Maintaining Outperform: Record Q4 With GM 370bps Above Guide, FQ1 Guide Clears the Street by ~$600M / ~250bps / ~$0.40, and the HBM Customer Base Widens to Six

Rating Action: Maintaining Outperform

This is the third report in our MU coverage arc. We are maintaining Outperform on the back of the Q4 FY2025 print and FQ1 FY26 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); (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 (Upgrading to Outperform) — June 26, 2025. Every Hold-gating concern resolved the right direction. Q3 GM expanded 110bps QoQ to 39.0% (vs. the 36.5% guide); FQ4 guide of 42% GM, $10.7B revenue, $2.50 EPS all comfortably cleared the Street; HBM revenue ~$1.5B (run-rate ~$6B), 12-high crossover pulled into FQ4, AMD MI355X added as a fourth HBM customer; DIO compressed 19 days to 139. We moved the FY26 EPS range to $11-13 and rated Outperform with downside triggers tied to HBM 12-high yield slip and tariff broadening.
• Q4 FY2025 (Today — Maintaining Outperform). The thesis compounds. Q4 revenue $11.3B beat the FQ4 guide of $10.7B by ~$600M; GM 45.7% beat the 42% guide by 370bps; EPS $3.03 beat the $2.50 guide by $0.53. The FQ1 FY26 guide of $12.5B / 51.5% GM / $3.75 EPS extends the trajectory and clears the 50% GM threshold a quarter early. HBM customer base expanded to six (from four at Q3); HBM revenue annualized run-rate is now ~$8B (vs. ~$6B); CY26 HBM3E pricing largely locked. DIO compressed another 15 days to 124 (within striking distance of the 120-day target). FY26 CapEx steps up to ~$18B (from $13.8B FY25), but the FQ1 guide already demonstrates that mix and pricing operating leverage is absorbing the depreciation re-base in real time.

The case for Maintaining Outperform: (1) The GM bridge to mid-50s is now visible. 39.0% Q3 → 45.7% Q4 actual → 51.5% Q1 guide is a 1,250bps three-quarter expansion against a Q2-call "up somewhat" framing that implied 38-40% zone for Q4. The CMBU segment GM of 59% in Q4 is a structural anchor for consolidated GM — with cloud memory now 40% of total company revenue and the segment running at a 59% GM, even modest mix improvement compounds. (2) The HBM customer base has scaled from three to six in two quarters, with pricing largely locked for CY26 HBM3E and HBM4 agreements about to follow. (3) Structural DRAM tightness is now management's base case for CY26 — constrained node migration (industry extends D4/LP4 EOL), longer lead times, higher costs for new wafer capacity, and HBM silicon-intensity creating displacement at the leading edge. (4) Inventory at 124 days is approaching target with DRAM "below target" and NAND inventory days improving sequentially. (5) FY26 free-cash-flow setup: management explicitly guided "significantly higher annual free cash flow YoY" in FY26 even with CapEx stepping to ~$18B.

Maintain-Outperform triggers met from the Q3 framing: "FQ4 GM at/above 42% with FQ1 guide implying 45%+ on credible bridge to 48%+" — FQ4 actual was 45.7%, FQ1 guide is 51.5%, both well above. "HBM customer-base expansion beyond fourth large customer" — explicit on the call: six customers. "DIO continues to compress toward 120-day target with FY25 exit framing intact" — DIO 124, DRAM below target, NAND improving. All three conditions met cleanly; the 51.5% FQ1 GM guide is the datapoint that decisively settles the prior-cycle GM-gap question.

Downside triggers from Outperform: HBM4 production-ramp execution slip in CQ2 2026 (the planned first-shipment window); CY26 hyperscaler capex pause that softens HBM/HC-DIMM pull; a NAND pricing reversal driven by competitor capacity additions; a tariff regime that materially impacts customer pull (management explicitly excluded tariff impact from FQ1 guidance). Upside triggers: HBM4E custom-base-die mix coming in higher than standard product, FY26 CapEx absorbed without GM compression even at the ramp peak, share gains in DC SSDs accelerating off the HDD-shortage tailwind.

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, again, the GM swing: 45.7% actual against the 42% guide is a 370bps upside surprise on top of a Q3 print that was already 250bps above its own guide. Two consecutive quarters of triple-digit-bps GM upside above guide is a pattern, not an artifact.

Quality of Beat

• The GM beat is again the headline. 670bps QoQ expansion against a guide of +300bps is a 370bps upside surprise. Mark Murphy attributed it to "favorable product mix, better DRAM pricing, and strong execution on cost reductions." The mix component is structural: data center revenue at 56% of FY25 total at 52% segment GM is now meaningfully above the legacy mix, and the new CMBU disclosure (59% GM, 40% of revenue) makes the pricing-power-at-the-edge dynamic explicit in the segment numbers for the first time.
• Revenue mix is the cleanest version of the bull case. DRAM $9.0B (+27% QoQ) on bit shipments low-teens with prices up low-double-digits — price + volume both contributing, the inverse of the Q3 setup where volume did the work against weak consumer pricing. NAND $2.3B (+5% QoQ) on bit shipments down mid-single-digits with prices up high-single-digits on favorable mix — the data-center NAND mix shift is now also a margin tailwind, not just a volume story.
• HBM run-rate is now ~$8B annualized. The ~$2B Q4 print represents another sequential record. The CMBU segment carries this growth: 40% of company revenue at 59% segment GM, +120bps sequentially. Critically, Sanjay confirmed the share-parity-with-overall-DRAM target is "on track" for CQ3 2025 — the precise quarter forecast at Q3.
• Cash flow has shifted from cycle-best to CapEx-investment phase. $5.7B operating cash flow funded $4.9B of CapEx with $0.8B of FCF. This is a deliberate trade-off: management explicitly framed FY26 CapEx as "higher" than FY25's $13.8B, with construction and front-end equipment for DRAM as the bulk. The signal: balance-sheet capacity is being deployed into the CY26-27 HBM/leading-edge-DRAM tightness window. With $11.9B cash + $15.4B liquidity and a 2033 weighted-average debt maturity, the runway is intact.
• Inventory cleanup continued. DIO compressed 15 days to 124 (vs. 158 at the Q2 initiation, vs. the 120-day target). Management framed FY26 DIO as remaining "at or better" than current levels, with DRAM "very tight" through the year and NAND DIO continuing to decline. The Q2 worry-list item is now decisively closed.

Segment Performance

Note on reporting structure: Q4 is the first quarter of the new business-unit segmentation flagged at Q3. The legacy CNBU/SBU/MBU/EBU structure is retired; Q4 reports under Cloud Memory (CMBU), Core Data Center (CDBU), Mobile & Client (MCBU), and Automotive & Embedded (AEBU). The arc-walk against Q3 is therefore a presentation discontinuity, but the economics tie: CMBU + CDBU = data center (54% of total Q4 revenue, vs. ~55% CNBU at Q3); MCBU = consumer DRAM/NAND (33%, vs. MBU 17% + parts of SBU); AEBU absorbs both auto and embedded (13%, vs. EBU 13%). The segment GM disclosure is the genuine new datapoint: each unit now reports its own GM, and the dispersion is wide (CMBU 59%, AEBU 31%).

CMBU: HBM Run-Rate ~$8B, Six Customers, 59% Segment GM

Cloud Memory revenue of $4.5B (+34% QoQ, 40% of total) is the segment that carries the equity case. HBM revenue grew to nearly $2B in Q4, a new quarterly record — an annualized run-rate of ~$8B vs. ~$6B at the end of Q3. Segment GM of 59% (+120bps sequentially) is supported by HBM mix and cost reductions on the 12-high ramp.

"Our HBM customer base has expanded and now includes six customers. We have pricing agreements with almost all customers for a vast majority of our HBM3E supply in calendar 2026. We are in active discussions with customers on the specifications and volumes for HBM4, and we expect to conclude agreements to sell out the remainder of our total HBM calendar 2026 supply in the coming months." — Sanjay Mehrotra, CEO

HBM datapoints from the call. (1) Run-rate ~$8B annualized on the Q4 pace; nearly $2B Q4 revenue, a new record. (2) Customer base expanded to six (vs. four at Q3, three at Q2). (3) HBM share parity with overall DRAM share "on track" for CQ3 2025 — the target Sanjay set several quarters ago, on schedule. (4) HBM4 samples shipped at industry-leading specs: bandwidth >2.8 TB/s per stack and pin speeds >11 Gb/s, both above the customer-required spec increase versus the JEDEC standard. (5) HBM4 first production shipments slated for CQ2 2026, with the ramp progressing through CY26. (6) HBM4E to follow with both standard and customized base-die options, the latter manufactured in partnership with TSMC and expected to deliver "higher gross margins than standard HBM4E."

LPDRAM-for-servers is now also a record line. LPDDR5 server revenue grew over 50% sequentially to a record level, with Micron sole supplier of LPDRAM to the data center for NVIDIA's GB family. GDDR7 product family added as another high-pin-speed (>40 Gb/s) leadership SKU for future AI systems.

Assessment: CMBU is the cleanest expression of the Outperform thesis. 40% of revenue at 59% GM is a structural anchor; the segment alone produces ~$10.6B of gross profit annually at run-rate — more than the entire FY24 Micron GP. The customer-base expansion from three to six in two quarters indicates that the HBM allocation conversation is now competitive across the entire AI-accelerator ecosystem, not just the NVIDIA/AMD core.

CDBU: 41% Segment GM Up 400bps, DC SSD Share Records Continue

Core Data Center revenue of $1.6B (+3% QoQ, 14% of total) carries enterprise/government DRAM and the data-center SSD business. The standout is the 41% segment GM, +400bps QoQ on pricing and mix. DC SSD market share reached a new record in FY25 with the first-to-market PCIe Gen6 SSDs on the G9 NAND node strengthening the portfolio. Near term, HDD supply shortages are expected to drive incremental NAND demand into the data-center storage tier, with management framing the supply-demand environment as "healthier" through CY26. Assessment: CDBU is a lower-growth but rapidly improving margin segment, with the 400bps GM step a meaningful contributor to the consolidated mix.

MCBU: +12 Percentage Points GM QoQ Is the Standout

Mobile & Client revenue of $3.8B (+16% QoQ, 33% of total) is structurally the lower-margin segment, but the GM expansion of 12pts to 36% is the largest single-segment swing in the print. Drivers: higher DRAM pricing and favorable mix into PC and AI-enabled smartphones. Smartphone unit shipments unchanged at low-single-digits CY25, but content growth is the lever — one-third of CQ2 flagship phones now ship with 12GB+ DRAM, with that mix expected to rise on the latest Apple/Samsung launches. PC unit growth raised to mid-single-digits CY25 (from low-single-digits) on Windows-10 EOL and AI-PC adoption. One strategic action worth noting: Micron ceased future mobile managed-NAND product development to focus resources on higher-ROI parts of the portfolio. Existing products continue to be supported. This is the first explicit product-line exit in the new structure and is mix-accretive over time. Assessment: MCBU is no longer a drag on the consolidated rating; the +12pt GM swing is doing the work the optimist case at Q3 had asked of it.

AEBU: D4/LP4 Allocation Pricing + Virginia Investment

Auto & Embedded revenue of $1.4B (+27% QoQ, 13% of total) at 31% segment GM (+540bps QoQ) reflects two drivers. (1) Auto demand strengthening on ADAS/AI-cabin content growth and improved profitability with greater D5/LP5 mix. (2) D4/LP4 allocation: Micron continues to be on allocation for legacy products, with the Virginia facility investment (announced June 2025) explicitly framed as supporting long-life-cycle customers' D4/LP4 demand for years to come. The "physical AI" framing — drones, advanced robots, AR/VR — is positioned as a medium-term embedded driver. Assessment: AEBU has gone from a small steady drag to a meaningful contributor, with the +540bps GM step among the cleanest in the print.

Key Topics & Management Commentary

Overall management tone: Forceful and explicit on the structural setup. Sanjay's lead framing — "Micron is positioned better than ever" entering FY26 — is a step up from the "exceeding the high end of guidance" formulation at Q3. Mark Murphy's GM commentary moved from describing what happened in the quarter to framing FQ2 sequentially up again on tight DRAM supply, with HBM and non-HBM both expected to remain healthy on the GM side through CY26. The capital-allocation conversation is openly tilted toward investment: $13.8B FY25 CapEx, ~$18B FY26 CapEx (implied from $4.5B/quarter baseline), with the bulk into DRAM front-end and Idaho/New York/Singapore/Japan capacity build-out for the CY26-CY28+ window.

The HBM Customer-Base Expansion: From Three to Six in Two Quarters

The Q3 worry on HBM was 12-high yield in early-life ramp. Q4 confirms the yield curve held: 12-high crossover happened in Q4 as planned, and the customer base widened from four to six in three months. Pricing agreements with "almost all customers for a vast majority" of CY26 HBM3E supply are concluded. HBM4 pricing/volume agreements expected to close out the remainder of total CY26 HBM supply "in the coming months." The combination — CY26 supply largely sold out before CY26 begins — is the cleanest demand-visibility signal management has provided in the cycle.

"In active discussions with customers on the specifications and volumes for HBM4, and we expect to conclude agreements to sell out the remainder of our total HBM calendar 2026 supply in the coming months." — Sanjay Mehrotra, CEO

HBM4 specs. Bandwidth exceeding 2.8 TB/s per stack (vs. the JEDEC spec of ~2 TB/s), pin speeds over 11 Gb/s. Built on Micron's proven 1-beta DRAM node with internally designed and manufactured advanced CMOS base die. Sanjay called it "industry-leading" on both performance and power efficiency. First production shipments slated for CQ2 2026; the ramp progresses through CY26 in line with customer-platform timing. HBM4E — standard products plus the option of customized base logic die manufactured by TSMC — targets CY27. Management explicitly flagged HBM4E custom base die as "higher gross margins than standard HBM4E," a key margin lever for FY27.

FY26 Industry DRAM/NAND Outlook: Tight Supply Becomes Management's Base Case

Sanjay's framing: CY25 industry DRAM bit demand growth in the high-teens (up from prior outlook); NAND bit demand growth in low-to-mid teens (up from prior). For CY26: "further DRAM supply tightness" and "continued strengthening in NAND market conditions." Medium-term industry bit-demand growth: mid-teens CAGR for both DRAM and NAND.

The supply-side framing is the load-bearing piece. Constrained node migration: as the industry extends D4/LP4 EOL, leading-edge-capacity migration is constrained — old nodes don't free up space for new-node ramps. Longer lead times. Higher costs globally for new wafer capacity: clean-room construction is expensive and slow. HBM silicon-intensity: HBM4's trade ratio (the wafer-area required per bit) is materially higher than HBM3E's, displacing leading-edge DRAM capacity at the same bit volume. The cumulative effect: Micron's view is that DRAM supply growth in CY26 will lag demand growth, and this gap is structural (not cyclical) through the planning horizon.

Capital Allocation: ~$18B FY26 CapEx, FCF Significantly Higher YoY

Management's FY26 framework: CapEx higher than FY25's $13.8B, with the implicit "reasonable quarterly baseline" of $4.5B implying ~$18B annual run-rate. Composition: DRAM front-end equipment for one-gamma node migration, fab construction (Idaho, New York, Singapore, Japan), and HBM capacity build-out. Despite the CapEx step-up, management explicitly guided "significantly higher annual free cash flow YoY" in FY26 — the GM operating leverage at 51.5% is doing the work.

One important nuance from CJ Muse's CapEx question: the $13.8B FY25 CapEx is net of $2B in government incentives ($15.8B gross). FY26 incentives have not been quantified yet, but with CHIPS grant disbursements continuing on Idaho fab construction, the gross/net gap is likely to widen, meaning the cash-cost of the CapEx step-up is somewhat softer than the headline number suggests. Idaho fab #1 first-wafer-out expected 2027; Idaho fab #2 design work has begun, providing capacity beyond 2028.

Mobile Managed-NAND Exit: First Explicit Product-Line Cull

Micron ceased future mobile managed-NAND product development to focus resources on higher-ROI portfolio segments. Existing products continue to be supported. This is the first explicit product-line exit in the new business-unit structure and signals management is willing to walk away from sub-scale categories where the ROI gap is wide. The mobile DRAM business remains a commitment, and the LP5X 10.7-Gb/s second-generation product qualified at OEMs in Q4. Read: margin-led portfolio discipline, of a piece with the new BU disclosure that surfaces segment GM ranges. Expect more of these decisions as the FY26 ROI prioritization works through the portfolio.

One-Gamma DRAM and G9 NAND: Technology Cadence on Track

One-gamma DRAM reached mature yields "in record time, 50% faster than in the prior generation." First revenue from a major hyperscaler on one-gamma server DRAM in Q4. One-gamma will be leveraged across the entire DRAM portfolio. The first EUV tool for the Japan fab installed in Q4 to enable one-gamma capability there, complementing existing one-gamma supply from Taiwan. G9 NAND ramp on track, with TLC and QLC ramping; G9 QLC qualified for enterprise storage; first-to-market PCIe Gen6 SSDs on G9. Read: the technology pipeline that supports the GM bridge through CY26 is intact.

Analyst Q&A: Selected Exchanges

Five questions on the call, all on supply/demand structure rather than execution. The dispersion is narrower than at Q3, where the questions had to grapple with the GM upside surprise itself; at Q4 the GM beat is taken as given and the conversation moves to whether the structural setup persists.

Tim Arcuri (UBS): FQ1 Mix and HBM TAM

Arcuri pressed Mark Murphy on the FQ1 sequential revenue mix between DRAM and NAND, and asked Sanjay for an update on the longer-term HBM TAM relative to the prior $100B-by-2030 framing given recent investment announcements from NVIDIA and others. Murphy on FQ1: heavier DRAM mix than NAND in the sequential growth; not breaking out bits/ASP, but the 580bps GM step is split across mix, pricing, and cost execution; "very constructive pricing environment" with DRAM tight and NAND improving. Mehrotra on HBM TAM: reaffirmed the $100B by 2030 framing and explicitly said the value proposition of HBM "continues to increase," with HBM bit-demand growth expected to outpace overall DRAM bits in CY26. Stopped short of raising the TAM number despite the prompt.

Vivek Arya (Bank of America): HBM3E→HBM4 Transition; CMBU GM Ceiling

Arya asked when HBM3E→HBM4 crossover happens in CY26 and whether HBM3E pricing is settling higher or lower vs. current; and on the gross-margin side, whether CMBU at 59% / 48% operating margin has further room. Mehrotra on HBM4: first production shipments CQ2 2026, ramp through CY26 in line with customer demand; share growing in 2026 vs. 2025; declined to comment on HBM3E pricing direction beyond "supply is tight" and "healthy demand-supply environment in 2026" benefiting both HBM and non-HBM profitability. Murphy on FQ2 GM: not providing out-quarter guidance, but explicitly expects sequential GM expansion FQ1→FQ2 on tight DRAM supply, NAND market continuing to improve, and ongoing mix-steering. Reiterated that GMs in both HBM and non-HBM are expected to remain healthy through CY26.

CJ Muse (Cantor Fitzgerald): Inference Inflection and FY-Q2 Seasonality; CapEx Detail

Muse asked about the recent inflection in DRAM demand led by inference hyperscalers, and whether the typically seasonally slower February (FQ2) holds normal seasonality this year given supply constraints; and on CapEx, asked Murphy to partition the gross figure between front-end equipment and clean-room space, plus the implied gross CapEx for FY26. Mehrotra on demand: AI strong in both training and inference; demand vector broadening across data center, AI PCs, AI smartphones; "healthy demand-supply environment" in CY26. Declined to offer FQ2 guidance directly but the structural setup (tight DRAM, healthy NAND) is positioned to override normal seasonality. Murphy on CapEx: declined to break out gross/net for FY26; the FY25 baseline was $13.8B net = $15.8B gross with $2B of government incentives; the $4.5B FQ1 baseline implies the run-rate; majority of FY26 spend is DRAM-related (construction, node-transition tools, greenfield).

Harlan Sur (JPMorgan): DIO, HBM4 Performance Achievement

Sur asked about exit-quarter DIO and lead-time visibility (DRAM is below target; what's the FY26 path?), and asked whether the team had to redesign the base-logic die to achieve the >11 Gb/s pin-speed and 2.8 TB/s bandwidth on HBM4, plus whether call schedules slipped or held. Murphy on inventory: DIO expected to remain "at or better" than current levels through FY26; DRAM very tight, will remain below target; NAND DIO continuing to decrease as that market improves. Mehrotra on HBM4: design win attributed to "innovative design, memory architecture, advanced CMOS in DRAM and the base die," with the base die manufactured internally by Micron as a competitive advantage. Customer schedules held; first production shipment CQ2 2026.

Krish Sankar (TD Cowen): HBM Sold-Out Quantification; HBM4 In-House vs. TSMC Mix

Sankar asked whether the HBM CY26 sold-out signal can be quantified in supply terms, and whether demand stronger than current supply could drive an upside response in CY26; and asked on HBM4 the mix split between Micron's in-house base die and the TSMC-manufactured base die for HBM4E, plus how easy it is to switch between the two based on customer demand. Mehrotra on supply: declined to break out volumes; HBM3E pricing/volumes essentially fixed with customers; HBM4 finalizing in coming months; flexibility comes from front-end fungibility (HBM uses the same 1-beta wafer base as other DRAM products) and from the assembly-and-test capacity Micron has built up over the past several quarters. The implicit answer: there is some swing capacity Micron can apply opportunistically. Mehrotra on HBM4 vs. HBM4E: HBM4 (CY26 product) uses Micron's internal base die; HBM4E (CY27 product) offers both standard and customized base-die options, the latter manufactured by TSMC; HBM4E custom base die expected to deliver higher GM than standard HBM4E.

The Q&A had a striking absence: nobody asked about pricing-cycle reversal risk, NAND oversupply, or hyperscaler capex pause. The Street is now substantially aligned with management's structural-tightness framing, and the dispersion is on execution timing rather than direction.

What They're NOT Saying

What management chose not to address — or addressed only obliquely — is informative.

• FY26 GM exit-rate. Murphy guided FQ1 GM to 51.5% and explicitly said FQ2 will be sequentially up again, but declined to provide out-quarter framing. Read: management is preserving optionality on the FY26 EPS path; with FQ1 at $3.75 and FQ2 sequentially up, the back-of-envelope math points to FY26 EPS comfortably north of $16. The reluctance to anchor that publicly suggests management views the upside-to-Street setup as a feature.
• HBM market-share-by-customer mix. The six-customer base is announced but the customer-by-customer mix is not. Read: at least one of the six is a smaller/emerging player, and the disclosure of the count without the names suggests management does not want to over-anchor expectations on any single ramp curve. The CY26 sold-out framing on HBM3E is the load-bearing datapoint — the customer mix is a derived quantity from there.
• HBM4 pricing direction. Sanjay explicitly declined to comment on HBM3E pricing direction (Vivek Arya's question) beyond "tight supply" and "healthy environment." The notable absence: no statement that HBM4 ASPs will be higher than HBM3E. Industry expectation is that HBM4 commands a meaningful premium on the trade-ratio expansion, but management is choosing not to anchor on that. Read: pricing negotiations are still open.
• Tariff impact specifics. Murphy explicitly excluded "any impacts that may occur due to potential new tariffs" from FQ1 guidance. No commentary on what scenarios the company has stress-tested. Read: this is a real but bounded risk that management is not yet quantifying. The Outperform thesis carries the assumption that tariff broadening does not cause customer pull-back materially before HBM3E CY26 supply is fully sold out.
• Mobile NAND exit financial impact. The decision to cease future mobile managed-NAND development is announced without quantification of the savings, the headcount impact, or the inventory tail. Read: this is more strategic signal than financial event. Management is communicating ROI discipline; the absolute financial magnitude is small enough that the disclosure of the principle matters more than the number.
• FY26 OpEx growth path. FQ1 OpEx guided to $1.34B, +12% sequentially, driven by R&D for "data center product innovation and development." FY26 has 53 weeks vs. 52 in FY25, with FQ4 FY26 absorbing the extra week. Beyond FQ1, no OpEx-trajectory guidance was provided. Read: R&D step-up is a forward investment in HBM4/HBM4E and one-gamma; the absolute level is manageable at the projected revenue scale, but the trajectory is upward.

Market Reaction

Context: MU's pre-print run had already absorbed a meaningful slug of the Q4 setup. The shares had advanced significantly through CY25 on the AI-memory thesis, with the August 11 mid-quarter guidance update (which raised the FQ4 framing into the $11B+ range) already pricing in much of the headline beat. The market reaction therefore reflects the GM upside and forward guide rather than the headline revenue, which was largely known.

The relative strength of the FQ1 guide is notable. Revenue $12.5B±$300M is a new all-time record; GM 51.5% breaches the 50% threshold a quarter ahead of even the optimistic post-Q3 trajectory; EPS $3.75±$0.15 is a new all-time record. With management framing FQ2 as sequentially up again, the implicit FY26 EPS run-rate is in the $16-19 range — a meaningful step up from where Street consensus sat coming into the print.

Street Perspective

The Street view going into the print was that the August 11 guidance update had largely fenced in the Q4 headline numbers, with the variable being CY26 HBM agreements and the FQ1 GM framing. The print resolved both: HBM CY26 supply largely sold out at the CY26 commencement, with HBM3E pricing/volumes effectively locked and HBM4 finalizing within months; FQ1 GM guide of 51.5% breaks the 50% ceiling that had bounded most of the prior consensus.

The bull case being made on the Street: Micron is now a structurally different company than at the FY24 trough. Data center 56% of FY25 revenue at 52% segment GM (vs. ~30% data center at substantially lower GM in FY24) is a different mix; CMBU 59% segment GM is a structural anchor; HBM CY26 supply largely sold out is a CY26 visibility floor; and the FY26 CapEx step-up is being absorbed by GM operating leverage in real time. The question is no longer whether the inflection is real but how high the FY26 EPS run-rate compounds: $16, $18, or $20+ depending on how aggressively the GM bridge into FQ4 FY26 and FY27 plays out.

The bear case being made on the Street: this is a memory cycle, and memory cycles end. The CY26 sold-out commentary on HBM3E doesn't anchor non-HBM DRAM pricing, which is the larger absolute revenue pool. Hyperscaler capex is at an unprecedented level; if even one major hyperscaler trims its CY26 build-out, the leading-edge-DRAM tightness narrative softens. CapEx at ~$18B is a material step-up; if GM operating leverage falters even modestly, the depreciation re-base hits hard. And the Q4 GM beat (370bps above guide) is the second consecutive quarter of triple-digit upside above guide — at some point, sandbagging-vs.-genuine-execution becomes a closer call.

Our read aligns with the bull case on this print. The bear case is real but bounded by the CY26 supply-side rigidities (constrained node migration, long clean-room lead times, HBM silicon-intensity displacement) that management framed as durable. The downside trigger to watch is HBM4 production-ramp execution into CQ2 2026 — if that timing slips, the FY26 GM bridge tightens.

Model Implications & Forward View

FY25 actuals: Revenue $37.4B (+49% YoY), non-GAAP GM 41% (+17pts YoY), non-GAAP EPS $8.29 (+538% YoY), operating cash flow ~$15B run-rate, net debt approaching neutral.

FY26 framework (revised up from Q3 upgrade):

FY27 placeholder: HBM4 ramp through CY26 supports FY27 EPS in the $20-25 range under base-case assumptions on margin persistence and CapEx absorption. The HBM4E custom base die mix is the largest single FY27 lever, with Sanjay explicitly framing custom HBM4E as "higher gross margins than standard HBM4E."

Thesis Scorecard: Q3 Upgrade Anchors vs. Q4 Outcomes

Six of seven anchors closed favorably; the one open watch item (tariff broadening) is the residual macro risk we can't price cleanly. The thesis is intact and the Q4 print/guide combination materially raises the FY26 EPS range.

Risks to the Outperform Rating

• HBM4 production ramp execution. First production shipment slated for CQ2 2026. Yield ramp on a new node with internal CMOS base die is the largest single technical execution risk. If timing slips by a quarter, FY26 GM bridge tightens.
• Hyperscaler capex pause. CY26 hyperscaler infrastructure spend is at an unprecedented level. A pause from even one major customer would soften HBM/HC-DIMM pull and could reverse the leading-edge-DRAM tightness narrative.
• NAND price reversal. NAND mix-led pricing strength in Q4 (prices up high-single-digits on bits down mid-single-digits) is partly cyclical. Competitor capacity additions or HDD-supply normalization could reverse this.
• Tariff regime broadening. Management explicitly excludes tariff impact from FQ1 guidance. A material expansion of tariffs in the customer-pull chain (consumer electronics, server OEMs) could compress demand even as supply remains tight.
• CapEx-driven depreciation re-base. $18B FY26 CapEx is a meaningful step-up. If GM operating leverage falters in CY26 (e.g., on HBM4 ramp slip or non-HBM DRAM softness), the depreciation re-base could compress GM faster than mix can offset.
• Memory-cycle risk. The structural tightness narrative is durable through CY26 in management's view, but DRAM/NAND remain commodity markets at the margin. A multi-quarter pricing softening would be a downgrade trigger.

Bottom Line

The Q4 FY2025 print extends the Q3 thesis cleanly. Revenue, GM, and EPS all printed materially above guide, and the FQ1 FY26 guide of $12.5B / 51.5% GM / $3.75 EPS clears the Street by ~$600M / ~250bps / ~$0.40 with management framing FQ2 as sequentially up again. The HBM customer base widened from four to six in three months; CY26 HBM3E supply is largely sold out before CY26 begins; HBM4 industry-leading specs sampled and on track for CQ2 2026 production. DIO compressed to 124 days, with DRAM below target. The FY26 CapEx step-up to ~$18B is being absorbed by GM operating leverage running well ahead of the depreciation re-base.

We maintain Outperform and widen the FY26 EPS range to $16-19 (vs. $11-13 at the Q3 upgrade). The structural setup is now a CY26-CY27 visibility window rather than a quarter-by-quarter beat-and-raise narrative. The downside triggers have moved from execution risk (HBM yield, NAND under-absorption) to macro risk (tariff regime, hyperscaler capex pause) and event-specific risk (HBM4 production ramp execution).