Initiating Coverage at Hold: HBM Crosses $1B and the Q3 Guide is a Record, But Gross Margin is Going the Wrong Direction Near-Term and the NAND Drag Stretches the Recovery Into FY26

Rating Action: Initiating Coverage

This is our initiation report on Micron Technology. We begin coverage at Hold — a patient stance on a name executing exceptionally well on the most strategically important product transition in memory in twenty years (HBM), but where the consolidated P&L still reflects the legacy NAND cycle and consumer-end overhang. Hold is a statement about the gap between the AI narrative and the consolidated margin print, not a critique of operational execution.

Three reservations underpin Hold rather than Outperform: (1) GM is going the wrong direction near-term — 37.9% Q2, 36.5% Q3 guide, Q4 "up somewhat" — the inverse of the trajectory the AI-memory bull case requires; recovery to mid-40s stretches into FY26. (2) NAND is a structural drag — revenue −17% QoQ, prices −high-teens, fabs at mid-teens-% under-utilization; under-absorption costs "weigh on gross margins in fourth quarter and into '26." (3) Inventory is high — DIO 158 days vs. 120-day target; management projects DIO below target by fiscal Q4 with limited margin for any further consumer-end pothole.

Outperform triggers: Q3 GM at/above 36.5% with Q4 guide implying 39%+ on credible bridge to 42%+; HBM trajectory accelerates on additional customer qualifications; CQ2 NAND pricing inflection; meaningful drawdown without thesis impairment. Underperform triggers: further consumer-end demand deterioration; HBM 12-high yield slip; tariff broadening with customer push-back; competitive datapoint from Samsung HBM3E or SK hynix HBM4.

Results vs. Consensus

The print was a beat on EPS, in line on revenue and gross margin within the prior guidance range, and headlined by a milestone HBM number. The Q3 guide was the constructive surprise — revenue and EPS both above Street, with management's record-revenue framing reinforced by bit-shipment growth in both DRAM and NAND.

Quality of Beat

• EPS beat: Driven primarily by lower-than-guided tax rate (10.7% on one-time items). Strip out the tax tailwind and the beat compresses materially.
• Revenue mix: DRAM $6.1B (76%) −4% QoQ with bits down high-single-digits offset by mid-single-digit price increases on portfolio mix — AI-memory mix shift doing exactly what the bull case requires. NAND $1.9B (23%) −17% QoQ with bits up modestly but prices down high-teens — NAND industry is mid-correction.
• HBM milestone: The $1B+ quarter is an inflection — HBM3E was sub-$200M two quarters earlier. Now ~12-13% of DRAM revenue and the structurally highest-margin product in the portfolio. 12-high volume production initiated this quarter.
• Gross margin: 37.9% at the low end of the prior guidance range (38.5% ±100bps). The trajectory of the next print (36.5% midpoint) is the more important signal.
• Cash flow: $3.94B operating cash flow, $3.1B net CapEx, $857M FCF. $14B FY25 CapEx unchanged — heavy-investment cycle (HBM, Idaho, Singapore, Taiwan); FCF generation is back-end-loaded.

Segment Performance

CNBU: The Whole AI Story

Compute and Networking revenue of $4.6B (+4% QoQ, third-consecutive quarterly record, ~57% of total) is the segment that defines the equity case. HBM grew >50% QoQ to cross $1B for the first time; combined high-capacity DIMM + data-center LPDRAM separately exceeded $1B; data-center DRAM hit a new all-time record. This segment is firing on every cylinder and justifies the Hold-with-constructive-bias.

"In fiscal Q2, data center DRAM revenue reached a new record. HBM revenue grew more than 50% sequentially to a new milestone of over $1 billion of quarterly revenue. Our HBM shipments were ahead of our plans." — Sanjay Mehrotra, CEO

HBM positioning. Three product datapoints: (1) HBM3E 8-high in NVIDIA GB200; (2) HBM3E 12-high entered volume production this quarter, designed into GB300; (3) volume shipments to a third large HBM3E customer began in Q2. Management frames HBM3E with a 30% power reduction vs. competition; 12-high with a 20% power advantage over competing 8-high and 50% higher memory capacity.

TAM and share. CY25 HBM industry TAM raised to ">$35B." CY25 output sold out; CY26 negotiations underway. CQ4 2025 share target is "similar to overall DRAM supply share" (low-20%s) — a multi-billion-dollar revenue unlock. "Multi-billion dollars in HBM revenue in fiscal 2025" is now the framing. HBM4 ramps in CY26 with >60% bandwidth gain vs HBM3E; trade ratio >4:1 with HBM4E (vs. 3:1 today), structurally tightening leading-edge DRAM supply for non-HBM products through CY26-27.

SOCAMM was co-developed with NVIDIA for GB300, extending Micron's data-center LPDRAM leadership as the form factor transitions from soldered to module. Micron remains the only company shipping LPDRAM into the data center in high volume.

Assessment: CNBU is the structural compounder. As a standalone segment, this is comfortably Outperform; dilution by the other three business units is what tempers the consolidated rating.

SBU: Data-Center NAND Inventory Adjustment, Consumer Pricing Pressure

Storage revenue of $1.4B (−20% QoQ) reflected the worst sequential print in the segment lineup. The decline was driven primarily by lower storage investments from data-center customers after several quarters of strong growth and overall NAND industry pricing weakness. The data-center NAND demand pause is framed as inventory-driven and short-term — Sanjay confirmed "a return to bit shipment growth in the months ahead." Strategic positioning is intact: record data-center SSD share in CQ4 2024, the 9550 SSD on the GB200 NVL72 approved vendor list, and G8 QLC components qualified for Pure Storage's 150TB DirectFlash module supporting the HDD-to-NAND data-center transition. Assessment: SBU is the segment where we want to see the bit-shipment recovery materialize in Q3 actuals.

MBU: Inventory Normalization, AI-Phone Tailwind Building

Mobile revenue of $1.1B (−30% QoQ) was the largest sequential decliner. Customers continue to improve inventory positions; the smartphone-content cycle is transitioning from inventory work-down to content-driven growth. AI-capable flagship phones feature 12GB+ DRAM (vs. 8GB last year). LP5X DRAM and UFS 4.0 NAND ship in the Samsung Galaxy S25 high-end. The 1TB UFS 4.1 sample is an industry first. Mobile DRAM and NAND bit shipments grow in fiscal Q3. Assessment: Cleaner sequential setup than the QoQ print suggests.

EBU: Late-Stage Inventory Adjustment

Embedded revenue of $1.0B (−3% QoQ) was the most resilient of the three non-CNBU segments. Auto OEMs are "in the later stages of adjusting inventory." Robotaxi platforms now contain "over 200GB of DRAM, or 20-30x higher than the average car." Micron's automotive LP5X DRAM (9.6Gbps) and the 4150 SSD (first automotive-qualified enterprise SSD) are now sampling. Small and steady; not a print-mover.

Key Topics & Management Commentary

Overall Management Tone: Confident on the HBM ramp and leading-edge DRAM technology position; matter-of-fact on consolidated margin pressure; clinical on NAND supply discipline. Sanjay led with "Micron is in the best competitive position in our history" — supported by HBM milestone, 1-gamma EUV launch, SOCAMM with NVIDIA, Singapore packaging groundbreak, Idaho fab CHIPS disbursement. Mark Murphy bounded the GM trajectory (Q4 "up somewhat") without committing to levels — right posture given the under-absorption flow-through is genuinely uncertain.

The HBM Ramp: Inflection Achieved, Now It Has To Scale

HBM revenue crossing $1B and 12-high entering volume production are the inflection points the company has telegraphed for several quarters. The 50%+ sequential growth and "ahead of our plans" framing are stronger than prior commentary.

"We have begun volume production of HBM3E 12-high and are focused on ramping capacity and yield. We anticipate HBM3E 12-high will comprise the vast majority of our HBM shipments in the second half of calendar 2025." — Sanjay Mehrotra, CEO

On the 12-high yield curve: Sanjay framed 12-high as carrying a premium over 8-high and remaining accretive to DRAM margins. Consistent with prior cycle product transitions but would benefit from more granular yield datapoints. Assessment: The questions we want answered over the next two prints: (i) is the third large HBM customer ramp accelerating, (ii) what is HBM4 customer-allocation visibility, (iii) does CQ4 2025 HBM share actually achieve overall-DRAM-share parity.

The Gross Margin Conversation: The Most Important Exchange of the Call

Chris Danely (Citi) asked the analytically pointed question: at the same revenue base as the prior cycle, why are gross margins ~10 points lower? Sanjay's answer — DRAM healthy, NAND drag, focus on high-value mix — is directionally true but does not fully resolve the question. The implicit answer is depreciation: the $14B FY25 CapEx envelope is materially higher than at the prior cycle-equivalent revenue point, and depreciation flows through COGS over time.

"Overall, our gross margins in DRAM have been healthy... NAND is what has weighed down on our margins." — Sanjay Mehrotra, CEO — in response to Chris Danely, Citi

Bull-case rebuttal: HBM scaling + trade-ratio tightness drive DRAM-mix toward high-margin HBM/HC-DIMM/LPDRAM; NAND pricing inflects on discipline; consolidated GM marches to mid-40s through FY26. Bear-case framing: depreciation has re-based COGS, FY25-26 CapEx wave re-bases again, cycle peak GM ~5 points below prior cycle. Both have merit; data over the next 2-3 prints resolves which framing is closer to right. Assessment: GM trajectory is the single biggest swing factor in the rating. Q3 actual GM at/above 36.5% with Q4 guide implying 39%+ would warrant an Outperform upgrade on margin progression alone.

NAND Supply Discipline: The Right Strategy, Bounded Cost Today

Micron is running NAND fabs at meaningfully underutilized levels (wafer output down mid-teens% from prior) and reusing a portion of underutilized NAND equipment for capital-efficient conversions to leading-edge nodes. The structural reduction is "over 10%" of NAND wafer capacity exiting FY25 vs. FY24. Management's explicit CQ2 calendar pricing-action framing positions Micron to lead an industry pricing inflection.

The cost is non-trivial. Per Mark Murphy, less period-cost drag in Q3 because capacity has been structurally reduced — but the under-absorption charge flows into inventory and weighs on gross margins in fiscal Q4 and into FY26. Assessment: The right strategic call. Upside payoff requires industry coordination that has historically proven fragile; we treat it as a real but probabilistic call option.

Tariff Exposure and Industry Bit-Demand Outlook

Sanjay framed tariff exposure as "very limited volume" today (Micron is U.S. importer of record on a small slice of the product mix) with cost pass-through to customers as the operational posture. The unanswered question is regime-broadening exposure — we treat tariff as a live but unquantified tail-risk into 2H 2025 rather than an immediate-print issue.

Industry bit-demand: CY25 DRAM raised to mid- to high-teens (from mid-teens prior); NAND held at low double-digits. CY24 NAND bit growth came in at ~10%, slightly below prior low-double-digit view — the underlying reason the NAND industry is mid-correction. Medium-term mid-teens CAGR for both. Micron's CY25 supply growth lower than industry demand for both, which is what enables the DRAM DIO tightening through fiscal Q4.

Guidance & Outlook

The Q3 guide is the constructive setup. $8.8B revenue midpoint is +9% QoQ, +~36% YoY, a new all-time record — ~$300M above Street consensus and EPS ~$0.10 above. Clean above-Street guide despite the further GM step-down.

The 36.5% GM guide (−140bps from Q2 actual) is the soft spot. Drivers: higher consumer-oriented volume mix, lower CQ1 consumer-end pricing, NAND under-utilization, partially offset by HBM growth and high-value DRAM mix. Q4 "up somewhat" suggests 38-40% zone with continued under-absorption headwind. Path to 42%+ GM stretches to fiscal Q1 FY26 in our base case.

FY25 implied: H1 FY25 revenue of $17.4B (Q1 $8.71B + Q2 $8.05B) plus Q3 guide of $8.8B places 9-month revenue at $26.2B. If Q4 lands $9.2-9.6B, full-year FY25 tracks toward $35.4-35.8B, +~33-34% YoY. EPS central tendency $6.80-7.20 (bull $7.50+, bear $6.50).

Capital structure: $9.6B cash, $12.1B liquidity, $14.4B total debt at quarter end. Post-quarter, revolver upsized to $3.5B. The $1B 10-year senior note + $1.7B term loan extended weighted-average maturity to 2032. Healthy balance-sheet posture for a heavy-CapEx fiscal year. No capital-return announcement — consistent with heavy-investment-cycle framing.

Analyst Q&A Highlights

Gross Margin Trajectory and Cost Curve

• Harlan Sur, JPMorgan: Asked whether the GM step-down through Q3 reverses in Q4. Mark Murphy framed Q4 GM "up somewhat" with tailwinds (market improvement, HBM/high-value mix) and headwinds (NAND under-absorption flowing through inventory, construction start-up costs, new-node DRAM start-up).
  Assessment: We read this as 38-40% GM zone for Q4, with bridge to mid-40s extending into fiscal H1 FY26.
• Tim Arcuri, UBS: Asked when "true goodness" from HBM and DRAM cost-downs flows through. Mark noted FY25 all-in DRAM costs flattish; FY25 NAND all-in costs improve in line with front-end cost reductions in the low double-digits.
  Assessment: HBM mix becomes the GM lever rather than commodity DRAM cost reductions; NAND cost-reduction helpful but not transformative.
• CJ Muse, Cantor Fitzgerald: Asked about under-absorption flow-through to fiscal H2 FY25 and FY26. Mark explained Q3 period costs were lower than projected (capacity structurally reduced), but the under-absorption charge into inventory weighs on Q4 GM as inventory clears and continues into FY26.
  Assessment: Most analytically important answer of the call. Under-absorption-as-inventory-charge is mathematically deterministic, not a function of external pricing. Bounds our Q4 GM expectation to 38-40%.

HBM Capacity, Customer Base, and Yield

• Krish Sankar, TD Cowen: Asked whether HBM3E 12-high yields are lower than 8-high. Sanjay framed 8-high as ahead of plan, 12-high in early-life yield-ramp but expected to carry a premium over 8-high and remain accretive to DRAM margins.
  Assessment: 12-high yield trajectory is a principal swing factor for FY25 H2 HBM revenue. More granular yield disclosure would help.
• Chris Caso, Wolfe Research: Asked about HBM TAM upside in CY26 given CY25 sold-out position. Sanjay framed CY26 share as higher than CY25 given the Q4-2025 share-parity exit run-rate and capacity expansion through 2026.
  Assessment: The CY26 HBM share dynamic is the next leg of the bull case.

Inventory and Long-Term Margin

• Joe Moore, Morgan Stanley: Asked about the path from 158 days DIO down to the 120-day target. Mark confirmed conditions are tighter in DRAM than NAND; DRAM DIO falls below 120 days by fiscal Q4.
  Assessment: DRAM DIO target achievable if Q3-Q4 bit shipments materialize as guided. NAND DIO trajectory is the looser piece.
• Chris Danely, Citi: Asked the most analytically pointed question of the call: why are gross margins ~10 points lower at the same revenue base as the prior cycle, and is mid-50% GM ever recoverable? Sanjay framed DRAM as healthy and NAND as the drag, with focus on high-value mix.
  Assessment: Not fully answered. Implicit answer is depreciation step-up from the FY24-25 CapEx wave. We model peak-cycle GM at ~46-48% (vs. prior-cycle 55%+) — a structural re-base of ~5-7 points is plausibly the new normal.

What They're NOT Saying

1. HBM customer concentration. Three large HBM customers referenced (NVIDIA explicit on GB200/GB300; third customer began volume shipments this quarter); mix not disclosed. Our prior: NVIDIA is the substantial majority, with AMD MI300/MI350 and one other large HBM3E customer splitting the rest.
2. HBM gross margin. Framed as "DRAM-margin accretive" but never disclosed. Street prior is 50%+ — meaningfully above the 37.9% consolidated print — but the mix-shift math requires this to be true at scale.
3. Q4 GM specific. "Up somewhat" framing implies 38-40% zone. Bull (40%+) and bear (37-38%) both inside the framing; company is conscious of not committing.
4. Tariff dollar exposure. "Limited volume" framing. Dollar impact of Canada/Mexico/China tariff actions not quantified; future-tariff exposure not bounded.
5. The prior-cycle GM gap. The depreciation-step-up explanation for the ~10-point GM gap to prior cycle is the missing piece; management is not volunteering it. We read this as conscious — implicit message is that mid-50s GM may not be the right benchmark for the new CapEx-base era.

Market Reaction

• After-hours move (March 20 evening): MU traded modestly higher initially on the EPS beat and above-Street Q3 guide; the HBM milestone narrative was the headline pickup. The reaction faded as analysts processed the consolidated GM trajectory (37.9% Q2 actual, 36.5% Q3 guide, "up somewhat" Q4 framing).
• Next-day intraday context: March 21 trading reflected the dual narrative — bulls anchoring on the HBM revenue inflection and record Q3 revenue guide, bears on the GM trajectory and NAND weakness. Late-March 2025 macro backdrop (tariff-regime uncertainty, semiconductor-tape volatility) added cyclical pressure beyond the print specifics. Tape behavior was net cautious.
• Pre-print context: MU had been a high-beta name through fiscal Q1 and the run-up incorporated meaningful HBM-narrative optimism. Expectations on consolidated GM were modestly above where the print and Q3 guide landed — the principal driver of the muted reaction.

The reaction is rational, in our view. The print delivered exactly what the AI-memory thesis required on the HBM line, but the consolidated GM trajectory is the gating metric for multiple expansion, and that trajectory disappointed at the margin. The next two prints (Q3 in mid-June 2025, Q4 in late September 2025) will determine whether the GM bridge to mid-40s is credible.

Street Perspective

Debate: Is the gross-margin trajectory a one-cycle issue or a structural re-base?

Bull view: 37.9% Q2 and 36.5% Q3 guide are the cycle trough. Drivers are non-secular (consumer mix, NAND under-absorption, CQ1 consumer pricing). HBM scaling, leading-edge DRAM trade-ratio tightness, and CQ2 NAND pricing inflection drive a sharp recovery into fiscal Q1 FY26 with cycle-peak GM in the high-40s / low-50s through CY26.

Bear view: The GM gap to prior cycle peak is a structural depreciation re-base from the FY24-25 CapEx wave; FY26 Idaho/Singapore CapEx adds further fixed-cost burden. Cycle-peak GM is mid-40s rather than high-40s, and the path stretches through FY26-27.

Our take: Bear view closer to right on the structural depreciation point; bull view closer to right on near-term recovery. We model FY25 GM high-30s, FY26 low- to mid-40s, peak-cycle mid- to high-40s rather than 50%+. A structural re-base of ~5-7 points of GM is the cost of the AI-memory CapEx wave.

Debate: How much of the HBM TAM does Micron actually capture?

Bull view: Micron is the technology leader on power efficiency (30% advantage HBM3E vs. competition; 20% advantage on 12-high vs. competing 8-high). NVIDIA GB200/GB300 design wins are anchor positions; the third large customer adds diversification. CY25 share parity is achievable, CY26 share gains the next leg.

Bear view: Samsung is aggressively requalifying HBM3E with NVIDIA; SK hynix retains dominant share and is ramping HBM4 on schedule. CY26 customer-allocation visibility is lower than the bull case implies.

Our take: Technology leadership on power efficiency is real and customer-validated; CY25 share parity exit run-rate is achievable. We model Micron at 20-23% HBM share through CY26 — consistent with overall DRAM share but below the bull-case 25%+.

Debate: Does NAND supply discipline actually deliver pricing inflection?

Bull view: Industry NAND wafer capacity is structurally reduced; CQ2 calendar pricing inflection is the management-guided event; NAND margin recovery through 2H CY25 unlocks consolidated GM expansion.

Bear view: NAND supply discipline has historically been fragile; Samsung has periodically broken discipline to defend share. CY25 NAND demand may not absorb even reduced supply unless the data-center pause resolves quickly.

Our take: The historical record on industry coordination is poor. We model a modest NAND pricing recovery through CY25, with the upside bounded. NAND will not be the primary driver of the GM recovery; the recovery will be DRAM-mix-led.

Model Implications

Valuation framing: At a low-double-digit forward P/E on FY26 — consistent with memory cycle-peak multiples historically — implied fair value lands roughly in line with current market price. The asymmetry is in the GM trajectory: faster-than-expected return to mid-40s GM through FY26 unlocks meaningful EPS upside; a depreciation-driven structural re-base caps EPS upside even with revenue compounding. Textbook Initiating-at-Hold — high-quality franchise, reasonable price, margin trajectory not yet validated.

Asymmetric scenarios that change our rating: (i) Q3 GM at/above 36.5% with Q4 guide implying 39%+ → Outperform pull; (ii) HBM customer-base expansion to a fourth large customer → Outperform pull; (iii) NAND pricing breakdown with no CQ2 inflection → Underperform; (iv) tariff-regime broadening with customer push-back → Underperform.

Thesis Scorecard at Initiation

Overall: Five bull points confirmed, six bear points live or suspected. The central tension is that the AI-memory operational story is on schedule with clean datapoints, but the consolidated P&L still reflects the legacy NAND cycle, consumer-end inventory work-down, and a depreciation base that stepped up materially with the FY24-25 CapEx wave. Hold reflects the gap between the AI narrative and the consolidated margin print — a gap we believe closes, but on a multi-quarter rather than next-quarter timeline.

Action: Initiating at Hold (constructive operational bias). Upgrade triggers: Q3 GM at/above 36.5% with Q4 guide implying 39%+ on a credible bridge to 42%+ by end of FY25; HBM customer-base expansion to a fourth large customer; CQ2 NAND pricing inflection; drawdown of 15-20% without thesis impairment. Downgrade triggers: Q3 GM below 36.5% with Q4 framing implying further compression; HBM 12-high yield issues; tariff-regime broadening with customer push-back; NAND supply-discipline breakdown. We revisit on the Q3 FY25 print in mid-June 2025.