MU — Earnings Review Q2 FY26 (Feb-26)

Date: 2026-03-19 | By: wsm007 | Position: Long MU, 10.2%

Verdict

I've been investing for over a decade and I have never — never — seen a quarter like this from a memory company. $23.86B in revenue against an $18.7B guide. 75% gross margins. $12.20 EPS against an $8.42 guide. And then they guide Q3 at $33.5B and 81% gross margins. Eish!! This isn't an earnings beat. This is a paradigm shift being repriced in real time.

Last month I wrote: "68% gross margins. At a memory company. I haven't written that sentence before." Now I'm writing: 75% gross margins. And 81% guided. At Micron. A company that was at -33% gross margins three years ago. The 108-percentage-point margin swing is the most extraordinary financial transformation I've tracked in any sector.

HBM4 qualified for Nvidia Vera Rubin. Five-year Strategic Customer Agreements signed. Supply meeting only 50-67% of demand. This is the structural thesis fully confirmed.

Revenue Trajectory

Revenue tripled YoY. $8.05B → $23.86B in 12 months. That's not growth. That's a phase transition.

EPS trajectory: $1.56 → $1.91 → $3.03 → 4.78 → **12.20** → guided $19.15. That's a 12x increase in 5 quarters. I've never seen anything like this outside of a turnaround from losses to profits — but these are massive absolute numbers, not base effects.

The Beat Pattern — Management Is Sandbagging at Industrial Scale

A $5.16B revenue beat. Let that sink in. That's larger than Micron's entire quarterly revenue was in early FY25. The guidance-to-actual gap is widening, not narrowing → management's visibility window is expanding but they're keeping guide conservative. Record guide that's sandbagged. The Q3 $33.5B guide will be beaten again.

Segment Breakdown — All Four Engines Firing

Three observations:

1. Core Data Center more than doubled QoQ (+139%). Record SSD shipments. NAND — yes, NAND — is inflecting. This is the "third S-curve" I've been watching: after DRAM recovery and HBM ramp, enterprise SSD is now in its own structural upcycle driven by AI inference storage demands.

2. Mobile & Client at 79% gross margin. Higher than Cloud Memory. Why? LPDDR5X/LPDDR6 for AI phones + Windows 10 EOL PC refresh cycle. Content per device is the multiplier — AI phones need 2-3x the DRAM of prior gen.

3. No dead zones. Every segment grew >45% QoQ. Every segment has >62% operating margins. This is not a one-trick HBM story. This is a company-wide pricing power event driven by structural supply constraints across all memory types.

HBM — Key Risk Resolved

The single biggest risk going into this quarter was HBM4 qualification for Nvidia's Vera Rubin platform. Reports were conflicting. Muji flagged it. I flagged it.

Resolved: HBM4 36GB 12-Hi volume shipments are underway, designed for Vera Rubin. HBM4 16-Hi (48GB) has been sampled — that's a 33% capacity increase. HBM4E development on track for CY2027.

Management has stopped breaking out HBM as a separate metric — they said HBM share reached parity with their overall DRAM share. Translation: HBM is no longer a separate product line to them. It's just... their DRAM business now. That's the structural shift in one data point.

Volume vs Price Decomposition — The Critical Question

This quarter's revenue growth is ~90% price, ~10% volume. That's a fact, not a narrative.

The trend is getting more price-heavy, not less:

Revenue tripled YoY but bits shipped barely moved. TrendForce reported Q1 CY2026 conventional DRAM contract prices up 90-95% QoQ, with PC DRAM contracts doubling.

Why volume is flat — structural, not demand weakness:

1. HBM cannibalises DRAM bit supply. HBM uses ~3x wafer area per bit vs standard DRAM. Every wafer converted to HBM removes 3x the bits from the standard DRAM pool. MU achieved HBM share parity with overall DRAM share → significant wafer starts now produce HBM instead of commodity DRAM.

2. Volume is supply-constrained, not demand-constrained. Mehrotra: "with some key customers, we fulfill only 50% to two-thirds of their demand." New capacity (Idaho mid-CY2027, Singapore CY2028) is 12-24 months away.

3. Content per device is the volume multiplier. Flagship phones with 12GB+ DRAM went from 20% to nearly 80% of shipments in one year. Industry bit demand growing 20%+ but absorbed by content increases, not unit growth. PC/smartphone units may actually decline low double-digits because memory supply is constraining device production.

The pricing durability question:

At 6.7x forward P/E, Mr Market is pricing scenario 3. I think we're in scenario 1 for the next 12 months minimum, with scenario 2 the base case for CY2027-2028. The transition to volume-AND-price growth happens when new fabs come online and content per device inflects.

Supply-Demand — The Moat Is Scarcity

From the call: "With some key customers, we fulfill only 50% to two-thirds of their demand" in the medium term. This hasn't changed from Q1 FY26. Demand is outrunning supply expansions.

Pricing: DRAM +65-67% QoQ. NAND +75-79% QoQ. Those are not normal pricing moves. That's supply-constrained panic buying across the industry.

Industry outlook: DRAM bit shipments growing low-twenties percent. NAND ~20%. Supply constraints persist through 2026+. PC and smartphone units may actually decline low double-digits because memory supply limits are constraining device production. The component is dictating system volume. That's pricing power.

Margin Story — The 108-Point Swing

81% gross margins. That's NVIDIA territory. That's higher than most SaaS companies. At a hardware semiconductor company that ships physical silicon.

Operating leverage: OpM expanded from 47% → 69% QoQ. OpEx was $1.4B non-GAAP on $23.86B revenue → 5.9% of revenue. The revenue is scaling while OpEx is essentially fixed. Every incremental dollar of revenue at 75% GM is falling almost entirely to the bottom line.

FCF: 6.9BinQ2(2913.8B FCF in a single quarter.** On a $512B market cap, that's a 10.8% quarterly FCF yield. Annualized, that's over 40%. The cash generation is becoming absurd.

Valuation (at 449after − hours, 512B market cap)

At 6.7x forward P/E with 81% gross margins and $33.5B guided revenue, the market is still pricing this as a cyclical company approaching peak earnings.

Scenario-Based Valuation

The market at $449 is pricing Scenario 2 almost exactly. 449/ 22 normalised EPS = 20x through-cycle P/E. Reasonable for a semi with structural HBM advantages. The opportunity: if Scenario 1 persists for 2-3 more quarters (5-year SCAs, 50-67% demand fulfilment gap, 12+ month capacity lead time), the re-rating to 700−1,000 is mechanical. The risk: Scenario 3 takes this to 150−200.

Why "Flat Bit Shipments" Is Misleading

Revenue growth is ~90% price-driven, but the "bits" metric is not apples-to-apples:

1. Revenue per bit has structurally changed. HBM3E earns ~6-8x the $/GB of standard DDR5. LPDDR5X earns ~2x. "Flat bits" at 6x the revenue per bit is value growth, not stagnation. It's like measuring a refinery's output in barrels when they shifted from crude to jet fuel.

2. HBM consumes 3x the wafer area per bit. TSV stacking, logic base die, yield losses. Converting commodity DRAM to HBM structurally reduces bit output but multiplies revenue per wafer start. Revenue per wafer start has nearly tripled in 12 months — that's the real productivity metric.

3. Node transitions mask volume. 1-gamma DRAM produces ~20-30% more bits per wafer than 1-beta. If MU weren't converting to HBM, bit shipments would be up 20-30% from the node shrink alone. The "flat" reading is the NET of node gains minus HBM conversion.

The right question isn't "are bits growing?" — it's "is the company creating more value per unit of manufacturing capacity?" The answer is unambiguously yes.

CapEx and Capacity — Building for the Decade

• FY26 CapEx: now >$25B (up from prior $20B estimate)
• FY27: "step up meaningfully" — construction spending increasing >$10B YoY
• Idaho Fab 1: First wafers pulled forward to mid-CY27
• New York Fab: Groundbreaking underway, supply 2030+
• Singapore NAND packaging: Targeting CY28
• Tongluo (Taiwan) fab acquisition closed

They're committing $25B+ this year and more next year because they see sustained demand. You don't build $50B+ in fabs for a 2-year cycle. This is the strongest management signal of structural durability.

SBC and Dilution

• SBC: $297M on $23.86B revenue → 1.2%. Negligible.
• Diluted shares: 1,149M (non-GAAP). Flat QoQ. No dilution concern.
• Dividend: increased 30% to $0.15/quarter. Token but directionally right.

Prior Beliefs → Updated Beliefs

Prior belief (Feb 25): Thesis strengthening. Hold 10.2%. Waiting for Q2 FY26 to confirm the $18.7B / 68% GM guide. If confirmed, re-rating opportunity.

Updated belief: Q2 FY26 didn't just confirm. It obliterated. $23.86B vs $18.7B. 75% vs 68% GM. $12.20 vs $8.42 EPS. Then guided $33.5B / 81% / $19.15. The re-rating I was waiting for should be in full motion — but Mr Market gave it a 2.7% haircut after hours. Stock was up 62% going into earnings, so the "sell the news" crowd is taking profits. Fine. The numbers speak for themselves.

Structural thesis: Three S-curves are now stacking simultaneously:

1. HBM — qualified on Vera Rubin, 5-year SCAs, share parity achieved
2. Enterprise SSD — Core Data Center +139% QoQ, record SSD shipments
3. AI content multiplier — LPDDR5X/6 for AI phones, Windows refresh cycle

This is not a single-product cycle peak. It's a multi-vector structural demand story. The 81% GM guide is the tell — management sees no pricing pressure in the next quarter. They're raising CapEx because the demand runway extends to 2028+.

What could go wrong:

• Samsung HBM yields recover → supply normalises → pricing pressure in CY2027+
• Hyperscaler capex pullback → demand cliff for data centre memory
• Macro recession → PC/smartphone replacement cycles stall
• CapEx overcommitment → if demand fades, $25B/year in capex becomes a balance sheet anchor

These are real risks. But they're 12-18 month risks, not next-quarter risks. The near-term setup is as clean as it gets.

Thesis Status: Strengthening → Strong Conviction

Every metric exceeded expectations. Leading indicators (supply constraints, pricing, 5-year SCAs) all confirm durability. HBM4 Vera Rubin risk resolved. Three concurrent growth vectors. The structural case has never been stronger.

Action

Hold 10.2%. Consider adding on any meaningful pullback. At 6.7x forward P/E with these margins and this growth trajectory, Micron is still being priced as if the cycle ends next quarter. It doesn't. The 5-year SCA is the proof.

Watch for:

• Q3 FY26 actuals (will they beat $33.5B? I think $38-40B is achievable)
• Samsung HBM4 qualification progress — timing of competitive supply entering market
• FY27 initial guide — if they guide revenue >$120B, the stock needs to be re-rated as a large-cap grower, not a cyclical
• Hyperscaler capex commentary from MSFT, GOOG, AMZN earnings in April/May

"75% gross margins at a memory company. 81% guided. Five-year customer agreements. HBM4 shipping for Vera Rubin. Supply meeting 50-67% of demand. And the stock went down 2.7% after hours."

The market will catch up. The numbers don't lie.

-wsm (Long MU, 10.2%)