and SG&A will be higher in absolute terms as Micron scales to $80B+ in revenue capacity. Depreciation from $20B+ annual capex cycles will weigh on net margins. Interest expense on debt supporting fab buildouts adds ~$300–400M annually. Tax rate normalizes at ~14–16% (consistent with Micron's current effective rate with CHIPS Act benefits). The net result: 23% net margin on $80B revenue produces ~$18.4B in net income.

Why Not Lower: Even the competing analyst's 22% estimate is close to ours. The structural shift to high-value products is real and confirmed by Q1 FY2026's business unit margins: Cloud Memory at 66% gross, Core Data Center at 51%, even Automotive/Embedded at 45%. A through-cycle average incorporating one downcycle year and one recovery year lands at 20–25%. We take the midpoint.

P/E Multiple 2031: 14x

Historical Precedent: Micron's 10-year average P/E is 15.88x. The 15-year average is 14.29x. The 20-year average is 13.51x. The median over 13 years is 14.58x. The stock has traded as low as 2.5x at cycle troughs and as high as 164x during earnings troughs (inflated denominators). The market has a deep, persistent habit of valuing memory as cyclical commodity — and 40 years of data largely justify this.

Why 14x and Not Higher: By 2031, the AI re-rating narrative will have matured. The initial "discovery" premium (the current phase where the market is learning that memory is AI infrastructure) will have been arbitraged away. New capacity from Idaho fabs, NY fabs, Samsung P5, and SK Hynix M15X will have normalized the supply picture. Chinese competitors (CXMT in DRAM, YMTC in NAND) will have expanded further. Micron's HBM competitive position is uncertain after the Vera Rubin exclusion — if subsequent generations (HBM4E, HBM5) show similar losses, the market will not grant a premium. The capital structure is clean (no preferred layers like QXO), but the $10B buyback authorization has only consumed $7.49B through November 2025, with just $300M repurchased in Q1 FY2026 despite record FCF — suggesting management is not prioritizing share count reduction during the peak. A 14x multiple prices Micron at the historical average, reflecting a mature, well-run memory company that benefits from AI but remains cyclical. This is deliberately conservative.

Why Not Lower: Even at 14x, we're assuming Micron doesn't fully re-rate as structural AI infrastructure. If multiyear contracted HBM revenues reach 40%+ of the total by 2031, reducing cyclical volatility, the market could award 16–18x (the bull case). Peers like SK Hynix trade at 10–13x forward, but they have lower institutional ownership and less liquidity. Broadcom (28x), which also serves AI infrastructure, provides an upper bound that memory will never reach but can aspire toward in structure. 14x is the number where we'd be comfortable being wrong in either direction.

Conclusion: $80B Revenue × 23% Net Margin = $18.4B Net Income

Diluted share count: ~1.13B shares today. Micron has a $10B buyback authorization ($2.5B remaining). At current capex intensity, buybacks will be modest — we model net share count of ~1.10B by FY2031, reflecting gradual repurchases offset by stock-based compensation dilution. This produces:

EPS 2031: $18.4B ÷ 1.10B = ~$16.70

Base Case Price: $16.70 × 14x = ~$234

This is the single-point normalized mid-cycle estimate. But memory doesn't produce "normalized" years — it produces peaks and troughs. The scenario range:

ScenarioProbabilityFY2031 EPSP/EPriceWeightedBear: FY2031 is a trough year15%$612x$72$11Below-Base: Shallow recovery, HBM share loss persists20%$1213x$156$31Base: Solid mid-cycle, HBM share partially recovered35%$1714x$238$83Above-Base: Strong recovery, HBM4E/5 wins20%$2415x$360$72Bull: Near-peak year, AI re-rating10%$3217x$544$54Probability-Weighted 2031 Value~$251

At $378, the stock is priced above the probability-weighted 2031 value of ~$251 on a 5-year hold-to-maturity basis. This does not mean it's a bad buy — it means the return is front-loaded into the FY2026–27 supercycle, and an investor must actively manage the position through the cycle.

The trade: Build a 5–7% position on pullbacks below $350 (the HBM4 news may create this opportunity near-term). Hold through FY2026–27 earnings expansion targeting $480–520 exit (13–15x on ~$35–40 FY2027 EPS). Reduce to 2% before FY2028 capacity additions hit. Re-enter at $150–200 during the FY2029 trough. Hold through recovery. An investor who executes both legs compounds at 15–25% annualized — far better than the -6% CAGR of buying at $378 and holding passively to the $251 probability-weighted target.

Position sizing rationale: 5–7% (not 10%) reflects the downgrade from the original thesis. The HBM4 Vera Rubin exclusion, aggressive insider selling (60 consecutive CEO sells, zero buys), and $20B capex intensity all introduce execution risk that merits a smaller position than a conviction overweight. Scale to 7% only on pullbacks below $350 where the risk/reward improves materially.