MU

MU — Micron Technology

TL;DR

Micron’s Feb-2026 quarter was among the largest single-quarter earnings beats in semiconductor history: $23.9Brevenue(vs.buy-side$20.1B), 75% gross margin (record), $19.15 EPS guided for May-2026 quarter. The thesis is a multi-year undersupply of DRAM and NAND driven by HBM capacity consumption, tech-node migration capacity loss, and AI datacenter demand that now exceeds 50% of total industry. The one structural risk: Micron is “effectively out of the picture” for Rubin HBM4 per SemiAnalysis — SK Hynix and Samsung lead on qualification.¹²

Business

Designs and manufactures DRAM, NAND, and HBM (High Bandwidth Memory). Three primary end markets: Data Center (AI training/inference DRAM + HBM + SSDs), PC/Smartphone (DRAM + NAND), and Automotive/Embedded (AEBU). Global fab footprint: Idaho (Boise HQ), New York (under construction), Taiwan (Tongluo acquired from PSMC, DRAM from FY2028), Singapore (NAND under construction, CY2H28). HBM advanced packaging in Singapore coming 2027.¹

Thesis

• Secular undersupply — structural, not cyclical. Management says it can only meet 50–67% of medium-term demand for some key customers. Internal DRAM inventory below 120 days (extreme tightness). Causes: (1) HBM production consumes significant wafer area that would otherwise produce standard DRAM; (2) technology node migration (1-gamma) creates capacity loss during transition; (3) cleanroom space constraints with multi-year fab lead times. FundaAI projects supply shortage to persist beyond 2026.¹
• HBM3E volume + HBM4 ramp. Volume shipment of 36GB 12-hi HBM4 began CY1Q26 for Nvidia’s Vera Rubin platform. Samples of 48GB 16-hi HBM4 (next tier) already delivered to customers. AWS Trainium3 switched from Samsung HBM3E (slow 5.7 Gbps) to Hynix/Micron HBM3E hitting 9.6 Gbps — highest HBM3E speeds achieved.³²
• Datacenter share expanding — AI the primary driver. Data Center DRAM + NAND will exceed 50% of total industry demand for the first time in 2026. AI + server demand is growing in low-teens percent range. Long-term roadmap: L4 autonomous driving requires >300GB DRAM per vehicle; humanoid robots described as a 20-year growth vector.¹
• ASP + mix shift driving massive margin uplift. Feb-2026: gross margin 75% (record); May-2026 guided 81%. 1-gamma DRAM and G9 NAND nodes becoming primary bit output mix by mid-2026 — lower cost per bit at same price = gross margin tailwind. First 5-year Strategic Customer Agreement (SCA) signed — provides pricing floor across multiple cycles.¹

Risks

• HBM4 disqualification for Rubin. SemiAnalysis: Micron is “well behind Samsung and Hynix” on HBM4 pin speed for Rubin (Nvidia targets 11 Gb/s) and “effectively out of the picture for Rubin HBM4.” Irrational Analysis pushes back — HBM4 qual at probe stations is notoriously noisy — but the divergence between Hynix/Samsung and Micron performance is real.²⁴
• PC + smartphone unit headwinds. Supply constraints pushing PC/smartphone shipments down low-double-digits in 2026. While per-unit memory capacity is rising (Agentic AI PCs recommended at 32GB, 2× traditional average), unit volume decline is a headwind to DRAM bit growth outside datacenter.¹
• Capex cycle execution. FY26 capex >$25B(upfrom$20B); FY27 construction spending +$10B. New fabs in Taiwan, NY, Singapore don’t produce meaningful bit output until 2027–2028. Execution risk on yield ramp at new nodes and sites.¹
• Customer concentration. 5-year SCA provides stability but also locks Micron into pricing/volume commitments. If AI capex pulls back, the SCA could become a margin risk at the bottom of the next cycle.

Recent catalysts

• 2026-04-14 — Edgewater Research (OUTPERFORM, $426.56; ST/LT Positive): HBM CY26 supply share: Hynix 60–70% (18–20B 1Gb), MU 7–8B (~15%), Samsung 10–12B (~20%); all three suppliers wary of adding HBM supply in favor of more profitable RDIMMs. DRAM price increases expected to moderate in C2H26. NVDA forecasting 23–25B 1Gb DRAM demand CY26, potentially growing to 60–70B in CY27. MU pricing tracking ahead of model.⁵
• 2026-04-14 — Cleveland Research: DRAM ASPs +45–55% Q/Q in CQ2, expected to moderate to +10–15% in CQ3. Guidance (Non-GAAP): Rev $8.4–$8.85B; EPS $12.00–$14.00. FY26 consensus estimate: Rev $34.58B(+29$41.34B (+19% Y/Y).⁶
• 2026-03-19 — FundaAI MU earnings review: FY2Q26 $23.9Brevenuevs.Street$19.2B; 75% gross margin; $12.2EPSvs.Street$8.6. FY3Q guidance $33.5B/81$19.15 EPS — all massively ahead of buy-side consensus.¹
• 2026-03-19 — MU Board approved 30% dividend increase to $0.15/quarter;record$6.9B free cash flow.¹
• 2026-03-12 — SemiAnalysis: HBM bit shipments inflecting sharply; rising capacity per accelerator (NVDA Rubin Ultra +4× HBM, AMD MI400 +50%) driving content gains independent of unit counts.³
• 2026-02-25 — SemiAnalysis Vera Rubin: Micron out of picture for Rubin HBM4; SK Hynix and Samsung progressing; initial Rubin shipments will come in at ~20TB/s vs. 22TB/s target.²
• 2026-02-12 — Irrational Analysis: HBM4 qual noise — SemiAnalysis “Micron failed” report may be premature given probe-station calibration difficulty at 11 Gbps; all three vendors working with Nvidia.⁴
• 2026-01-16 — Irrational Analysis Memory Madness: MU director insider buy of $8M personal cash — unusual conviction signal; author re-entered MU long on this signal.⁷

Second-order reads

• 2026-04-14 — Edgewater Research, Memory/Storage Update — NVDA forecasting 23–25B 1Gb DRAM CY26, growing to 60–70B in CY27; HBM suppliers (Hynix, MU, Samsung) all wary of adding HBM capacity in favor of more profitable standard DRAM → bullish for server/datacenter DRAM pricing. HBM4 quals delayed for all three suppliers, Rubin timeline risk → negative tail for NVDA GPU shipment cadence.⁵
• 2026-03-19 — FundaAI, MU Review — SNDK EPS estimates revised up alongside MU; memory supercycle benefits both → positive for SNDK.
• 2026-03-12 — SemiAnalysis, Great AI Silicon Shortage — HBM capacity consuming wafer area that would otherwise produce standard DRAM → tighter DRAM pricing positive for MU/SNDK; also negative second-order for NVDA (accelerator COGS, HBM4 supply risk).
• 2025-12-04 — SemiAnalysis, Trainium3 — AWS switching from Samsung HBM3E (5.7 Gbps, subpar) to Hynix/Micron (9.6 Gbps) for Trn3 → MU wins HBM3E allocation in AWS supply chain; negative for Samsung HBM business.

Valuation & positioning

FundaAI projects MU CY26/27 EPS of $92/$143. May-2026 quarter: ~$19EPSrun-rateimpliestrailing$76/year. FY28 gross margin structure depends on new fab output hitting plan. First 5-year SCA provides pricing floor.¹ Edgewater (OUTPERFORM, $426.56)seesFY26pricingtrackingaheadofmodel;viewsMUasST/LTpositivewithupsidetoconsensus.ClevelandResearchFY26consensusestimate:Rev$34.58B (+29% Y/Y), FY27 $41.34B (+19% Y/Y).⁵⁶

Sources

Related

SNDK — NAND companion; memory cycle moves in lockstep NVDA — primary HBM customer (Hopper/Blackwell/Rubin); Micron out of Rubin HBM4 per SemiAnalysis AMD — HBM3E customer (MI355X/MI450X); Micron competing for Instinct socket TSM — CoWoS-L packaging for HBM; TSMC capacity is gating factor for HBM qualification ONTO — metrology equipment; HBM yields are measurement-intensive KLAC — process control equipment; HBM ramp benefits equipment vendors

Footnotes

1. FundaAI — Review: MU Results and Guidance Crushes Expectations — 2026-03-19 ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹ ↩¹⁰

2. SemiAnalysis — Vera Rubin: Extreme Co-Design — 2026-02-25 ↩ ↩² ↩³ ↩⁴

3. SemiAnalysis — The Great AI Silicon Shortage — 2026-03-12 ↩ ↩²

4. Irrational Analysis — Market Memo: Tower+GloFo, HBM4 Qual Noise — 2026-02-12 ↩ ↩²

5. SNDK) Update — 2026-04-14 ↩ ↩² ↩³

6. NAND Update — 2026-04-14 ↩ ↩²

7. Irrational Analysis — Memory Madness — 2026-01-16 ↩