Upgrading to Outperform: The Margin Inflection Arrives One Quarter Early, the FQ4 Guide Clears the Street by ~$800M and ~250bps, and Every Hold-Gating Concern Resolves the Right Way

Rating Action: Upgrading from Hold to Outperform

This is the second report in our MU coverage arc. We are upgrading to Outperform on the back of the Q3 FY2025 print and FQ4 guide. The arc:

• Q2 FY2025 (Initiating at Hold, constructive bias) — March 21, 2025. We opened coverage at Hold on three concerns: (1) consolidated GM compression — 37.9% Q2 actual, 36.5% Q3 guide, "up somewhat" Q4 framing; (2) NAND under-absorption flowing through inventory and weighing on Q4 GM "and into '26"; (3) DIO at 158 days vs. the 120-day target with limited margin if consumer-end demand softened. The HBM ramp was real and the leading-edge DRAM positioning was industry-best, but the consolidated P&L still reflected the legacy NAND cycle.
• Q3 FY2025 (Today — Upgrading to Outperform). Every Hold-gating concern resolved the right direction. GM expanded 110bps QoQ to 39.0% — 250bps above the 36.5% guide and the inverse of the trajectory we modeled. The FQ4 guide of 42% GM (with revenue $10.7B and EPS $2.50, all comfortably above Street) demonstrates that mix and pricing have overrun the NAND under-absorption charge. DIO compressed 19 days to 139, with management explicit on exit-FY25 DIO "near our target levels." HBM ramp is ahead of plan: ~$6B run-rate, 12-high crossover pulled into FQ4, fourth customer added. The bridge to mid-40s GM is now visible in the FQ4 print itself rather than stretching through FY26.

The case for Outperform: (1) GM trajectory has inflected — 37.9% Q2 → 39.0% Q3 actual → 42% Q4 guide is a 410bps two-quarter expansion against a Q2-call "up somewhat" framing that implied 38-40% zone for Q4. The structural depreciation re-base from the FY24-25 CapEx wave is real but bounded; mix and pricing do most of the work. (2) The HBM ramp continues to compound — ~$6B run-rate, 12-high crossover pulled forward, AMD MI355X design win, four large customers, CY26 bit-demand growth "significantly exceeding" DRAM industry. (3) NAND inventory and pricing have stabilized — DIO down 19 days, structural 10% wafer-capacity reduction on track, leading-edge NAND fully utilized. (4) Inventory at 139 days is near target, with tight DRAM and reduced NAND inventories framing into FQ1 FY26.

Upgrade-to-Outperform triggers met from the Q2 framing: "Q3 GM at/above 36.5% with Q4 guide implying 39%+ on credible bridge to 42%+" — Q3 actual was 39%, Q4 guide is 42%, both well above. "HBM customer-base expansion to a fourth large customer" — explicit on the call: "shipping HBM in high volume to 4 customers." "CQ2 NAND pricing inflection" — NAND bit shipments up mid-20%s with prices down high-single-digits (sequential), but combined with the bit-shipment surge and consolidated GM expansion, the gating-concern is bounded. Three of four conditions met cleanly; we treat that as sufficient for the upgrade.

Downside triggers from Outperform: HBM 12-high yield slip on the planned FQ4 crossover; tariff-regime broadening with customer push-back; competitive datapoint from Samsung HBM3E or SK hynix HBM4 ramp accelerating faster than modeled; CapEx step-up beyond ~$14B FY25 footprint that re-bases depreciation faster than mix-led GM expansion can absorb.

Results vs. Consensus

The print is a beat on every line that matters — revenue, GM, EPS, segment composition, cash flow, inventory, and the forward guide. The defining datapoint is the GM swing: 39% actual against 36.5% guide is a 250bps upside surprise, with HBM mix and DRAM pricing better than expected as the underlying drivers.

Quality of Beat

• The GM beat is the headline. 110bps QoQ expansion against a guide of -140bps is a 250bps upside surprise. Mark Murphy attributed it to "better-than-expected pricing" and "good cost performance," with mix shift to DRAM and away from consumer aiding through the back half of the quarter. Critically, this is not a tax-rate-driven beat — the 12.3% effective tax rate was lower than the ~14% guide on one-time discrete items, but even at a normalized rate the EPS would have cleared the Street comfortably.
• Revenue mix is exactly what the bull case requires. DRAM $7.1B (+15% QoQ) on bit shipments >+20% with prices down low-single-digits — the bit-shipment volume more than absorbed the consumer-mix price weakness. NAND $2.2B (+16% QoQ) on bit shipments mid-20%s with prices down high-single-digits — the bit-volume recovery confirms the data-center NAND inventory pause from Q2 has resolved.
• HBM is ahead of plan. The ~50% sequential growth came on top of the >$1B Q2 print, putting the run-rate at ~$6B+. Sanjay's framing that the share-parity goal "could be" achieved earlier than the 2H CY25 timeline is the cleanest indication yet that Micron is winning the HBM allocation conversation. The fourth customer ramp (vs. three at Q2) is real.
• Cash flow is the cycle-best print. $4.6B operating cash flow, $2.7B CapEx, $1.9B FCF — the highest quarterly FCF in over six years. $12.2B cash + investments at quarter end. The balance-sheet posture for the FY25-26 CapEx wave is in pristine shape; the $1.7B refinancing extended weighted-average debt maturity to 2032 with net leverage at $3B.
• Inventory cleanup is decisive. DIO compressed 19 days to 139 — the largest single-quarter improvement in the DIO series since the cycle began. Management's exit-FY25 framing ("DIO near our target levels") implies the 120-day target gets hit in FQ4. This was the single biggest risk into the Q2 initiation; resolved.

Segment Performance

Note on reporting structure: management completed a strategic reorganization of business units in June, and starting FQ4 will report revenue, GM, and operating margin across the new market-segment-focused units. Q3 was reported under the legacy CNBU/SBU/MBU/EBU structure for the last time. The arc-walk against Q2 is therefore clean; FQ4 will introduce a presentation discontinuity.

CNBU: HBM, HC DIMMs, and LP Server DRAM — Three Records in One Segment

Compute and Networking revenue of $5.1B (+11% QoQ, quarterly record, ~55% of total) is again the segment that defines the equity case. HBM grew ~50% sequentially; high-capacity DIMM + LP server DRAM combined delivered another quarterly record. Data-center DRAM revenue reached a new record for the fourth consecutive quarter. Micron remains the sole supplier in volume production of LPDRAM in the data center.

"Data center revenue more than doubled year-over-year and reached a record level... DRAM revenue reached a new record driven by a nearly 50% sequential growth in HBM revenue. We remain the sole supplier in volume production of LPDRAM in the data center." — Sanjay Mehrotra, CEO

HBM datapoints from the call. (1) Run-rate at $6B+ on the FQ3 print pace (Sanjay confirmed in response to CJ Muse). (2) HBM3E 12-high yield ramp "going faster than 8-high yield ramp" — built on the learnings from the prior generation. (3) 12-high "shipment crossover" — meaning 12-high becomes the majority of HBM shipments — expected in FQ4, ahead of the prior 2H CY25 framing. (4) Now shipping HBM in high volume to four customers (up from three at Q2), spanning both GPU and ASIC platforms. (5) HBM3E 36GB 12-high designed into AMD's Instinct MI355X — announced at AMD's Advancing AI event during the quarter. (6) CY25 HBM industry revenue tracking $35B vs. ~$18B in CY24, on Sanjay's framing.

HBM4 visibility. Sampled to multiple customers; volume production in CY26. Built on the well-proven 1-beta DRAM technology with internally developed and manufactured advanced CMOS logic base die — bandwidth >2 TB/s per stack (>60% higher than HBM3E), 20% lower power than HBM3E 12-high. Trade ratio >3 (vs. 3 for HBM3E); HBM4E will go higher again "toward 4." This is the structural lever for leading-edge DRAM tightness through CY26-27.

Assessment: CNBU is comfortably Outperform on a standalone basis. The dilution from other segments — the Q2 framing's basis for Hold — is now actively narrowing as MBU and EBU recover.

SBU: Bit-Shipment Recovery Is the Q3 Datapoint We Wanted

Storage revenue of $1.5B (+4% QoQ) confirms the recovery we flagged into Q3 in our initiation. The data-center NAND inventory pause from late CQ4 2024 / early CQ1 2025 has resolved; consumer-oriented revenue growth led the sequential recovery. Critically, Micron achieved a record-high market share for data-center SSDs and client SSDs in CQ1 per third-party data, and for the first time ever became the #2 brand by share in data-center SSDs in CQ1. The 9550 SSD on the GB200 NVL72 recommended-vendor list completed additional OEM qualifications; the 6550 ION 60TB capacity SSD continues to ramp. Assessment: SBU is no longer a drag on the consolidated rating — bit shipments mid-20%s sequential growth in NAND is a clean beat against the Q2 worry list.

MBU: The +45% Sequential Print Is the Standout

Mobile revenue of $1.6B (+45% QoQ) is the segment that turned hardest. Customer inventory normalization is complete; DRAM content growth is the driver. Smartphone units expected to grow low-single-digits in CY25 with AI-enabled phones and Windows-11-cycle PCs as the structural tailwinds. 1-gamma LP5X DRAM — first qualification samples shipped in Q3 — is positioned for 2026 flagship smartphones with 25%+ faster recommendation performance and 20% lower power. UFS 4 NAND on G9 ramped to high-volume production with a key customer design win; seven smartphone OEMs awarded Micron top quality awards during the quarter. Assessment: MBU went from -30% in Q2 to +45% in Q3 — one of the cleanest sequential turns in the equity story.

EBU: Industrial Resumption Plus D4/LP4 Allocation Pricing

Embedded revenue of $1.2B (+20% QoQ) reflects two distinct drivers. (1) Industrial: customers increasing investment for AI adoption, including in factory automation, with Micron driving price improvements against constrained D4/LP4 supply and low distributor channel inventory. (2) D4/LP4 EOL: Micron sent EOL notices several months ago to high-volume segments with final shipments occurring in 2-3 quarters; high-volume customers are starting to see "increasing shortages of D4 products" and Micron is "now on allocation." Auto/industrial/defense longevity customers continue to be supplied with 1-alpha D4/LP4 for several years. Assessment: EBU is now contributing positively to the consolidated mix narrative rather than serving as a small steady drag.

Key Topics & Management Commentary

Overall management tone: Confident on every dimension — HBM execution, GM trajectory, capacity ramp, customer relationships, capital deployment. The tone is materially more assertive than three months ago. Sanjay's lead framing — "record revenue in fiscal Q3 with revenue, gross margin and EPS all exceeding the high end of our guidance ranges" — is precisely the formulation the Q2 framing had set as the trigger for the upgrade. Mark Murphy's GM commentary moved from "up somewhat" hedging to explicit Q4 framing of 42% with "favorable mix Q3 to Q4" and a constructive forward read.

The HBM Ramp: Crossover Pulled Into FQ4, Fourth Customer Added, Share Parity Earlier

The Q2 worry on HBM was 12-high yield in early-life ramp. Three months later, the yield curve is "going faster than 8-high yield ramp." The crossover datapoint — when 12-high becomes the majority of HBM shipments — was framed at Q2 as 2H CY25; now expected in FQ4 (June-Aug 2025). Share-parity-with-overall-DRAM "could be" achieved earlier than the prior end-of-CY25 framing.

"Our yield and volume ramp on HBM3E 12-high is progressing extremely well, and we expect shipment crossover in FQ4. We expect to reach HBM shares similar to our overall DRAM share sometime in the second half of calendar 2025." — Sanjay Mehrotra, CEO

"At AMD's Advancing AI event earlier this month, we announced that Micron's HBM3E 36-gigabyte 12-high has been designed into AMD's Instinct MI355X GPU platform. We are now shipping HBM in high volume to 4 customers, spanning both GPU and ASIC platforms." — Sanjay Mehrotra, CEO

HBM4 specification leadership. Bandwidth >2 TB/s per stack — 60% higher than HBM3E. 20% lower power than HBM3E 12-high. Internally developed and manufactured advanced CMOS logic base die — a structural advantage given Micron's vertical integration on the logic-die portion. Sampled to multiple customers; volume production in CY26 aligned with customer platforms. The HBM4 trade ratio >3 (vs. 3 for HBM3E) means a given silicon footprint produces fewer HBM bits relative to non-HBM DRAM — the structural lever for non-HBM DRAM tightness.

HBM TAM. Sanjay framed CY25 HBM industry revenue at ~$35B (vs. ~$18B CY24). For CY26, HBM bit-demand growth "will significantly exceed the overall DRAM industry demand growth," with the 8-high to 12-high transition and HBM4 introduction both adding value per bit. This frames CY26 HBM revenue meaningfully above $35B even before share-gain math. Assessment: The HBM thesis is strengthening on every axis — technology, capacity, customer base, and TAM.

The Gross Margin Conversation: From "Up Somewhat" to 42% in One Quarter

The Q2 call's most analytically important exchange was the Citi question on the prior-cycle GM gap. Three months later, that question has been substantially overrun by the print: 39% Q3 actual + 42% Q4 guide, with FQ1 FY26 GM "can be up" per Mark Murphy. The depreciation re-base concern is bounded; mix and pricing do most of the lifting.

"We had strong sequential volume growth in the quarter. But versus the guide, the defining factor was we had — prices were down, as I said in the opening comments, but we had better-than-expected pricing that drove the margin improvement versus expectations. I'll add, too, that we had good cost performance and all. But certainly, the price was better than we had expected." — Mark Murphy, CFO

"In Q3 to Q4, we have favorable mix effects that is helping drive margins up to the current guide of 42%... The market environment does remain constructive, particularly in DRAM versus NAND. Again, we're focused on pricing and making sure to put our bets in the right places... our inventories are very tight, particularly on leading edge and then now even pockets of some of the legacy and DRAM. So we're going to have some bit constraints going into first quarter, but we're going to mix to DRAM, higher-value DRAM and higher-value NAND products. And so we think gross margins can be up." — Mark Murphy, CFO

The bridge from 39% Q3 actual to 42% Q4 guide is constructed from: (a) more DRAM growth than NAND in the FQ4 mix, (b) more data-center than consumer-oriented mix (vs. the consumer-mix headwind that weighed on Q3), (c) continued cost improvements on leading-edge nodes, (d) HBM scaling as a %-of-DRAM. All four are mechanically supportable in the data we have. Assessment: The GM trajectory we modeled into Hold three months ago (mid-40s in mid-FY26) gets pulled forward by 2-3 quarters. We now expect FY25 exit GM in the low-40s and FY26 cycle-peak in the mid- to high-40s, vs. the prior view of mid-40s.

NAND: Discipline Is Working; Bit-Shipment Recovery Confirms the Q2 Framing

NAND revenue of $2.2B (+16% QoQ) on bit shipments mid-20%s sequential growth is the clean confirmation of the Q2 supply-discipline framing. Structural NAND wafer capacity reduction of 10% exits FY25 vs. end-FY24 as planned. Leading-edge NAND fully utilized; legacy nodes still partially under-utilized but actively being converted. G9 QLC NAND ramping at a pace consistent with demand; record-high QLC mix in Q3.

The under-absorption flow-through that bounded our Q2 Q4 GM expectation to 38-40% has been overrun by mix and pricing. Mark Murphy's FQ1 FY26 framing — "gross margins can be up" — with "tight DRAM inventories" and "significantly reduced NAND inventories" exiting FY25 implies the structural drag is materially bounded. Assessment: NAND is no longer the gating concern. Recovery is in line with Q2 framing.

Capital Deployment: $200B U.S. Investment and the FY25 CapEx Cadence

Two weeks before the call, Micron announced plans to invest approximately $200B in the U.S. over 20+ years — $150B in manufacturing + $50B in R&D — with $30B incremental beyond previously announced plans. Components: (1) second leading-edge fab in Boise (ID2); (2) expansion and modernization of the Manassas, VA fab serving auto/aerospace/defense/industrial; (3) advanced packaging capabilities in the U.S. to support long-term HBM growth (after sufficient DRAM wafer scale is established). ID1 first DRAM wafer output expected 2H CY27. ID2 ramps before the first New York fab.

FY25 CapEx unchanged at ~$14B with the overwhelming majority supporting HBM, facility construction, back-end manufacturing, and R&D. FQ3 CapEx of $2.7B was lighter than expected; FQ4 step-up implied to hit the $14B target. FY26 CapEx framing not yet provided — the call asked about the 35%-of-revenue prior reference; Mark declined to commit. The $200B U.S. plan is a long-arc build with grants, construction, and equipment install lumpiness; FY26 CapEx envelope should be one of the FQ4 datapoints to watch.

Tariff Exposure: Modest Pull-In, Limited Volume Impact, Still Unquantified Going Forward

Sanjay framed customer inventory levels as "healthy overall across end markets" with "some tariff-related pull-ins by certain customers" the impact of which is "relatively modest." The constructive demand framing for the remainder of CY25 remains intact. The unquantified piece — future tariff regime broadening — is still a tail risk into 2H 2025, but no longer a near-term print risk. Assessment: Tariff exposure remains a live but bounded overhang.

1-Gamma DRAM: First EUV Node Yielding Ahead of Plan

1-gamma DRAM yields are ramping ahead of the record pace achieved on 1-beta. First qualification sample shipments of 1-gamma LP5 DRAM completed during the quarter. The node provides 30% bit-density improvement, >20% lower power, and up to 15% higher performance vs. 1-beta. 1-gamma will leverage across the entire DRAM portfolio. Assessment: Technology positioning is strengthening; 1-gamma yield trajectory is a leading indicator of FY26-27 DRAM cost curve.

Guidance & Outlook

The FQ4 guide is the upgrade-anchoring datapoint. Revenue $10.7B at midpoint is +15% sequential, ~+38% YoY, a new all-time record — ~$800M above the ~$9.9B Street midpoint. GM 42% is ~250bps above Street and ~300bps above the Q3 actual. EPS $2.50 is ~$0.40 above Street even with tax normalization. The bridge: more DRAM than NAND in the mix; more data-center than consumer; continued cost improvements; HBM scaling.

FY25 implied: H1 actual $16.8B (Q1 $8.71B + Q2 $8.05B) + Q3 actual $9.3B + Q4 guide midpoint $10.7B = $44.8B FY25 at the midpoint — a substantial step up from our Q2-initiation FY25 framing of $35.4-35.8B. FY25 EPS now tracking ~$8.50-9.00 at the midpoint vs. our Q2 framing of $6.80-7.20. This is the most significant model upward revision since initiation.

FY26 implications: With FY25 exit-quarter at $10.7B revenue and 42% GM, the FY26 starting baseline is materially higher than we modeled at Q2. Continued HBM ramp + HBM4 introduction + non-HBM DRAM tightness + NAND mix-led recovery puts FY26 revenue range at $46-50B (vs. our prior $43-47B) with FY26 GM in the mid- to high-40s (vs. our prior 42-45%). FY26 EPS range $11-13, with bull case approaching $14 if HBM4 ramps cleanly into customer platforms and NAND maintains pricing discipline.

Capital structure: $12.2B cash & investments, $15.7B liquidity (including untapped credit facility), $15.5B total debt, weighted-average debt maturity to 2032. $1.7B in new notes refinanced the $900M 2027 notes during the quarter, extending the maturity ladder. Net leverage at $3.3B is the lowest in several quarters. The balance-sheet posture for the FY25-26 CapEx wave is in pristine shape.

Analyst Q&A Highlights

HBM Trajectory and TAM Scaling

• Tim Arcuri, UBS: Asked how HBM TAM scales with the accelerator TAM and whether there's an asymptote on HBM-attach. Sanjay framed HBM TAM growing from $18B CY24 to ~$35B CY25, with CY26 HBM bit-demand "significantly exceeding" overall DRAM industry demand. The 8-high → 12-high transition, HBM4 introduction in CY26, HBM4E in CY27, and customization in HBM in CY28 all add value per bit. The trajectory in HBM value proposition continues to be strong; Micron is well-positioned across the road map.
  Assessment: The HBM TAM ramp is accelerating, not flattening. Customization in CY28 (presumably tighter integration with specific accelerator customers) is a long-cycle differentiation lever Micron is positioned to capture given its vertical integration on the logic-die portion of HBM4.
• CJ Muse, Cantor Fitzgerald: Asked about HBM share trajectory in CY26 and the contributions from bits vs. ASPs. Sanjay confirmed Q3 HBM run-rate >$6B and framed share-parity-with-overall-DRAM "could be" achieved earlier than the prior end-of-CY25 framing. The HBM execution — output, yield ramp ahead of plan, 12-high yield ramp faster than 8-high — is the basis for accelerated share-parity expectations.
  Assessment: The "could be earlier" framing is a notable upward revision. We model Micron CY26 HBM share at 22-24% (vs. 20-23% in our Q2 initiation), with upside to 25%+ if HBM4 customer-allocation conversations break Micron's way.
• Tom O'Malley, Barclays: Asked about CY26 HBM share normalization given competitor qualification difficulties. Sanjay declined to forecast CY26 share specifically but framed HBM as part of the broader portfolio with "some movement" in share around different end-market segments. HBM uses 1-beta technology and is "fungible capacity with the rest of the portfolio" — a flexibility lever for managing mix and profitability.
  Assessment: The fungibility framing is the cleanest competitive answer. If HBM allocation softens, the 1-beta capacity flexes to non-HBM leading-edge DRAM at high margins; if HBM allocation tightens, more flexes in. This is the structural advantage of using a proven node for HBM rather than a bleeding-edge node.
• Harlan Sur, JPMorgan: Asked whether CY26 HBM supply is fully committed and pricing locked. Sanjay confirmed CY25 HBM is sold out and CY26 negotiations are underway "in the middle of '25" with customer mix between HBM3E 12-high and HBM4 still being finalized. HBM3E 12-high is in volume production with strong yield ramp; HBM4 sampled to multiple customers; CY26 customer needs being addressed across both.
  Assessment: CY26 visibility is solidifying but not yet locked. The HBM3E 12-high vs. HBM4 mix in CY26 is the next datapoint we want clarified — HBM4 carries a higher trade ratio and is a higher ASP product, so the mix matters for CY26 revenue and margin math.
• Vijay Rakesh, Mizuho: Asked about HBM4 qualification progress and whether the power leadership extends from HBM3E to HBM4. Sanjay confirmed early units delivered to multiple customers, very pleased with execution and specs, planning to maintain the strong power-position established with HBM3E. HBM4 uses well-proven 1-beta technology with internally developed and manufactured advanced CMOS logic die; volume ramp is a 2026 product.
  Assessment: The internally manufactured CMOS logic base die is a structural advantage. Other HBM suppliers have to source the logic die externally, which constrains optimization across the package; Micron's vertical integration enables a tighter performance-power envelope.

Gross Margin Sustainability and FY26 Trajectory

• Vivek Arya, Bank of America: Asked whether 39% Q3 GM is the new baseline and what drives the next several quarters. Mark framed Q3 outperformance vs. guide as price-driven (better-than-expected pricing despite reported price-down dynamics). Q3-to-Q4 bridge is favorable mix (more DRAM than NAND, more DC than consumer). Beyond Q4, declined to provide FQ1 FY26 guide but framed the trajectory as positive: tight inventories, mix to high-value DRAM and NAND products, "gross margins can be up."
  Assessment: The most important answer of the call. Mark explicitly told the market FQ1 FY26 GM "can be up" from the FQ4 42% guide — meaning mid-40s GM is on the table for fiscal Q1 FY26 (Sep-Nov 2025), pulling our prior Q2-initiation expectation forward by 2-3 quarters.

NAND Utilization and Inventory

• Tom O'Malley, Barclays (second part): Asked where NAND utilization stands today and the plan for FY26. Sanjay confirmed structural NAND wafer capacity reduction of ~10% exits FY25 vs. FY24 was implemented for capital-efficient G9 node transition. Underutilization has come down as capacity has been reduced; "leading edge of NAND is fully utilized" while "part of NAND continues to remain underutilized." Future utilization decisions are market-driven.
  Assessment: The under-absorption flow-through that we worried about three months ago is bounded. Leading-edge NAND fully utilized + structural 10% wafer capacity reduction means the period-cost drag is materially contained.

HBM Trade Ratios and Customer Mix

• Krish Sankar, TD Cowen: Asked about HBM4 trade ratio given the double I/O count via TSV, and whether GPU vs. ASIC HBM customers carry different margin profiles. Sanjay confirmed HBM4 trade ratio >3 (vs. ~3 for HBM3E), with HBM4E going "even greater" toward 4. Larger die size + higher performance drive the trade-ratio expansion. On margin difference between GPU and ASIC HBM: HBM commands high value across both platforms; content per accelerator continues to increase (200 to 288 GB today, going higher with 12-high capacity); declined to differentiate margin profiles between GPU and ASIC.
  Assessment: The trade-ratio progression (3 → >3 → toward 4) is the cleanest structural argument for non-HBM DRAM tightness. Each generation of HBM consumes more silicon per delivered bit, which constrains the wafer share available for non-HBM DRAM and supports DRAM pricing across the cycle.

CapEx and Capital Returns

• Joe Moore, Morgan Stanley: Asked whether 35%-of-revenue is a reasonable CapEx envelope for FY26 or if Micron needs to invest ahead of that. Mark framed the generational tech-transition opportunity, the leaner inventory levels (especially leading edge), HBM silicon requirements with high trade ratios, and the need for greenfield capacity. FQ3 CapEx of $2.7B was lighter than expected; FQ4 step-up implied to hit the unchanged $14B FY25 target. FY26 CapEx not provided; lumpy timing on equipment installs and grant disbursements. Net debt down to $3B, $900M 2027 notes refinanced with $1.75B further out, "great position with a flexible balance sheet to invest in the business... and then return capital to shareholders through dividend and opportunistic buyback."
  Assessment: The capital-return reference is notable — first time in several quarters that "opportunistic buyback" has been mentioned alongside dividend. With $1.9B FCF in Q3 (highest in 6+ years) and tightening leverage, the FY26 capital-return envelope expands materially. We do not assume buyback in our base case but flag it as upside.

Tariff Pull-Ins and Demand

• Chris Caso, Wolfe Research: Asked about tariff-related pull-in impact and how it affects 2H bit-demand visibility. Sanjay confirmed customer inventory levels healthy overall, some tariff-related pull-ins relatively modest, customers continuing to signal constructive CY25 demand environment driven by AI-data-center, auto/industrial recovery, distribution channel restocking, and AI-content growth in smartphones and PCs.
  Assessment: The pull-in is real but not material to the print or guide. The 2H CY25 demand environment described — constructive across end markets — is consistent with the FQ4 guide and supports our model upward revision.

What They're NOT Saying

1. FQ1 FY26 GM specific. "Can be up" framing implies mid-40s zone. Bull (45%+) and bear (43%) both inside the framing; the explicit decline to provide a number is appropriate posture given Q3-Q4 mix delta is unusually favorable.
2. HBM4 customer mix in CY26. Multiple customers sampled, qualifications ahead, but the HBM3E 12-high vs. HBM4 mix in CY26 is the key driver of CY26 revenue and margin math — not yet quantified. Watching for FQ4 datapoint.
3. FY26 CapEx envelope. Not provided. The $200B U.S. plan is a 20+ year arc but CY26-27 cadence (ID1 first wafer 2H CY27, ID2 ahead of NY1, Singapore advanced packaging on schedule) implies elevated near-term spend. FY26 CapEx-as-%-of-revenue is the next datapoint to track.
4. Capital return cadence. "Opportunistic buyback" was referenced for the first time in several quarters but no specific framing on size or trigger. With $1.9B FCF and $12.2B cash, the capacity is there; the timing remains opaque.
5. Tariff dollar exposure for 2H CY25. "Modest" pull-ins and "limited volume" framing remain — future regime broadening still unquantified. The "any impacts that occur due to potential new tariffs are not included in our guidance" framing is the explicit hedge.
6. The new business-unit structure granularity. FQ4 will be the first quarter reported under the new market-segment-focused units, but the mapping from CNBU/SBU/MBU/EBU to the new units has not been disclosed. The presentation discontinuity will make Q3-Q4 segment trends harder to track until the prior-period restated data is published.

Market Reaction

• After-hours move (June 25 evening): MU initially traded up >5% in the immediate after-hours print on the clean revenue, EPS, and FQ4 guide beats. The HBM milestone narrative ($6B run-rate, ahead-of-plan crossover, fourth customer) was the headline pickup; the GM beat (39% vs. 36.5% guide) and FQ4 GM guide (42%) were the secondary pickup that took the bid higher into the second hour of after-market trading.
• Next-day intraday context: June 26 trading saw a "sell-the-news" dynamic settle in. The stock had rallied meaningfully into the print on HBM-narrative momentum and AI-cycle beta; the print delivered the upgrade Wall Street wanted but the magnitude of the pre-print run-up absorbed much of the upside. MU traded down modestly in regular session and aftermarket on June 26, closing at ~$126 zone — net negative tape on the day despite the print being a clear above-Street result. The reaction is a positioning artifact, not a thesis-impairment signal.
• Pre-print context: MU had been a high-beta name through fiscal Q3 and the run-up incorporated meaningful HBM-narrative optimism. Buy-side expectations on FQ4 GM were running ahead of the ~37% Street, but no one had 42% on the card. The print delivered the upgrade; the tape just had it priced.

The reaction is rational from a positioning standpoint, but the print itself is the cleanest above-Street result Micron has delivered in this cycle. The thesis the tape is fading is precisely the thesis we are upgrading on. Outperform calls are best made on prints where the operational beat outpaces the price reaction; this is one of those.

Street Perspective

Debate: Is the GM trajectory now structurally re-based higher, or is Q4's 42% the cycle peak?

Bull view: 39% Q3 + 42% Q4 + "can be up" FQ1 FY26 framing pulls the GM bridge to mid-to-high 40s through fiscal H1 FY26. HBM mix expansion + HBM4 introduction at higher trade ratios + leading-edge DRAM tightness + tight inventories — the structural drivers compound. Cycle-peak GM in the high-40s with FY26-27 outperforming.

Bear view: The Q4 42% guide is mix-favored (DRAM>NAND, DC>consumer) and not sustainable as the FY26 NAND mix re-normalizes. Depreciation step-up from FY25-26 CapEx wave (potentially >$14B in FY26 given the $200B plan ramp) eats into mid-40s GM from the back end. Cycle peak is mid-40s with limited expansion from there.

Our take: Bull view closer to right on near-term trajectory; the structural depreciation re-base concern from Q2 has been bounded by the actual print. We model FY25 exit GM low-40s, FY26 GM mid- to high-40s, peak-cycle high-40s. The 50%+ peak is unlikely (depreciation step-up is real) but the mid-40s prior-cycle peak we modeled into Hold is now visible in the FQ4 guide itself. The right framing has shifted.

Debate: Can Micron sustain HBM share parity with overall DRAM into CY26?

Bull view: HBM3E 12-high yield ramp ahead of plan; fourth customer added; AMD MI355X design win; HBM4 sampled with internally manufactured logic die; CY25 sold out, CY26 negotiations underway. Share parity is achievable in 2H CY25 and sustainable through CY26.

Bear view: Samsung HBM3E qualification with NVIDIA may finalize in 2H CY25; SK hynix HBM4 ramp is on schedule; CY26 customer-allocation visibility is lower than the bull case implies. Micron's CY26 share could trail back toward 18-20%.

Our take: The Q3 print is a stronger datapoint than we had at Q2. Four customers (vs. three), AMD MI355X explicit, 12-high yield ahead of 8-high pace. We model Micron at 22-24% HBM share through CY26, with upside to 25%+ if HBM4 customer mix breaks Micron's way. The bear-case downside (18-20%) requires both Samsung qualification and SK hynix HBM4 acceleration to outpace Micron's execution — possible but not the base case.

Debate: Does the $200B U.S. investment plan stress the FY26-27 P&L?

Bull view: $200B is a 20+ year arc with grants, construction subsidies, and customer-validated demand backing it. FY25 CapEx unchanged at $14B; FY26-27 CapEx envelope manageable within the cash-generation profile (Q3 FCF $1.9B annualizes to ~$8B with continued mix-led GM expansion). Idaho ID1 first wafer 2H CY27; ID2 ramps before NY1. Capital deployment is disciplined, not ahead of demand.

Bear view: $200B over 20 years is $10B/year baseline before HBM-specific build. Combined with FY25-26 HBM and Singapore packaging spend, FY26-27 CapEx could push to $16-18B, eating into FCF and constraining the capital-return cadence. Depreciation step-up flows through GM with a 2-3-year lag.

Our take: The $200B framing is back-loaded; near-term FY26-27 CapEx envelope is the right concern. We model FY26 CapEx at $14-16B (vs. $14B FY25), with depreciation flowing through GM beginning FY27. The mix-led GM expansion through FY26 absorbs the depreciation step-up; the bigger risk is FY27 if CapEx steps materially beyond $16B and HBM mix levels off.

Model Implications

Valuation framing: At a low-double-digit forward P/E on FY26 EPS midpoint of ~$12, MU trades at a meaningful discount to AI-cycle peer multiples. The HBM ramp delivers the structural growth Wall Street demanded, the GM trajectory has inflected ahead of the consensus path, and the balance sheet is in pristine shape for the FY25-26 CapEx wave. The asymmetry has shifted: faster-than-modeled mid-40s GM achievement is now the base case rather than the bull case; cycle-peak GM in the high-40s is now plausible rather than aspirational. Outperform reflects this asymmetry — faster GM-recovery than we modeled at Q2, with HBM customer-base expansion and CY26 share-parity visibility ahead of plan.

Asymmetric scenarios that change our rating: (i) HBM 12-high yield slip on planned FQ4 crossover → reverts to Hold pending the next print; (ii) tariff regime broadening with customer push-back → Hold; (iii) Samsung HBM3E full qualification at NVIDIA + SK hynix HBM4 acceleration → Hold pending share-loss datapoint; (iv) FY26 CapEx step beyond $16B with no offsetting demand visibility → Hold. Outperform is the right rating for a print that resolved every Q2 Hold-gating concern and pulled the GM bridge forward by 2-3 quarters.

Thesis Scorecard

Overall: Six bull points confirmed (five strengthened from Q2), three bear points resolved or bounded, two bear points neutral or live-but-bounded, one new bear point (CapEx envelope) added but contained. The Hold-gating concerns from Q2 were specific and testable; the Q3 print resolved each one in the right direction. The remaining open questions — HBM4 customer mix in CY26, FY26 CapEx envelope, tariff regime broadening — are all bounded relative to the upside from the GM trajectory inflection.

Action: Upgrading from Hold to Outperform. Reversion-to-Hold triggers: HBM 12-high yield slip on planned FQ4 crossover; FQ4 GM below 41%; FY26 CapEx step beyond $16B without demand visibility; Samsung HBM3E full qualification with NVIDIA combined with SK hynix HBM4 acceleration; tariff regime broadening with customer push-back. Reversion to Underperform: same triggers compounded with NAND pricing breakdown or HBM customer-base contraction. We revisit on the FQ4 FY25 print in late September 2025, which will be the first print under the new business-unit structure and the first datapoint on the FY26 CapEx envelope.