| Sep 30, 2026 | +3.03% | Q4 FY2026 | Read transcript briefingQuarter in one view- FQ4 FY2026 was a blowout on every line: revenue $54.2B (+31% q/q, +379% y/y, sixth consecutive record), gross margin 87% (+210 bps q/q), non-GAAP EPS $33.42 (+33% q/q) — all above the high end of guidance. FY2026 revenue $133.2B (+256%), GM 81.1% (+40 pts y/y), EPS $75.52 (+811%).
- FQ1 FY2027 guide: record $61.5B ±$1.5B revenue, GM ~86.25% (explicitly the FY27 floor), EPS $38.15 ±$1.00, opex ~$2.06B, tax ~15.5%. FY27 guided to sequential revenue growth every quarter with higher GM after Q1.
- SCA book scaled again: 26 signed (from 16), ~$32B customer financial commitments (from $22B+, "vast majority" cash), RPO ~$150B, estimated >35% of revenue through 2030, with extensions into 2031. New disclosure: three-quarters of SCA revenue has a defined pricing framework, a majority of that with floor/ceiling bands; the rest is periodically market-priced.
- Tightness language hardened further: "much tighter" supply-demand in CY27 and CY28 than CY26; no line of sight to balance; >75% of CY2027 output already committed (SCA + non-SCA POs); HBM CY2027 "vast majority" sold with significant price increases narrowing the GM gap to conventional DRAM.
- CapEx stepped up again: FQ1 ~$11.5B, 1H FY27 ~$25B, 2H higher (implies FY27 >$50B), construction-weighted, aimed at cleanroom availability late CY2028+. Capital return increase confirmed from December 9, 2026, with intent to return 100% of excess cash over time, primarily buybacks; new authorization to be sought (current $2.2B).
What management is focused on- Framing FY27 as better than FY26 with margins still rising: Q1 GM guided down 75 bps sequentially but explicitly called the FY27 floor; the ~$1B Q1 cost headwind (incentive comp absorbed into inventory, startup costs) was proactively quantified by Murphy and framed as partly "variable or temporal."
- SCA architecture as the structural-reset proof point: Mehrotra and Murphy repeatedly tied CapEx confidence, RPO conservatism ("revenue to well exceed RPO"), and margin durability ("even at floor prices, margins meaningfully above any prior cycle peak") to the SCA book.
- Pre-empting the peak-pricing/despec bear thesis: Mehrotra answered Arya's 2028 pricing question and Sankar's HBM-despec question with the same argument — content optimizations let customers ship more units (high-teens server unit growth CY26 and CY27), don't reduce latent memory need, and have "diminishing" effect; supply-side headwinds (HBM trade ratios, lower node-transition productivity, slow cleanroom ramps) keep 2028 tight.
- Capacity build-out with discipline language attached: six fab/packaging projects enumerated (NY first output CY2030, ID1 mid-CY2027, ID2 late CY2028, Japan late CY2028, Tongluo mid-CY2027, Singapore HBM packaging early CY2027, Singapore NAND 2H CY2028); equipment to be racked "as appropriate."
- Organizational restructuring: Bhatia to President/COO (end-to-end demand-to-supply, operating P&L), DeBoer to President/CTPO — framed as execution acceleration.
- Physical AI as the next demand leg: L4+ vehicles >200GB memory and multiple TB storage; humanoids comparable; "significant driver by end of this decade"; customers sampling next-gen products.
Key numbers and quarter mechanics- FQ4: revenue $54.2B; DRAM $39.8B (73%, +27% q/q; bits +mid-single digits, ASP +high teens); NAND $14.1B (26%, +42% q/q; bits ~+10%, ASP ~+30%); GM 87%; opex $2.6B (+$1.1B q/q on incentive comp + $300M community contribution); operating income $44.6B (82.3% margin); tax 15%; EPS $33.42.
- Segments (all records): CMBU $16.3B (30%), GM 83% flat (HBM mix offset pricing); CDBU $18.0B (33%, +56% q/q), GM 90% (+290 bps); MCBU $13.1B (24%), GM 90% (+260 bps); AEBU $6.8B (13%, +47% q/q), GM 84% (+470 bps).
- Cash flow: operating cash flow $44B; CapEx $10.8B; FCF $33.2B; SCA deposits received in FQ4: $12.3B, reported in financing activities (excluded from FCF); deposit balance $12.7B at quarter end (unrestricted, returnable back-half of terms if minimums met); cash/investments $73.5B; debt $5.2B; net cash $68.3B; two rating upgrades to BBB+ equivalent.
- Inventory: $10.4B, 129 days (+9 q/q) — attributed to no EOL build-ahead and incentive comp absorbed into inventory; DIO expected to decline.
- FY27 mechanics: ~$1B Q1 cost headwind (incentive comp inventory effect + startup + other); FY27 opex +~$2.5B (R&D + incentive comp); "more moderate rate of price increases" after Q1; tax ~15.5%.
- Industry outlook: CY26 DRAM bits +mid-20s% (Micron in line), NAND bits +low-20s% (Micron below); CY27–28 DRAM bits ~+low-20s%, NAND ~+mid-20s%, supply-constrained both years; HBM bits growing faster than conventional DRAM through CY2028.
- Data-center SSD revenue ~$10B in FQ4, >10x y/y, >two-thirds of NAND revenue; fifth consecutive record share year on track.
Product and launch scorecard- HBM: FQ4 HBM revenue grew faster than total company; CY2027 bit supply "vast majority" committed with significant y/y price increases narrowing the GM gap to conventional DRAM (timing: calendar 2027 pricing, negotiated now). HBM4 ramp "executing well"; NVHBM — industry's first custom HBM4E with NVIDIA for next-gen GPUs and NVLink Fusion — is the headline new product disclosure. HBM4E products "late next year." No share, yield, or 16-Hi qualification data given.
- DRAM nodes: 1γ is the largest production node and tracking to be the highest-volume node in company history; next-gen DRAM and NAND nodes on track for volume production 2H CY2027. Nearly half of MCBU FQ4 revenue from 1γ products.
- NAND/SSD: G9 is largest NAND node; data-center SSD momentum quantified ($10B quarter, 10x y/y); KV-cache offload and HDD displacement expanding TAM; participation in NVIDIA SCADA (direct GPU-to-storage) surfaced by Sur and not denied; Singapore NAND fab 2H CY2028.
- SOCAMM/DDR/LP server DRAM: referenced via press release only; no new attach or revenue data this call.
- Physical AI: content claims quantified (>200GB / multiple TB per L4+ vehicle); sampling underway; still a narrative, not a revenue line.
Sell-side read-through- Five questioners; tone congratulatory, and the two hardest questions (Arya on 2028 peak-pricing/despec, Sankar on HBM despec) were aimed at the bear narrative, not the quarter. No challenge to the numbers, the SCA economics, or the CapEx escalation.
- Arcuri (UBS): extracted the most new capital-return detail — target cash level reached by end of FQ1, FQ1 FCF "significantly higher" than $33.2B, buyback authorization increase coming. His FY27 CapEx math ("maybe $55B+, high-teens % of revenue") was not corrected; his 20–25% long-term capital-intensity probe got only "structurally reset, disciplined" — no number.
- Muse (Cantor): got the ~$1B Q1 margin-headwind quantification and confirmation that FY27 incentive comp hits manufacturing margins from Q2; HBM price increases confirmed as calendar-2027.
- Arya (BofA): pressed the $100B+ FY27 cash-generation math for capital return — Murphy gave intent, not quantum. His 2028 pricing question produced the call's most important forward statement: tighter in 27/28 than 26, no line of sight to balance, >75% of CY27 output committed, customer conversations already on 2028 and extensions to 2031.
- Sankar (TD Cowan): HBM-despec question was answered with the "optimizations have diminishing effect / enable more units" framing — no direct denial of any specific customer despec. His equipment-constraint hypothesis for construction-heavy CapEx was declined (lead times and greenfield visibility cited instead).
- Sur (JPMorgan): pushed SCA coverage toward 60–70%; Mehrotra held the line at ~50% of revenue through 2030 and clarified the >75% CY27 commitment figure includes non-SCA POs — a subtle walk-back of any implication that SCAs alone cover three-quarters of output.
Management credibility- Delivery vs. prior guidance is clean: revenue, GM, and EPS all above the high end; the FQ4 CapEx (~$10.8B vs ~$10B flagged) and startup-cost framing from last quarter landed as advertised. The December 9 capital-return commitment was reiterated on schedule with added specificity (target cash by end of FQ1, new authorization).
- SCA disclosure continued to deepen and is internally consistent: 16 → 26 agreements, $22B+ → $32B commitments, and new structural detail (75% of SCA revenue with defined pricing, majority with bands; RPO $150B on committed volumes at minimum pricing). The deposit accounting answer (financing activities, unrestricted, returnable) is now explicit — investors can model it.
- New, checkable claims to track: "even at floor prices, margins meaningfully above any prior cycle peak"; "revenue to well exceed RPO"; Q1 as FY27 GM floor; FQ1 FCF "significantly higher" than $33.2B. These are specific and falsifiable — credibility-positive if they hold.
- Evasions and soft spots logged: FY27 full-year CapEx still unguided (Arcuri's ~$55B uncorrected); capital-return quantum refused twice (Arcuri, Arya); HBM despec not directly denied (Sankar); long-term capital intensity target refused; SCA band widths, floors vs. spot, and counterparty concentration still undisclosed; HBM revenue dollars and share still not given.
- Language watch: "disciplined" appears alongside a CapEx trajectory of ~$27B FY26 → >$50B FY27; the ~50% SCA coverage target was reaffirmed despite Sur's 60–70% probe — consistent, but also a cap on the "contracted revenue" thesis.
What changed versus the prior quarter- From Q&A-only to full results: the prior transcript had no actuals; this quarter delivered records across every segment with full P&L, cash flow, and guidance.
- SCA book: 16 → 26 agreements; $22B+ → $32B commitments; new RPO metric (~$150B); coverage now ">35% of revenue through 2030" with 2031 extensions signed; pricing structure split disclosed (75% defined framework / 25% periodic market). The ~50%-of-revenue target was reaffirmed, not raised.
- Deposit mechanics clarified: deposits sit in financing cash flow, not FCF; $12.3B received in FQ4; $12.7B balance — versus ~$18B cash implied last quarter (the $32B is "vast majority" cash; the balance-sheet figure is lower than the cumulative commitment, worth reconciling in the 10-K).
- CapEx: from FY27 "above low-to-mid-40s% of revenue" to concrete near-term numbers — FQ1 ~$11.5B, 1H ~$25B, 2H higher, construction-weighted; Arcuri's ~$55B FY27 estimate uncorrected. Greenfield bit contribution timing unchanged (late CY2028+ for most; ID1 mid-CY2027).
- Tightness: "beyond 2027" → "much tighter in 27 and 28 than 26" with no line of sight to balance; HBM sold-out language now "vast majority of CY2027" with significant price increases — a change from "demand exceeds supply" to contracted pricing.
- Capital return: from pre-announced intent to dated mechanics — target cash by end of FQ1, 100% of excess cash over time, authorization increase coming.
- Margin story inverted: prior quarter emphasized startup-cost headwinds; this quarter delivered 87% GM and guided Q1 as the FY27 floor with the headwind quantified at ~$1B.
- Demand-destruction framing returned in softened form: double-digit PC/phone unit declines acknowledged but offset by premium-mix strength; server content growth "modestly lower than prior expectations" — a new, notable concession, framed as unit-enabling.
Bull case- The numbers now match the narrative: 87% GM, 82% operating margin, $33.2B quarterly FCF, and guidance for another record with GM rising through FY27 — the "structurally reset" claim is currently being validated by the P&L.
- Contracted visibility is real and growing: $150B RPO at minimum pricing (management: revenue will "well exceed" it), >75% of CY27 output committed, 2031 extensions, and floor-price margins "meaningfully above any prior cycle peak" — the downside case is increasingly collateralized.
- HBM economics inflecting: CY27 HBM repriced significantly higher, GM gap to conventional DRAM narrowing, custom NVHBM with NVIDIA extends the roadmap into NVLink Fusion — evidence against the HBM-commoditization and despec theses.
- NAND is a second engine at scale: ~$10B data-center SSD quarter (>10x y/y), two-thirds of NAND revenue, five straight record share years, KV-cache/HDD-displacement TAM expansion.
- Capital return is about to be very large: FQ1 FCF guided "significantly higher" than $33.2B, target cash reached within one quarter, 100% excess-cash return intent, authorization increase imminent — Arya's $100B+ FY27 cash math went uncorrected.
Bear case- CapEx is compounding faster than disclosure: >$50B FY27 (uncorrected), construction-weighted, with most bits arriving late CY2028 or CY2030 (NY). If the cycle turns before capacity lands, the fixed-cost base is roughly double FY26's. Management still refuses a full-year number or a long-term capital-intensity target.
- Ceilings cap the upside on a third-plus of revenue: the majority of defined-price SCA revenue has floor *and ceiling* bands; band widths remain undisclosed, so the upside surrendered in a rising market is unquantifiable. The 25% market-priced slice and ~50% coverage cap mean cyclicality is reduced, not eliminated.
- Server content growth was quietly revised down ("modestly lower rate of content growth than prior expectations") and PC/phone units face double-digit declines — the despec question was answered with framing, not a denial, and HBM despec by "one large customer" was not refuted.
- Deposit optics need scrutiny: $12.3B of FQ4 financing inflow is customer cash that is largely returnable; the $12.7B balance vs. $32B commitments gap and the return schedule require 10-K verification. FCF excludes deposits, but headline cash ($73.5B) includes them.
- Margin guide embeds a deceleration: "more moderate rate of price increases" after Q1, ~$1B of new ongoing costs, opex +$2.5B in FY27 — the easy pricing leverage is behind; from here, beats require volume and mix.
- Tax rate drifting up (15% → 15.5%) and incentive comp is now a structural, growing cost line.
Next-quarter watchlist- FQ1 print vs. guide: revenue $61.5B ±$1.5B, GM ~86.25% (must be the FY27 floor — any slip breaks a fresh promise), EPS $38.15 ±$1.00, and whether FCF is indeed "significantly higher" than $33.2B.
- December 9 capital-return event: size of the new buyback authorization, pace vs. the $2.2B remaining, and whether "100% of excess cash" becomes a defined policy with a target cash level disclosed.
- FY27 CapEx hardening: a full-year number vs. the uncorrected ~$55B; construction/equipment split; any change to ID1 (mid-CY2027), Tongluo (mid-CY2027), or Singapore HBM packaging (early CY2027) timelines.
- SCA progression: count beyond 26, movement in the ~50% coverage target, RPO growth from $150B, any disclosure of band widths/floors, and 10-K treatment of deposits (balance vs. the $32B commitments, return schedule).
- HBM execution: CY2027 pricing realization and GM-gap closure, HBM4 ramp, NVHBM milestones, HBM4E timing ("late next year"), and any confirmation or denial of large-customer despec.
- Demand gauges: server unit growth (high teens CY26/27) vs. the lowered content-growth assumption; PC/phone unit declines vs. premium mix; any replacement for the retired fulfillment metric.
- Margin trajectory: Q2 onward GM recovery as the $1B Q1 headwind rolls off; whether price increases stay "moderate" while GM still rises; opex tracking to +$2.5B FY27.
- Still unaddressed: SCA floor levels vs. spot, counterparty concentration, China/CXMT-YMTC trajectory, tariff exposure, HBM revenue dollars and share, and terminal GM.
|
| Jun 24, 2026 | +15.74% | Q3 FY2026 | Read transcript briefingQuarter in one view- Fifth consecutive record, another massive beat-and-raise: FQ3 revenue $41.5B (+74% QoQ, +346% YoY), GM 84.9% (+10 pts QoQ, record), EPS $25.11 (+106% QoQ) — all above the high end of the guide ($33.5B ±$750M / ~81% / $19.15 ±$0.40). The $17.6B sequential revenue increase eclipses last quarter's $10.2B record.
- FQ4 guide is another step-function: revenue $50B ±$1B (+21% QoQ), GM ~86%, EPS $31 ±$1 — but with the first explicit deceleration language: the guide "reflects a meaningful moderation in the rate of price increases."
- The SCA blackout broke — this is the quarter's real news. 16 SCAs signed (from 1 last quarter), ~20% of DRAM volume and ~1/3 of NAND volume covered, take-or-pay structure, floor/ceiling price bands with ceilings at CQ2 market prices, floors at GM "well above" prior-cycle peaks, ~$100B RPO at minimum pricing, and $22B of customer cash deposits/commitments (~$18B cash). When complete, ~40% of revenue under fixed/ceiling pricing and ~50%+ of revenue under SCAs.
- Pricing still parabolic but decelerating at the margin: DRAM price +low-60s% QoQ (bits +LSD), NAND price +mid-80s% QoQ (bits +MSD). Volume remains nearly flat; price is still the entire quarter.
- Tightness horizon extended again: now "beyond calendar 2027" (from "beyond 2026"), with no line of sight to supply catching demand even as industry supply improves gradually in 2028.
- Balance sheet transformed: cash and investments $30.2B record, net cash $24.4B (from $6.5B), debt cut to $5.7B, FCF $18.3B record, BBB+ upgrade. Capital return to increase after 12/9/2026 (CHIPS anniversary); intent to return 100% of excess cash over time.
What management is focused on- SCAs as a completed business-model transformation, now with economics disclosed: 16 signed across data center and auto (4 very large, 3 medium, rest smaller auto); typically 5-year terms CY2026–CY2030 (auto 3-year); take-or-pay with binding volume commitments; largest agreements carry ceiling prices at current CQ2 market levels and floor prices through the term; a modest portion are fixed-price or unbanded/market-priced. Mehrotra's key claim: floor-price GM is "well above our peak quarterly margins in any past cycle."
- Deposits as mutual commitment: $22B projected deposits/financial commitments (~$18B cash, ~$4B letters of credit); ~$10B arrives in FQ4, sits in financing cash flows (not FCF), unrestricted/fungible, returned to customers weighted to the back half of terms. Murphy framed it as evidence of binding commitment, not prepayment.
- Supply maximization with pulled-in timelines: Tongluo existing fab now supports meaningful shipments mid-CY2027, ~a quarter earlier than prior expectations; second EUV-capable cleanroom under construction; ID1 first wafers mid-CY2027, ID2 late-CY2028; Singapore HBM packaging meaningful contribution 1H CY2027; multiyear EUV supply agreement with ASML concluded for 1δ and beyond; Manassas 1α DDR4 production start for legacy auto/industrial/medical/aerospace/defense.
- Structural supply-constraint narrative deepened: greenfield lead times, trade-skill labor shortages, permitting, energy infrastructure, slowing bit growth per node transition, HBM trade-ratio pressure on non-HBM supply, and NAND cleanroom diversion to DRAM — all cited as reasons supply cannot catch demand on any visible horizon.
- Cost inflection acknowledged for the first time: blended DRAM cost per bit projected to rise from current levels (node transitions, HBM mix, greenfield ramps); SCAs provide for future price premiums on new products to be negotiated.
- Demand framing: CY2026 server units now +high-teens (raised from low-double-digits), enabled by a modest reduction in server DRAM content growth (allocation-driven); data center DRAM/NAND industry bits to more than double vs two years ago; humanoid robots carry 10x the memory of an L2+ vehicle; L2+ mix >20% this year, >40% by 2030.
Key numbers and quarter mechanics- FQ3: revenue $41.5B; DRAM $31.3B (76%, +343% YoY, +67% QoQ; bits +LSD, price +low-60s%); NAND $9.9B (24%, +361% YoY, +99% QoQ; bits +MSD, price +mid-80s%). Data center revenue >$25B in the quarter (>$100B annualized); data center SLC (SSD) revenue >$5B, more than doubled sequentially.
- Segments (all records): CMBU $13.8B (33%, +78% QoQ, GM 83% +9 pts); CDBU $11.5B (28%, +103% QoQ, GM 87% +12 pts); MCBU $11.5B (28%, +49% QoQ, GM 87% +9 pts, lower bits); AEBU $4.6B (11%, +71% QoQ, GM 79% +11 pts). Data center (CMBU+CDBU) = 61% of revenue (from 56%).
- Margins/EPS: GM 84.9% (+10 pts); OpEx $1.5B (+$97M, variable comp); operating income $33.7B, op margin 81.2% (+12 pts QoQ, +54 pts YoY); tax $5.1B at 14.9%; EPS $25.11.
- Cash/balance sheet: OCF $25.4B; CapEx $7.1B; FCF $18.3B record; cash and investments $30.2B record; debt reduced $4.4B (incl. $4.3B tender) to $5.7B; net cash $24.4B; weighted avg maturity April 2035; upgrades from all three agencies incl. BBB+.
- Inventory: $8.6B, 120 DIO (from 123); DRAM DIO below 120 and "very tight."
- SCA accounting: RPO >$5B at FQ3-end (small/auto agreements; ~$1.8B next-12-months); ~$100B RPO across 14 of 16 SCAs at minimum price/volume, to be disclosed in the 10-K with next-12-month layering; ASC 606 framing, "inherently conservative," revenue expected to "well exceed" RPO.
- FQ4 guide: revenue $50B ±$1B; GM ~86%; OpEx ~$1.65B; tax ~15% for FQ4/FY26; EPS $31 ±$1 on 1.15B shares. FY27 OpEx +$1B (R&D, weighted to 2H). FCF to "increase substantially again" in FQ4.
- CapEx: FQ4 ~$10B; FY26 ~$27B (from >$25B); FY27 quarterly CapEx above FQ4 levels, >half the YoY increase from construction as cleanroom capacity is pulled in.
Product and launch scorecard- HBM4: 12-Hi ramp tracking 2x faster than HBM3E 12-Hi; >$1B of HBM4 revenue already shipped (first dollar figure for any HBM product in the packet); mature yields expected "significantly faster" than HBM3E (still no yield data). HBM share deliberately held near DRAM share — confirmed as strategy, not shortfall — because of trade-ratio pressure on non-HBM supply.
- HBM margin question dodged: Muse asked directly whether HBM margins can bridge to DDR5 levels in CY2027; Mehrotra declined to comment on pricing, reiterated "strong ROI," and did not repeat or update last quarter's admission that non-HBM margins exceed HBM margins. The hierarchy question is now open again.
- Data center SSDs: data center SLC revenue >$5B, more than doubled sequentially (second consecutive doubling); AI context-store and HDD displacement cited as TAM expansion.
- Nodes: 1γ DRAM and G9 NAND "on track to become the highest volume nodes in Micron's history"; next-gen DRAM and NAND nodes on track for volume production 2H CY2027; ASML EUV multiyear agreement signed for 1δ+.
- Legacy/auto: Manassas 1α DDR4 first production start (auto, industrial, medical, aerospace, defense); auto SCAs (3-year) signed with smaller customers — auto is being used as the SCA proving ground for smaller counterparties.
- No updates given: HBM TAM, 16-Hi qualification status, LP server DRAM, HBF — none addressed on this call.
Sell-side read-through- Five questioners; tone congratulatory, and the SCA economics questions from last quarter were largely answered in prepared remarks — the Q&A was about filling in residual gaps rather than prying open a blackout.
- Arcuri (UBS): did the key math live — $100B floor revenue over ~5 years ≈ $20B/year, far below the ~$50B quarterly run rate, implying limited floor coverage. Mehrotra's answer: ~20% DRAM / ~30% NAND volume ≈ ~25% of revenue under signed SCAs, rising to ~40% under price-band/fixed structures when complete; floor GM above any prior peak. Murphy gave the modeling mechanics: RPO disclosures with next-12-month revenue per agreement cohort in the 10-Q/10-K.
- Moore (Morgan Stanley): pressed on deposit mechanics — escrow? collateral? Murphy: not a prepayment, unrestricted cash, held during performance, returned back-half-weighted; the deposit exists as a commitment device alongside take-or-pay binding volumes.
- Muse (Cantor): confirmed deposits are fungible for CapEx and won't change near-term gross-cash targets; his HBM margin-bridge question (can HBM reach DDR5 margins by CY2027?) was declined — the one notable evasion of the call.
- Arya (BofA): extracted that large/medium SCAs span data center, consumer, and auto, and that ceilings at CQ2 prices are "reflected in FQ3 results and FQ4 guidance" — i.e., the ceiling is roughly today's spot. His terminal-GM question (normalize to mid-70s?) got no number: Murphy repeated "incremental price yields less GM expansion," cited tightness beyond 2027, mix optimization, and mid-2027 volume/operating leverage — a deliberate non-answer on the long-term GM range.
- Sankar (TD Cowen): tried to back into per-GB floor pricing ($10–12/GB floor vs ~$25/GB current on 64GB server DRAM); refused on specifics, but Mehrotra reiterated floor GM "well beyond" prior peaks. His 2027 supply-gap quantification question got only "2027 overall tight," no magnitude.
- Notable absences: no questions on China, tariffs, consumer demand destruction, CapEx escalation, or the rising cost-per-bit disclosure. Only five analysts asked questions.
Management credibility- Beat-and-raise streak now five quarters, with the guide beaten by ~$7B on revenue. Last quarter's FQ3 guide ($33.5B/81%/$19.15) was conservative by a wide margin — the pattern of guiding well below delivery continues and is itself now a modeling fact.
- The SCA disclosure promise was kept and exceeded: last quarter's three-refusal blackout converted into structure (take-or-pay, bands, ceilings at CQ2, floors above prior-peak GM), coverage (~20% DRAM / ~33% NAND volume), RPO (~$100B), and deposits ($22B). This is the single largest credibility deposit in the packet — though per-GB pricing and cancellation terms remain undisclosed (Sankar refused).
- "FCF roughly double sequentially" delivered: $6.9B → $18.3B (+165%). Tongluo timeline pulled in a quarter. HBM4 "faster ramp than 3E" now supported by the 2x ramp-rate claim and $1B shipped — still no yield data, but the trajectory claim is gaining evidence.
- CapEx escalation continues: FY26 now ~$27B (from >$25B, ~$20B, ~$18B, $13.8B across five quarters), FY27 quarterly above $10B. "Disciplined" language persists; the SCA deposits and contracted volumes are now the stated justification for the spend.
- New disclosure with forward risk: blended DRAM cost per bit to rise — an unusually candid margin-structure admission, paired with the SCA mechanism to negotiate new-product premiums. Watch whether this becomes a margin headwind narrative in FY27.
- Evasions logged: per-GB SCA pricing (Sankar); HBM margin trajectory vs DDR5 (Muse); terminal/normalized GM range (Arya); 2027 supply-gap magnitude (Sankar); HBM share beyond "near DRAM share"; cancellation/termination mechanics of SCAs (Moore's question answered on deposits, not on what happens to deposits on cancellation).
What changed versus the prior quarter- SCAs went from 1 signed and zero economics to 16 signed with structure disclosed: take-or-pay, 5-year terms, floor/ceiling bands at CQ2 ceilings, ~$100B RPO, $22B deposits, path to ~50%+ of revenue contracted. The packet's top open item is substantially closed.
- Tightness horizon extended: "beyond calendar 2026" → "beyond calendar 2027," with no line of sight to supply catching demand even in 2028.
- First deceleration language on pricing: FQ4 GM guide of 86% "reflects a meaningful moderation in the rate of price increases" — the first plateau signal in the packet, though GM still rises ~1 pt.
- Server units revised up (low-double-digits → high-teens CY2026) with the mechanism disclosed: customers trading DRAM content per server for unit volume under allocation.
- Cost-per-bit inflection disclosed: blended DRAM cost per bit now projected to rise — a reversal of the industry's structural deflation assumption, with SCA premium-negotiation clauses as the offset.
- Balance sheet regime change: net cash $6.5B → $24.4B; debt $10.1B → $5.7B; BBB+; explicit commitment to raise capital return after 12/9/2026 and to return 100% of excess cash over time.
- HBM got its first revenue number: >$1B HBM4 shipped; ramp 2x HBM3E. But the HBM-vs-non-HBM margin hierarchy was not reiterated and the bridge question was dodged.
- Data center mix rose to 61% of revenue (from 56%); data center SLC revenue doubled sequentially again to >$5B.
- FY26 CapEx up again (>$25B → ~$27B) and FY27 quarterly CapEx guided above $10B.
Bull case- The cyclicality argument is now structurally impaired: ~40% of revenue heading into fixed/ceiling pricing with floors at GMs above any prior-cycle peak, take-or-pay volumes, $18B of cash deposits as commitment collateral, and ~$100B of conservative RPO. Even a price collapse scenario now has a disclosed floor that resembles prior-cycle *peak* profitability.
- Pricing still rising into FQ4 (GM 84.9% → ~86% guided) with tightness affirmed beyond 2027 and no supply-catch-up line of sight even in 2028; every supply lever (Tongluo mid-2027, ID1 mid-2027, ID2 late-2028, Singapore) is dated and pulled in, not slipped.
- Cash generation is outrunning the CapEx escalation: FCF $18.3B vs $7.1B CapEx; FQ4 FCF guided "substantially" higher alongside ~$10B CapEx; plus ~$10B of customer deposits arriving in FQ4. Net cash $24.4B with a stated intent to return 100% of excess cash.
- HBM execution now has hard numbers: $1B+ HBM4 shipped, 2x ramp rate vs HBM3E, share deliberately at DRAM parity — the strategy is margin-optimal mix management, not share maximization, and it's validated by the 87% CMBU/CDBU margins.
- NAND is a second engine at full power: +mid-80s% pricing, data center SLC >$5B and doubling, record segment margins.
- Demand broadening: server units revised up, agentic-AI CPU/storage racks, auto L2+ mix doubling, robotics/humanoids quantified (10x vehicle memory) — the non-accelerator demand story keeps thickening.
Bear case- The ceiling is now contractual for a large minority of the business: SCA ceilings are set at CQ2 market prices. If spot prices keep rising through CY2027, ~40% of revenue is capped at roughly today's levels while uncontracted competitors reprice higher — the SCAs truncate the upside tail they were signed during.
- Deceleration has begun: "meaningful moderation in the rate of price increases" is the first explicit slowdown language of the cycle. Bits remain ~flat (LSD DRAM, MSD NAND); when price stops rising, revenue stops rising — and FQ4's +21% guide embeds continued, if slower, price gains.
- Cost per bit is now guided to rise — node transitions, HBM trade ratios, and greenfield start-up costs (acknowledged for mid-2027 ramps) hit the cost line just as price momentum fades. The 86% GM may be the cycle print.
- Deposit mechanics carry a tail liability: ~$18B of customer cash must be returned, back-half-weighted toward 2029–2031 — an off-FCF financing inflow now, an outflow later, arriving precisely when new capacity (and potential oversupply) comes online. Cancellation treatment of deposits was never explained.
- CapEx keeps escalating into the exact window management can't see past: ~$27B FY26, >$10B/quarter FY27, with supply arriving 2027–2028 into a demand environment where management admits it has "no line of sight." The 2028 gradual supply improvement is the first easing signal in the packet.
- Terminal GM deliberately unanswered: Arya's mid-70s normalization framing was neither endorsed nor rejected; the market is anchoring on 84–86% prints with no management guidance on where GM settles as price bands bind and costs rise.
- Content-per-server is being cut to stretch supply — the server-unit revision embeds lower DRAM content growth, an early elasticity response inside the strongest end market.
Next-quarter watchlist- FQ4 delivery vs $50B / ~86% / $31 — and the first FY27 framing: does "meaningful moderation" in price increases become flat pricing? Track the incremental-price/GM language for the plateau call.
- SCA completion and the 10-K disclosures: remaining SCAs to reach ~50%+ revenue coverage; the ~$100B RPO with next-12-month layering; any disclosure of cancellation terms or deposit forfeiture mechanics; whether new SCAs extend beyond data center/auto into consumer at scale.
- Deposit flows: ~$10B of cash deposits due in FQ4 (financing cash flow); balance-sheet presentation; any change to the "unrestricted, fungible" treatment or capital-return counting.
- Price/deceleration mechanics: DRAM/NAND QoQ price vs this quarter's +low-60s%/+mid-80s%; bit growth still near zero; MCBU bits (fell again) as the demand-destruction gauge; server content-per-unit trend.
- HBM: HBM4 revenue trajectory off the $1B base, 16-Hi qualification, any yield disclosure, and whether the HBM-vs-DDR5 margin bridge question gets answered as CY2027 pricing negotiations begin.
- Cost line: evidence of the disclosed rising DRAM cost per bit — watch segment GMs for the first mix/cost headwind against moderating price, and FY27 OpEx +$1B phasing.
- CapEx and capacity: FQ4 ~$10B spend; FY27 quarterly run-rate hardening above $10B; Tongluo mid-2027 shipment prep and second-cleanroom progress; ID1/ID2 and Singapore packaging milestones.
- Capital return: the 12/9/2026 CHIPS anniversary passes before next quarter's report — size and form of the promised increase (buyback vs dividend), and whether deposits count toward "excess cash."
- Still unaddressed: China exposure, tariff quantification (excluded from guidance again), per-GB SCA pricing, deposit-on-cancellation treatment, 2027 supply-gap magnitude, and the normalized GM range — now the single most important missing number in the story.
|
| Mar 18, 2026 | -3.78% | Q2 FY2026 | Read transcript briefingQuarter in one view- Another massive beat-and-raise, bigger than the last: FQ2 revenue $23.9B (+75% QoQ, +196% YoY, record), GM 75% (+18 pts QoQ, record), EPS $12.20 (+155% QoQ, +682% YoY) — all above the high end of the guide ($18.7B ±$400M / 68% ±100 bps / $8.42 ±$0.20). Fourth consecutive record revenue quarter; the $10.2B sequential revenue increase is the largest in company history.
- FQ3 guide is another step-function: revenue $33.5B ±$750M (+40% QoQ), GM ~81% (+6 pts QoQ), EPS $19.15 ±$0.40. Mehrotra noted FQ3 single-quarter revenue guidance exceeds full-year revenue for every year in company history through FY2024. No FQ4 GM guidance given; Murphy reiterated that at these levels incremental price increases have less GM effect.
- Pricing went parabolic: DRAM price +mid-60s% QoQ (bits +MSD), NAND price +high-70s% QoQ (bits +LSD). This is the entire quarter's mechanics — volume is nearly flat, price is doing everything.
- First SCA signed: a five-year strategic customer agreement with a large customer — the LTA theme from last quarter converted into a signed contract, but all economics (pricing structure, cancellation terms, CapEx linkage, downside GM protection) were refused across three separate analyst attempts.
- CapEx raised again: FY26 to above $25B (from ~$20B), driven by the Tongluo (Powerchip) fab acquisition and US construction; FY27 construction CapEx alone to rise >$10B YoY, plus higher equipment spend. FY27 R&D also to increase "meaningfully."
- Capital return stepped up: 30% dividend increase to $0.15/quarter, $350M buyback, $1.6B debt paydown, record net cash $6.5B, two credit upgrades to solid BBB.
What management is focused on- SCAs as the business-model transformation: renamed from LTAs ("strategic customer agreements"), multiyear with "specific commitments," described as having "robust provisions" for both sides across tight and loose industry periods. First five-year SCA signed with a large customer; discussions with multiple customers across multiple markets. Mehrotra confirmed SCAs deepen R&D collaboration and roadmap planning (Sur's custom-base-die question), but refused every structural detail.
- Supply maximization at a new scale: Tongluo fab acquisition closed ahead of schedule (shipments from FY2028, second cleanroom construction to begin by 2026); second Idaho fab ground prep begun; New York groundbreaking done and ahead of plan; new Singapore NAND fab groundbreaking (first wafers 2028); India assembly/test now in commercial shipments; Singapore HBM packaging on track for meaningful CY2027 contribution.
- Demand-supply gap unchanged and reaffirmed: the 50%-to-two-thirds key-customer fulfillment figure from last quarter "still remains the case" (confirmed to Moore). Tightness expected "beyond calendar 2026" for both DRAM and NAND.
- Technology narrative: 1γ now claimed as fastest ramp to mature yields and fastest volume ramp in company history, majority of DRAM bits by mid-CY2026; increased EUV adoption planned at 1δ; G9 majority of NAND bits by mid-CY2026; record QLC mix.
- New demand vectors emphasized: robotics framed as a "20-year growth vector" and potentially one of the largest tech product categories; agentic AI PCs (32GB+ recommended configs), personal AI workstations (128GB), flagship phones at 12GB+ now ~80% mix (from <20% a year ago).
- Demand-destruction acknowledgment widened: PC and smartphone units could decline low double digits in CY2026 due to supply constraints and pricing — a harder statement than last quarter's elasticity concession.
Key numbers and quarter mechanics- FQ2: revenue $23.9B; DRAM $18.8B (79% of revenue, +207% YoY, +74% QoQ; bits +MSD, price +mid-60s%); NAND $5.0B (21%, +169% YoY, +82% QoQ; bits +LSD, price +high-70s%). Both records; HBM revenue also a record (no figure given, as is now policy).
- Segments (all records): CMBU $7.7B (32%, +47% QoQ, GM 74% +9 pts); CDBU $5.7B (24%, GM 74% +23 pts); MCBU $7.7B (32%, +81% QoQ, GM 79% +25 pts); AEBU $2.7B (11%, +57% QoQ, GM 68% +23 pts; auto+industrial revenue >$2B). Data center (CMBU+CDBU) = 56% of revenue, same as last quarter.
- Margins/EPS: GM 75% (+18 pts); OpEx $1.4B (+$87M, R&D-driven); operating income $16.5B, op margin 69% (+22 pts QoQ, +44 pts YoY); tax $2.5B at 15.1% — fifth consecutive quarter at this below-original-guide rate, now formally the FQ3/FY26 guided rate; EPS $12.20.
- Cash/balance sheet: OCF $11.9B; CapEx $5.0B; FCF $6.9B record (+77% vs the FQ1 record); cash and investments $16.7B record; liquidity >$20B; debt $10.1B after $1.6B reduction (2029/2030 senior notes redeemed; >$5B debt cut over three quarters); net cash $6.5B, highest ever; $350M buyback; dividend +30% to $0.15; weighted avg debt maturity August 2034; two credit upgrades (solid BBB).
- Inventory: $8.3B, 123 DIO (+$62M QoQ, -3 days) — back inside the old 124-day commitment; DRAM DIO still below 120 and "especially tight."
- FQ3 guide: revenue $33.5B ±$750M; GM ~81%; OpEx ~$1.4B; tax ~15.1% for FQ3 and FY26; EPS $19.15 ±$0.40 on 1.15B shares. GM bridge: higher price, lower cost, favorable mix. FQ4 OpEx to reflect 53rd week. Tariffs/geopolitics excluded. Murphy: FQ3 cash flow could "roughly double sequentially."
- CapEx: FY26 above $25B (from ~$20B; majority of increase is cleanroom/facility, largest factor Tongluo, then US fab construction); FQ3 CapEx ~$7B; FY27 to step up "meaningfully," construction CapEx +>$10B YoY plus higher equipment spend; construction growth to outpace equipment growth in FY26 and FY27.
- Return on capital: over 30%, "headed towards 50%" (Murphy — new disclosure).
Product and launch scorecard- HBM4: volume shipments of 36GB 12-Hi begun in 2026, designed-in for NVIDIA Vera Rubin (announced at GTC); mature yields expected faster than HBM3E (reiterated claim, now with ramp underway). HBM4 16-Hi (48GB, +33% capacity vs 12-Hi) sampled. HBM4E development "well underway," volume ramp CY2027 on 1γ. HBM4D customization options cited. The 81% FQ3 GM guide explicitly contemplates HBM4 mix growth (Murphy).
- HBM economics: Mehrotra stated plainly that non-HBM margins are currently higher than HBM margins — the first direct margin comparison in the packet, and a notable admission given the HBM-led narrative. HBM TAM not updated ($100B by 2028 stands); Muse's implied ~$50B CY2026 math went unconfirmed. HBM share disclosure again declined (Arya's 20–25% Vera Rubin question); Mehrotra corrected a slip to reaffirm share parity with DRAM was achieved in 2025.
- LP server DRAM: industry's first 256GB LP SoC-M2 sampled on 1γ — 2TB per CPU, 4x content vs a year ago (prior quarter: 192GB sampled). NVIDIA "Grok 3 LPX" rack-scale architecture cited with up to 12TB DDR5 per rack.
- Data center SSDs: data center NAND revenue more than doubled sequentially to a "substantial" record; G9 PCIe Gen6 SSDs now in high-volume production; 122TB SSD seeing "strong adoption" (16x sequential read throughput per watt vs capacity-matched HDD); data center SSD share up for the fourth consecutive calendar year to a record in 2025. Sur's estimate: ESSD ~50% of NAND mix, doubled sequentially — management did not dispute. Further growth guided for FQ3.
- Mobile/PC: LPCAM2 qualified at a major OEM; industry's first Gen5 QLC client SSD launched (G9); LPDDR5X designed into personal AI workstations with high volumes shipping; 1γ LPDDR6 sampling interest "strong"; 10.7 Gbps 1γ LP5X 16Gb in mass production.
- Auto/industrial: record AEBU revenue; first automotive-grade 1γ LPDDR5 sampled; first G9-based UFS 4.1 automotive solution; L2+ ADAS content math given (16GB today vs 300GB+ at L4).
- High-bandwidth flash (HBF): Mehrotra gave a cautious first assessment — capacity positives but NAND write-speed, power, and retention limits; "really early," under study, no R&D commitment (Sur's question).
Sell-side read-through- Seven questioners; tone congratulatory throughout. The dominant theme was SCA structure — three separate analysts (Sankar, Arcuri, Muse) probed pricing mechanism, cancellation terms, downside GM protection, and CapEx linkage, and all were refused. The market is clearly trying to model whether SCAs convert Micron into a contracted, floor-protected business; management is deliberately keeping that unanswerable.
- Sankar (TD Cowen): got the only forward GM color — no FQ4 guide, tightness beyond 2026, 81% includes HBM4 growth, and the repeated caveat that incremental price matters less at these GM levels. His SCA pricing/cancellation questions were declined.
- Moore (Morgan Stanley): extracted confirmation that the 50–66% fulfillment figure is unchanged three months later — supply tightness has not eased despite the capacity announcements. Allocation philosophy: stay diversified across end markets, data center gets a growing share.
- Arcuri (UBS): asked the sharpest strategic question — do SCAs create a GM floor on the other side of the cycle? Flatly declined. His cash question ($35–40B FY26 FCF, >$50B cash by year-end, CHIPS buyback restrictions) produced the capital-allocation hierarchy: balance sheet, organic investment, dividend (just raised), buybacks for dilution offset plus opportunistic. No commitment to rework CHIPS restrictions.
- Muse (Cantor): got two useful facts: the signed SCA is with a large customer, and SCA discussions span multiple markets (not just hyperscalers). His HBM TAM check (~$50B implied CY2026) got "no update"; his DDR5-vs-HBM margin question produced the admission that non-HBM margins currently exceed HBM margins.
- Sur (JPMorgan): confirmed ESSD momentum (~50% of NAND mix, doubled sequentially — unchallenged) and got the HBF brush-off. His thesis that custom HBM base-die design cycles drive SCA length was endorsed directionally ("brings us closer in R&D collaboration and roadmap planning").
- O'Malley (Barclays): raised the LPU/SRAM architecture risk; Mehrotra answered with the complementarity argument (Grok LPX uses 12TB DDR5 per rack; works alongside Vera Rubin HBM) and "efficiency grows the pie." His capacity-benchmarking question (how do you avoid oversupply in years 3–5?) went essentially unanswered — no bottoms-up methodology disclosed.
- Arya (BofA): HBM share question declined again; his historical-peak GM question (prior peaks low-60s) drew Murphy's "revert-to-mean should be revisited" argument — AI as secular driver plus structural supply constraints as durable margin supports. Notably, no one asked about China, tariffs, or the consumer demand-destruction number.
Management credibility- Beat-and-raise streak extended to an extreme degree: FQ2 beat the high end on all three metrics for the fourth straight quarter; the FQ3 guide implies revenue +40% and GM +6 pts to 81% — 20 points above the prior-cycle peak Arya cited. Last quarter's "GM up beyond FQ2, more gradual" was delivered (75% actual, 81% guided).
- LTA-to-SCA conversion delivered: last quarter's "in discussion, unsigned" became a signed five-year SCA with a large customer — the packet's top pending item partially closed. But the refusal to disclose any economics (price structure, floors, cancellation, CapEx linkage) means the market cannot yet verify the "stability and visibility" claim; credibility here is asserted, not demonstrated.
- HBM4 execution on schedule: CQ2 2026 ramp commitment from two quarters ago is now "volume shipments underway" for Vera Rubin; the "faster yield ramp than 3E" claim is repeated but still unverified by any yield data.
- CapEx credibility is the emerging tension: FY26 CapEx has gone $13.8B → ~$18B → ~$20B → >$25B in four quarters, with FY27 construction alone +$10B. "Disciplined" language persists throughout. The Tongluo acquisition is a genuine new supply lever, but the escalation rate is now the fastest in the packet.
- Mehrotra's HBM-share slip (said CQ3 2026, corrected to 2025 mid-answer) is minor but notable — he self-corrected promptly.
- Tax rate now fully institutionalized at 15.1% — five quarters at this level, now the formal guide; the EPS tailwind is baked in rather than disclosed as discrete items.
- Evasions logged: all SCA economics (Sankar, Arcuri, Muse — three refusals); HBM share trajectory (Arya); HBM TAM update (Muse); capacity-benchmarking methodology / oversupply guardrails (O'Malley); FQ4 GM (Sankar); CHIPS buyback-restriction rework (Arcuri); HBM revenue dollars (standing policy).
What changed versus the prior quarter- Pricing went from surging to unprecedented: DRAM +mid-60s% QoQ (from +20%), NAND +high-70s% (from +mid-teens). Bit growth remains minimal (MSD/LSD) — the quarter is entirely a price event.
- Margins re-based to a different regime: GM 56.8% → 75% actual → 81% guided; every segment GM now 68–79%, with MCBU (consumer-facing) the highest at 79% — a remarkable inversion given elasticity concerns.
- LTAs became SCAs and the first one signed: five-year term, large customer, multiyear specific commitments — but disclosure discipline tightened rather than loosened on the economics.
- CapEx stepped up twice in one quarter's narrative: FY26 >$25B (Tongluo acquisition closed early is the largest single factor) and FY27 construction +$10B YoY — the capacity response is now concrete and dated (Tongluo shipments FY2028, Singapore NAND 2028, Idaho mid-CY2027).
- Demand-destruction language hardened: from "pricing may trim some PC/phone units" to "PC and smartphone units could decline low double digits in CY2026." Simultaneously, content-growth data strengthened (12GB+ flagship mix ~80% vs <20% a year ago).
- HBM margin hierarchy revealed: non-HBM margins now stated to exceed HBM margins — reframes the HBM mix-management commentary from prior quarters and validates the deliberate HBM/non-HBM mix choice.
- Capital return expanded: dividend +30% (first dividend action in the packet), buyback $350M (from $300M), net cash $6.5B (from >$250M), two credit upgrades.
- New demand narratives added: robotics as a 20-year vector; agentic AI PCs/workstations; LPU/SRAM architectures addressed for the first time (framed as complementary, DRAM-intensive).
- Inventory back inside commitment: 123 DIO vs 126 — the minor miss from last quarter is repaired.
Bull case- The pricing cycle has no visible ceiling: DRAM +mid-60s% and NAND +high-70s% in a single quarter, GM guided to 81%, tightness affirmed "beyond 2026," and the 50–66% fulfillment constraint unchanged — supply relief (Tongluo FY2028, Singapore 2028, Idaho mid-2027) is years away while HBM trade ratios keep rising.
- FQ3 guide implies FCF roughly doubling sequentially (Murphy) on ~$7B CapEx — the cash machine is accelerating faster than the CapEx escalation; net cash $6.5B, ROC >30% heading toward 50%.
- SCA flywheel starting: first five-year agreement signed, multiple customers across multiple markets in discussion, and SCAs explicitly enable "investing with confidence" — if these contain downside protection (unconfirmed), the cyclicality argument weakens structurally.
- HBM4 position validated at the highest level: volume shipments for Vera Rubin underway, 16-Hi sampled, HBM4E on 1γ for CY2027, custom HBM4D engagement — and the admission that non-HBM margins exceed HBM margins means the mix is being managed for profit, not share.
- NAND transformed: data center NAND revenue doubled sequentially, Gen6 SSDs in HVM, 122TB adoption strong, share up four straight years — NAND is now a second pricing engine (+high-70s%) rather than a laggard.
- Content growth offsets unit destruction: 12GB+ flagship mix ~80% (4x YoY), AI PCs at 32GB+, workstations at 128GB, auto content 16GB→300GB+ at L4, robotics as a new category — bits per device rising even as units fall.
Bear case- The quarter is 100% price: bits +MSD in DRAM, +LSD in NAND. If pricing flattens, revenue flattens — and management itself keeps repeating that at 81% GM, incremental price adds little margin. The beat mechanics are exhausted; FQ4 GM guidance was withheld.
- Consumer demand destruction is now guided, not hypothesized: low-double-digit PC/phone unit declines in CY2026. MCBU and AEBU bit shipments already fell sequentially. The elasticity self-limiting mechanism is operating at these price levels.
- SCA opacity cuts both ways: three analysts failed to extract whether SCAs have price floors, annual renegotiation, or cancellation terms. If SCAs lock pricing near current levels, they cap upside in a still-rising market; if they don't protect the downside, the "stability" claim is marketing. Unverifiable either way.
- CapEx escalation is now the fastest-moving number in the story: >$25B FY26, +$10B construction in FY27, plus "meaningfully" higher R&D and OpEx. If the cycle turns before Tongluo/Singapore/Idaho output arrives (2027–2028), Micron carries a vastly larger fixed-cost base into a downturn — exactly the prior-cycle mistake management says it's avoiding.
- HBM margins below non-HBM — the flagship AI product is not the profit leader; if HBM mix rises into CY2027 (HBM4E ramp), it could be margin-dilutive at the portfolio level.
- EPS quality: tax at 15.1% for five straight quarters is now structural in guidance; any normalization is a headwind. Tariffs and geopolitics remain excluded from guidance; China unaddressed again.
- O'Malley's unanswered question hangs over everything: with demand concentrated in a few data-center customers signing multiyear agreements, management disclosed no methodology for avoiding oversupply in years 3–5 — the very risk SCAs are supposed to mitigate.
Next-quarter watchlist- FQ3 delivery vs $33.5B / 81% / $19.15 — and whether any FQ4 GM framing emerges; Murphy's "less GM effect from incremental price" caveat is the language to track for a plateau signal.
- SCA count and disclosure: additional signings, counterparty type (hyperscaler vs other markets), and any crack in the economics blackout — price structure, duration mix, cancellation terms, CapEx linkage. Arcuri's GM-floor question will be re-asked until answered.
- DRAM/NAND price trajectory: does pricing decelerate from +mid-60s%/+high-70s%, and does bit shipment growth stay near zero? Watch MCBU/AEBU bits for demand destruction and the low-double-digit PC/phone unit forecast for revision.
- HBM4 ramp proof points: Vera Rubin volume trajectory, 16-Hi qualification progress, any yield data behind the "faster than 3E" claim, and whether the non-HBM-over-HBM margin hierarchy persists as HBM4 mix grows in the 81% guide.
- CapEx execution: FQ3 ~$7B spend, Tongluo integration and second-cleanroom construction start, FY27 number hardening, and whether FCF "roughly double sequentially" materializes alongside it.
- Fulfillment gauge: any change to the 50–66% key-customer fulfillment figure — the cleanest tightening/easing indicator on the call.
- Data center SSD scaling: whether data center NAND doubles again or normalizes; Gen6 and 122TB/245TB qualification conversions; ESSD as % of NAND mix vs Sur's ~50% estimate.
- Capital allocation: buyback pace vs the $350M run rate, CHIPS restriction flexibility, debt paydown continuation, and whether cash builds toward Arcuri's >$50B year-end framing.
- Still unaddressed: China exposure, tariff quantification, HBM share trajectory, SCA economics, oversupply guardrails for 2028+ capacity, and the terminal GM level — now acute with 81% guided and no ceiling or floor stated.
|
| Dec 17, 2025 | +10.21% | Q1 FY2026 | Read transcript briefingQuarter in one view- Massive beat-and-raise, again: FQ1 revenue $13.6B (+21% QoQ, +57% YoY, record), GM 56.8% (+11 pts QoQ), EPS $4.78 (+58% QoQ, +167% YoY) — all above the high end of the guide ($12.5B ±$300M / 51.5% ±100 bps / $3.75 ±$0.15). Third consecutive record revenue quarter. Driver: DRAM price +20% QoQ (bits only up slightly), NAND price +mid-teens.
- FQ2 guide is a step-function unlike anything in this packet: revenue $18.7B ±$400M (+37% QoQ), GM 68% ±100 bps (+11 pts QoQ, 7 pts above the prior record), EPS $8.42 ±$0.20. Murphy also guided GM to keep expanding beyond FQ2, "more gradual" than the recent pace.
- The biggest open item is closed: price AND volume agreements completed for the entire CY2026 HBM supply, including HBM4. Last quarter's "HBM4 agreements in the coming months" promise is delivered — and this time pricing is explicitly included in the agreements, though direction/level is still undisclosed.
- HBM TAM raised for the first time in the packet: ~$35B in 2025 → ~$100B by 2028 (~40% CAGR), two years earlier than the prior 2030 framing. This answers Arcuri's two-quarters-running TAM probe.
- FY26 CapEx raised to ~$20B (from ~$18B), weighted to H2; FY27 CapEx expected up again. Idaho 1 first wafers pulled in to mid-CY27 (from 2H CY27); New York groundbreaking early CY26, supply 2030+.
- FCF inflection delivered: $3.9B record (vs $803M in FQ4), debt cut $2.7B, company returned to net cash (>$250M), and a $300M buyback restarted under CHIPS-agreement terms.
What management is focused on- The supercycle narrative, now stated in maximal terms: "the gap between demand and supply for all of DRAM, including HBM, is really the highest that we have ever seen"; tightness expected "through and beyond calendar 2026." Mehrotra quantified the shortfall: Micron can meet only ~50% to two-thirds of demand from several key customers in the medium term.
- Multiyear LTAs as the next structural leg: in discussion with several key customers across DRAM *and* NAND, spanning data center and other segments, with "specific commitments" and "far stronger contract structures" than prior LTAs. Not signed yet — timing, size, and terms all withheld (Danely pressed; no answer). Arcuri's bundling observation (DDR5 with HBM, even NAND) was implicitly confirmed.
- Supply maximization, not just pricing: pulling in equipment orders, accelerating installs, doubling brick-and-mortar construction CapEx FY25→FY26, adding cleanroom in Hiroshima, Singapore HBM packaging on track for CY27, India assembly/test in pilot production. Muse's "you're being conservative on capacity" challenge was met with "clean room space takes time... there is no near-term solution."
- 1-gamma as the CY26 supply engine: majority of DRAM bit output in 2H CY26; 1-delta and 1-epsilon development disclosed for the first time. G9 to become the largest NAND node later in FY26; QLC mix at a record.
- Demand forecasts raised across the board again: CY25 server units to high teens (from ~10%), PCs to HSD (from MSD), DRAM bit demand to low 20s% (from high teens), NAND to high teens (from low-to-mid teens). CY26 industry DRAM and NAND bit shipments both ~+20%, supply-constrained; Micron plans ~20% bit growth in both.
- AI-internal productivity story updated: 80%+ of professional workforce using GenAI (10x usage YoY), root-cause ID time cut in half, 30%+ coding gains.
Key numbers and quarter mechanics- FQ1: revenue $13.6B; DRAM $10.8B (79% of revenue, +69% YoY, +20% QoQ; bits up slightly, price +20%); NAND $2.7B (20%, +22% YoY, +22% QoQ; bits +mid-to-high single digits, price +mid-teens). Both records.
- Segments (all records): CMBU $5.3B (39%, +16% QoQ, GM 66% +620 bps); CDBU $2.4B (17%, +51% QoQ, GM 51% +990 bps); MCBU $4.3B (31%, +13%, GM 54% +17 pts); AEBU $1.7B (13%, +20%, GM 45% +14 pts). Data center (CMBU+CDBU) = 56% of revenue.
- Margins/EPS: GM 56.8% (+11 pts); OpEx $1.3B (+$120M, in line, R&D-driven); operating income $6.4B, op margin 47% (+12 pts QoQ, +20 pts YoY); tax $977M at 15.1% (vs 16.5% guided — a fourth below-guide tax quarter); EPS $4.78.
- Cash/balance sheet: OCF $8.4B; CapEx $4.5B (on baseline); FCF $3.9B record, >20% above the prior FQ4 2018 record; cash $12B; liquidity $15.5B; debt $11.8B after $2.7B reduction ($1B term loans + $1.7B senior notes); net cash >$250M; $300M buyback.
- Inventory: $8.2B, 126 DIO (-$150M QoQ, +2 days) — DRAM DIO below 120 days and "tight." The "at or better than 124" commitment technically missed by 2 days; dollar inventory fell.
- FQ2 guide: revenue $18.7B ±$400M; GM 68% ±100 bps; OpEx $1.38B ±$20M; tax ~15.5% for FQ2 and FY26 (lower than the 16.5% framed last quarter); EPS $8.42 ±$0.20 on ~1.15B shares. Tariffs excluded. FQ2 GM bridge: higher prices, lower costs, favorable mix.
- CapEx: FY26 ~$20B (from ~$18B), weighted to H2, primarily HBM supply capability and 1-gamma; brick-and-mortar construction roughly doubling FY25→FY26; FY27 expected up. Capital intensity acknowledged as below the historical ~35% frame — attributed to constructive markets and efficiency, not a policy change.
Product and launch scorecard- HBM (3E + 4): record FQ1 revenue (no dollar figure given — the prior $8B run-rate disclosure was explicitly declined when O'Malley asked). Entire CY26 supply sold out on price and volume, including HBM4. HBM4: >11 Gbps, in-house base die and core dies, production ramp CQ2 2026 on customer schedules, and a new claim — HBM4 yield ramp expected to be faster than HBM3E's. CY26 will carry both 3E and 4 with mix set by customer demand; Sur's ASIC/XPU (TPU, Trainium) 3E-upside question got no order-book specifics.
- HBM share: parity with DRAM share reiterated as achieved in CQ3, but forward share deliberately unquantified (Sankar's "20%?" question declined) — management now frames HBM vs non-HBM as an active mix-management choice since both are highly profitable.
- HBM4E: customized engagements highlighted as differentiation; no new structural detail beyond last quarter's TSMC disclosure.
- LP server DRAM: new milestone — 192GB LP SOCAM2 sampled, +50% capacity per module, >50TB rack-scale LPDRAM density; one-third the power of DDR5 modules. Sole-supplier claim not repeated this quarter but not contradicted.
- Data center SSDs: portfolio revenue >$1B in FQ1 (new disclosure); world's first PCIe Gen6 SSD (G9) seeing "rapidly increasing qualification commitments" including hyperscalers; 122TB/245TB G9 QLC SSDs entering qualification at multiple hyperscalers. Sur cited third-party data of ~25% QoQ enterprise SSD growth; management claimed share gains and confirmed data center SSDs are inside the multiyear LTA discussions.
- Mobile DRAM: 1-gamma 16Gb LPDDR6 sampled to OEMs (>50% performance uplift); 1-gamma LP5X 24Gb sampled, 16Gb in volume shipments. Flagship 12GB+ mix hit 59% in CQ3, >2x YoY.
- Auto/industrial: ASIL-rated LP5X and UFS 4.1 with "billions of dollars in design wins"; D4/LP4X demand strong, Manassas investment continues.
- PC: 1-gamma 16Gb DDR5 and G9 PCIe Gen4 QLC SSD OEM quals completed; management acknowledged memory pricing may trim some PC/phone unit demand — elasticity explicitly conceded for the first time (Arya's question), but said to be embedded in forecasts.
Sell-side read-through- Seven questioners; tone congratulatory, focused on LTAs, CapEx sufficiency, HBM share/pricing, and margin durability. A muted-line glitch briefly disrupted the session; no content lost.
- Arcuri (UBS): extracted the LTA framing (multiyear, DRAM+NAND, "specific commitments," stronger structure — no specifics) and the CapEx architecture: ~$20B FY26, construction doubling, FY27 CapEx up, capital intensity below the old 35% frame. His two-quarter TAM campaign was answered in the prepared remarks ($100B by 2028).
- Muse (Cantor): challenged capacity conservatism directly — got the cleanest supply statement of the call: bit growth is supply-constrained, no near-term fix, and Mehrotra's 50–66% demand-fulfillment admission. His HBM3E→4 cost-transition question produced the "HBM4 yields ramp faster than 3E" claim plus a startup-cost caveat (new fabs, small margin impact, begins '26–'27).
- Sur (JPMorgan): ASIC XPU 3E upside — acknowledged implicitly, no order-book color. Enterprise SSD: confirmed LTAs include data center SSDs; inference-vs-training storage intensity answered directionally (GenAI video drives SSD demand), not quantified.
- O'Malley (Barclays): asked for HBM as % of DRAM dollars and FQ1/FQ2 HBM revenue — both declined; the $8B run-rate metric is retired. Competitive-position question (Samsung HBM4 certification risk) answered with spec/power claims only.
- Sankar (TD Cowen): got the most useful margin guidance: GM expected to expand beyond FQ2 but more gradually, with the mathematical caveat that at high GM levels the same price increase yields less margin percentage. His HBM share-through-2028 question was declined.
- Danely (Citi): pressed LTA signing timing/holdups and whether AI customers would co-fund fabs — declined on all specifics; the fab co-investment question went unanswered. His HBM price-float question got the confirmation that CY26 HBM pricing is fixed by completed agreements, not floating.
- Arya (BofA): the only demand-skeptic question — price elasticity in consumer/enterprise. Management conceded possible unit-demand impact and customer mix-down in phones/PCs, said it's in the forecast, and countered with AI content growth. Notably, no one asked about China, tariffs, or the mobile managed NAND exit.
Management credibility- The under-promise/over-deliver machine is now extreme: FQ1 beat the high end on all three metrics for the third straight quarter; the FQ2 guide implies GM +11 pts sequentially to 7 pts above the prior record. Last quarter's verbal "GM up in FQ2" was not just kept but demolished.
- The CY26 HBM promise chain is fully closed: "agreements in the coming months" (FQ4) → "completed agreements on price and volume for our entire calendar 2026 HBM supply, including HBM4" (FQ1). This was the packet's top checkable item, and it was delivered early and in stronger form (pricing included) than promised.
- FCF promise kept: "FCF strengthens in FQ1" delivered as a $3.9B record; net cash restored; buyback resumed. The CapEx-catch-up FCF compression narrative is retired.
- TAM credibility improved by finally moving: the $100B-by-2030 hold (two quarters) gave way to $100B-by-2028 with a defined $35B 2025 base and ~40% CAGR — a falsifiable path now exists.
- New specificity on supply: the 50–66% demand-fulfillment figure and the ~20% CY26 bit-growth plan give the market concrete supply math for the first time.
- Persistent quality flags: tax at 15.1% vs 16.5% guided — fourth consecutive below-guide tax quarter; the FY26 rate was then lowered to ~15.5%, partially institutionalizing the tailwind. HBM revenue disclosure was withdrawn (no run-rate update) precisely as HBM becomes the most important line — a transparency step backward. DIO at 126 vs the "at or better than 124" commitment is a minor miss.
- Evasions logged: HBM revenue dollars and % of DRAM (O'Malley, twice); HBM share trajectory (Sankar); LTA timing, size, structure, and customer fab co-investment (Danely); CY26 HBM pricing level/direction (agreements confirmed, economics undisclosed); ASIC 3E order-book detail (Sur); FY26 gross CapEx and full-year revenue (Arcuri); cost-down trajectory for CY26 (Muse — declined).
What changed versus the prior quarter- Pricing accelerated dramatically: DRAM +20% QoQ (from +low double digits), NAND +mid-teens (from +HSD, and now on tight supply, not just mix). The pricing cycle went from rising to surging.
- CY26 HBM went from "largely locked, pricing direction hidden" to fully sold out on price and volume including HBM4 — but HBM revenue disclosure was simultaneously withdrawn, so the economics are now less visible, not more.
- HBM TAM raised two years forward: $100B by 2028 vs 2030; first TAM revision in the packet, in the direction Arcuri had been pushing.
- Margin trajectory went vertical: 45.7% → 56.8% actual → 68% guided → "up beyond FQ2, more gradual." MCBU GM jumped another 17 pts to 54%; CMBU hit 66%.
- CapEx raised again mid-year-planning: ~$18B → ~$20B, with FY27 now flagged as up and construction doubling — the "disciplined" language persists but the numbers keep stepping up.
- Demand outlooks raised for the second consecutive quarter: servers ~10% → high teens; DRAM bit demand high teens → low 20s; NAND low-to-mid teens → high teens; PCs MSD → HSD. New: CY26 industry bit growth framed at ~20% for both DRAM and NAND, supply-capped.
- LTAs emerged as a new structural theme: from unmentioned to the dominant Q&A topic — multiyear, cross-product, cross-segment, unsigned, with "specific commitments." This is a business-model change in progress.
- Balance sheet inflected: from deleveraging to net cash + buyback resumption in one quarter.
- First elasticity concession: management acknowledged memory pricing may dent consumer unit demand — new language, absent from all prior quarters.
- Idaho 1 pulled in (mid-CY27 vs 2H CY27); New York timeline firmed (groundbreak early CY26, supply 2030+).
Bull case- The pricing upcycle is now a pricing explosion: DRAM +20% in one quarter with bits flat; GM guided to 68% with expansion expected beyond; management states the demand-supply gap is the largest ever and can serve only half to two-thirds of key-customer demand. Supply additions (cleanroom lead times, HBM 3:1 trade ratio rising with future generations) structurally lag into 2027+.
- CY26 is substantially de-risked: all HBM sold out on price and volume; ~20% bit growth planned; LTAs in negotiation would extend visibility into 2027–2028 with "specific commitments" — potentially converting a cyclical business into a contracted one.
- HBM franchise validated: TAM pulled forward two years, HBM4 ramping CQ2 2026 with claimed faster yields than 3E, industry-leading >11 Gbps and 30% power advantage reiterated, custom HBM4E engagements expanding.
- Mix and breadth: data center at 56% of revenue with CMBU at 66% GM; data center SSDs >$1B and inside LTA discussions; auto design wins in the billions; even mobile/client GM at 54%.
- Cash machine confirmed: $3.9B FCF in one quarter against $4.5B CapEx; net cash; buybacks resumed; FY26 FCF guided significantly higher even at $20B CapEx. The bear case's FCF objection from last quarter is answered.
- Cost tailwinds still ahead: 1-gamma and G9 ramps proceeding well, HBM4 yields claimed better than 3E — margin support beyond pricing.
Bear case- Expectations are now parabolic: a 68% GM guide with "up from here" commentary leaves almost no room for deceleration; Murphy himself flagged the math — at these levels, the same price increase yields less GM expansion, and growth will be "more gradual." Any flattening reads as disappointment.
- HBM economics are now a black box: CY26 price and volume are fixed by completed agreements (no float, per the Danely answer), and management simultaneously stopped disclosing HBM revenue. If CY26 HBM pricing was conceded to lock volume, the packet cannot detect it until mix math reveals it.
- Elasticity is now on the record: management conceded phone/PC unit demand and mix-down risk from memory pricing. The consumer leg of the demand story has a self-limiting mechanism at these price levels.
- LTA execution risk: the contracts are unsigned, timing unknown, and the structure ("specific commitments") could cut both ways — volume commitments to customers at fixed terms could cap upside if spot pricing keeps rising, and Danely's fab co-investment question suggests customers may want supply without funding capacity.
- CapEx creep: $13.8B → ~$18B → ~$20B in two quarters, FY27 guided up, startup costs arriving '26–'27. Discipline language is constant while the numbers escalate; if the cycle turns, the fixed-cost base is much larger.
- Tax tailwind institutionalized: fourth straight below-guide quarter, and the FY26 rate was revised down to ~15.5% — EPS quality remains assisted.
- Supply-constrained growth caps the beat mechanics: ~20% bit growth is a ceiling, not a choice; upside must come almost entirely from price, which is the most cyclically fragile variable.
- Tariffs remain excluded from guidance; China unaddressed for another quarter.
Next-quarter watchlist- FQ2 delivery vs $18.7B / 68% / $8.42 — and whether "GM up beyond FQ2, more gradual" survives into formal FQ3 guidance; the gradualism caveat is the new language to track.
- LTA signatures: any announced multiyear contracts — counterparties, duration, product scope, price structure (fixed vs floating), and whether customers contribute capital to capacity. This is now the single largest pending disclosure.
- HBM economics without disclosure: with the run-rate metric retired, watch CMBU revenue/GM and DRAM ASP math for evidence of what CY26 HBM pricing actually looks like; HBM4 ramp milestones vs the CQ2 2026 commitment and the "faster yield ramp than 3E" claim.
- DRAM price trajectory: does +20% repeat, decelerate, or accelerate in FQ2; NAND price sustainability; any sign the elasticity concession is showing up in MCBU bit shipments (already negative in FQ1).
- CapEx and FY27 framing: H2-weighted $20B execution; whether FY27 commentary hardens into a number; startup-cost drag quantification.
- Inventory and fulfillment: DIO vs the 124-day commitment (126 this quarter); DRAM DIO below 120; any update to the 50–66% demand-fulfillment figure as a tightening/easing gauge.
- Tax: does 15.5% hold, or does a fifth discrete-item quarter appear.
- Enterprise SSD momentum: PCIe Gen6 qualification conversions, 122/245TB QLC hyperscaler quals, and whether data center SSD revenue scales meaningfully past $1B/quarter.
- Consumer demand: PC/phone unit commentary for evidence the elasticity management conceded is materializing; flagship memory-content mix (59% at 12GB+) as the offset.
- Still unaddressed: China exposure, tariff quantification, HBM share trajectory through 2028, LTA counterparty economics, and the terminal GM level — now an urgent modeling question with 68% guided and no ceiling stated.
|
| Sep 23, 2025 | -2.82% | Q4 FY2025 | Read transcript briefingQuarter in one view- FQ4 beat-and-raise delivered in full: revenue $11.3B (+22% QoQ, +46% YoY, record), GM 45.7% (+670 bps QoQ), EPS $3.03 (+59% QoQ, +157% YoY) — all above the high end of the updated guide ($10.7B ±$300M / 42% ±100 bps / $2.50 ±$0.15). Drivers: DRAM pricing (+low double digits — the first positive DRAM price print in this packet's history), favorable mix, and cost execution.
- FQ1 FY26 guide is another step-function: revenue $12.5B ±$300M (record), GM 51.5% ±100 bps (+580 bps QoQ), EPS $3.75 ±$0.15 (record). Murphy went further: GM expected to improve again sequentially in FQ2. The 42% "new baseline" question from last quarter is answered — it was a waypoint, not a ceiling.
- FY25 final: revenue $37.4B (+49%), GM 41% (+17 pts), EPS $8.29 (+538%), FCF $3.7B (10% of revenue). HBM + high-cap DIMM + LP server combined revenue hit $10B, >5x YoY.
- HBM: ~$2B in FQ4, ~$8B annualized run rate (vs >$6B last quarter); share parity with DRAM share declared "on track... in calendar Q3" — the pulled-forward target is being claimed as delivered. Customer base now six (from four). Critically, pricing agreements are done with almost all customers for the vast majority of CY26 HBM3E supply, with HBM4 agreements expected "in the coming months" to sell out the remainder of CY26 HBM — this closes the biggest open item from last quarter (Sur's unanswered CY26 question).
- HBM4 specs raised and sampled: >2.8 TB/s bandwidth, >11 Gbps pin speeds (vs >2 TB/s disclosed last quarter); first production shipments CQ2 2026. HBM4E base die will be made by TSMC (standard and custom); custom base-die HBM4E expected to carry higher GM than standard.
- Demand outlooks raised across the board: CY25 server units ~+10% (from +MSD), traditional servers flat→+MSD, PCs LSD→MSD, NAND bit demand low double digits→low-to-mid teens; DRAM held at high teens ("somewhat higher"). CY26 framed as further DRAM tightness and strengthening NAND.
- CapEx steps up hard: FY26 ~$18B net implied ($4.5B quarterly baseline) vs $13.8B net FY25 — the FY26 number Murphy declined to give last quarter is now effectively out.
What management is focused on- The pricing cycle has turned from "falling less than planned" to actively rising: DRAM price +low double digits QoQ on "tight industry DRAM supply, pricing execution, and favorable mix"; NAND price +HSD on mix. Murphy's FQ1 bridge: "price, cost, and mix to all contribute." Language has shifted from defensive (trough calls) to offensive ("constructive pricing environment," "durable" supply-demand factors).
- CY26 HBM sell-out as the credibility centerpiece: Mehrotra led with pricing agreements locked for the vast majority of HBM3E CY26 supply and committed to concluding HBM4 agreements to sell out remaining CY26 HBM supply "in the coming months." This is a direct answer to last quarter's evasion — but note pricing *direction* on CY26 HBM3E was explicitly not commented on (Arya).
- Structural tightness narrative for CY26: low supplier inventories, node migration constrained by extended D4/LP4 EOL support, longer lead times, higher global wafer-capacity costs, HBM silicon intensity. Management is building the case that tightness is multi-year and supply-driven, not just demand-driven.
- 1-gamma as the cost/supply engine: mature yields in record time, 50% faster than prior gen; first revenue from a major hyperscaler on 1-gamma server DRAM; 1-gamma provides the majority of CY26 DRAM supply growth while 1-beta shifts to HBM support. Japan EUV installed (record install time) to complement Taiwan 1-gamma.
- Portfolio pruning for ROI: ceased future mobile managed NAND development (existing products supported) — a notable admission that a product line didn't clear the return bar.
- AI as internal productivity story: 30–40% productivity uplift in select GenAI use cases (code generation), 5x wafer images analyzed, 2x fab telemetry data — framed as yield/time-to-market advantage.
- Data center mix milestone: 56% of FY25 revenue at 52% GM; new segment disclosure puts CMBU's 59% GM / 48% op margin (per Arya) on display.
Key numbers and quarter mechanics- FQ4: revenue $11.3B; DRAM $9.0B (79% of revenue, +69% YoY, +27% QoQ; bits +low teens, price +low double digits); NAND $2.3B (20%, -5% YoY, +5% QoQ; bits -MSD, price +HSD on mix).
- New segment disclosure (first report): CMBU $4.5B (40% of revenue, +34% QoQ, GM 59% +120 bps, record HBM); CDBU $1.6B (14%, +3%, GM 41% +400 bps); MCBU $3.8B (33%, +16%, GM 36% +12 pts); AEBU $1.4B (13%, +27%, GM 31% +540 bps). Data center (CMBU+CDBU) = 54% of Q4 revenue.
- Margins/EPS: GM 45.7% (+670 bps); OpEx $1.2B (+$81M, in line, R&D-driven); operating income $4.0B, op margin 35% (+820 bps QoQ, +12 pts YoY); tax $471M at 12% — below guide again on discrete items (third consecutive tax-assisted quarter); EPS $3.03.
- Cash/balance sheet: OCF $5.7B; CapEx $4.9B (the flagged FQ4 catch-up arrived); FCF $803M (down from >$1.9B on the CapEx surge); cash $11.9B; liquidity $15.4B; debt $14.6B after $900M paydown ($700M term loans + ~$200M notes repurchase); WAM 2033.
- Inventory: $8.4B, 124 DIO (-15 days, -$372M) — the "near target" exit delivered; DRAM DIO now below target, NAND improved. Murphy guided DIO "at or better" than FQ4 going forward, DRAM below target through the year, NAND DIO to decrease.
- FQ1 FY26 guide: revenue $12.5B ±$300M; GM 51.5% ±100 bps; OpEx ~$1.34B ±$20M (data center R&D); tax ~16.5% (the flagged FY26 step-up, better than the "high teens" prior frame); EPS $3.75 ±$0.15 on ~1.15B shares; CapEx ~$4.5B as quarterly baseline. FY26 is a 53-week year (extra week lands in FQ4 FY26 OpEx). Tariffs excluded from guidance.
- CapEx detail: FY25 $13.8B net / $15.8B gross / $2B government incentives (US, Singapore, Japan); FY26 "vast majority DRAM" — construction, node-transition tools, greenfield install starts. CHIPS grant disbursement received on an Idaho construction milestone; ID1 first wafers 2027; ID2 design begun (capacity beyond 2028); Singapore HBM assembly/test contributes CY27.
- FY25 cost execution: DRAM all-in costs (incl. HBM) down LSD; NAND all-in costs down ~low teens.
Product and launch scorecard- HBM3E 12-high: ~$2B quarter, ~$8B run rate, share parity claimed for CQ3, six customers, CY26 supply largely priced and agreed. The multi-quarter commitment chain (run rate → crossover → parity → CY26 lock) is now substantially closed.
- HBM4: samples shipped with raised specs — >2.8 TB/s, >11 Gbps (vs >2 TB/s at sampling last quarter); claimed to outperform all competing HBM4; first production shipments CQ2 2026, ramping through 2026; agreements to sell out remaining CY26 supply expected in coming months. Qualifications still ahead — the key remaining risk.
- HBM4E: standard + custom base logic die, manufactured by TSMC — a strategic shift from the in-house base die used for HBM4; custom versions carry higher GM. Framed as a 2027 product. Trade ratio not discussed this quarter.
- LP5 server DRAM: +50% QoQ, record revenue; sole supplier of LPDRAM to the data center since NVIDIA's GB family launch — the sole-source claim is now explicit and uncontested on the call.
- GDDR7: >40 Gbps pin speeds, positioned for "certain future AI systems" — new emphasis this quarter.
- 1-gamma DRAM: mature yields 50% faster than prior gen; first hyperscaler revenue on server DRAM; first OEM qual of 16Gb 1-gamma D5 with volume shipments started; 1-gamma LP5X quals referenced from last quarter. Now the designated CY26 supply-growth vehicle.
- G9 NAND: TLC and QLC both ramped; G9 QLC qualified for enterprise storage; first-to-market PCIe Gen6 data center SSDs; record client SSD revenue in Q4 and FY25. HDD shortages flagged as a NAND demand tailwind.
- Mobile managed NAND: future development ceased — resources redirected to higher-ROI opportunities; existing products supported. Mobile DRAM commitment reiterated (10.7 Gbps 1-beta LP5X quals at 16/24GB).
- Auto/industrial: demand strengthened above forecast; profitability improving on pricing and D5/LP5 mix; D4/LP4 constraints persist; Virginia (Manassas) investment supports long-lifecycle customers.
Sell-side read-through- Five questioners; tone congratulatory, focused on guide mechanics, HBM CY26, and CapEx. No questions on China, tariffs (beyond guide exclusion), buybacks, or the mobile NAND exit.
- Arcuri (UBS): got the FQ1 revenue split (heavier DRAM than NAND) and the GM bridge (mix, pricing, cost). His HBM TAM probe — does the $100B number get raised given hyperscaler capex inflation? — got a reaffirmation of "$100B by 2030," not a raise. Mehrotra corrected the framing (2030, not 2028) and leaned on value-proposition language instead of a bigger number. Second straight quarter without a TAM raise.
- Arya (BofA): asked the two sharpest questions. (1) HBM3E→HBM4 crossover timing and CY26 HBM3E pricing direction — declined ("we are not commenting on the pricing of HBM3E"), though share is expected to grow in 2026. (2) Is 51.5% the baseline? Murphy gave a concrete answer: GM up sequentially in FQ2 on tight DRAM, NAND improvement, mix, and cost — and reiterated margins healthy in both HBM and non-HBM in '26.
- Muse (Cantor): tested demand breadth/sustainability and February seasonality — Mehrotra broadened the demand vector (inference, traditional servers, AI phones, AI PCs) but gave no FQ2 seasonality answer. His CapEx math extracted the ~$18B net FY26 figure and the FY25 gross/net/incentive decomposition ($15.8B/$13.8B/$2B); FY26 gross declined.
- Sur (JPMorgan): confirmed inventory at target with DRAM below target, and got forward commitment: DIO at or better, DRAM below target through the year, NAND DIO falling. Lead-time extension question was answered only anecdotally ("customers are fully aware... supply is very tight") — no quantified lead-time or order-visibility disclosure.
- Krish Sankar (TD Cowen): asked to quantify CY26 HBM supply opportunity assuming sell-out — declined (no volume breakdown). His HBM4 vs HBM4E base-die question produced the cleanest product-structure disclosure of the call: HBM4 = internal base die (2026 ramp); HBM4E = TSMC partnership, standard + custom, 2027 product, customization expected to carry high gross margin.
- Notable absences in questioning: no one pressed the mobile managed NAND exit, the 12% tax rate (third straight below-guide quarter), or FCF compression ($803M vs >$1.9B) from the CapEx ramp.
Management credibility- The guidance machine is now consistently conservative: FQ4 came in above the high end on all three metrics for the second straight quarter, and the FQ1 guide (+580 bps GM) plus "GM up in FQ2" extends the pattern of under-promise/over-deliver. The FQ3 trough call and the FQ4/FQ1 inflection path have both validated.
- HBM commitments keep closing: >$6B run rate → ~$8B; four customers → six; share parity "sometime in H2 CY25" → declared on track for CQ3; and the biggest one — CY26 supply commitments, dodged by Sur last quarter, are now substantially done on HBM3E with HBM4 agreements promised "in the coming months." That promise is now the checkable item.
- Inventory promises kept again: "near target" exit delivered at 124 DIO (-15 days); DRAM below target as flagged.
- CapEx transparency improved: last quarter's refusal to frame FY26 CapEx resolved into a ~$18B net figure with a quarterly baseline and a gross/net/incentive decomposition. The $200B long-term plan now has a near-term funding number attached.
- Persistent quality flags: tax at 12% vs ~13% guided — third consecutive quarter of discrete-item EPS help; the 16.5% FY26 rate ends the tailwind but also confirms it existed. FCF fell to $803M on the CapEx catch-up — flagged last quarter, so not a surprise, but the "FCF strengthens in FQ1" promise is now on the record.
- Language watch: HBM TAM reaffirmed at $100B by 2030 despite Arcuri practically inviting a raise — either discipline or a ceiling signal; unverifiable from the transcript. CY26 HBM3E pricing direction deliberately withheld even while announcing agreements are done — the agreements could be flat-to-down; the transcript does not say. "Vast majority" and "almost all customers" leave a residual uncommitted slice.
- Evasions logged: CY26 HBM3E pricing direction (Arya); CY26 HBM supply quantification (Sankar); FY26 gross CapEx (Muse); FQ2 seasonality (Muse); lead-time quantification (Sur); HBM4 qualification status detail (not asked directly this quarter).
What changed versus the prior quarter- Pricing flipped from falling to rising: FQ3 was DRAM -LSD / NAND -HSD with "better than expected" as the beat driver; FQ4 is DRAM +low double digits / NAND +HSD. The pricing-execution narrative is now showing up in actual price prints, not just beats versus plan.
- CY26 HBM went from uncommitted to largely locked: last quarter Sur's demand-vs-supply question went unanswered; this quarter pricing agreements cover the vast majority of CY26 HBM3E supply, with full CY26 HBM sell-out promised in coming months. Pricing direction, however, remains undisclosed.
- HBM4 specs were raised mid-qualification: >2.8 TB/s and >11 Gbps vs >2 TB/s disclosed at sampling — management says customer performance requirements increased and Micron's product met them. HBM4E base die moved to TSMC partnership (from the in-house CMOS framing emphasized for HBM4).
- Margin trajectory steepened again: 39.0% → 45.7% actual → 51.5% guided → "up in FQ2." Last quarter's 42% guide was beaten by 370 bps.
- Demand outlooks raised broadly: servers +MSD→~+10%, traditional servers flat→+MSD, PCs LSD→+MSD, NAND bit demand low double digits→low-to-mid teens. DRAM held at high teens. The AI-agent-driven traditional server thesis is new language.
- CapEx stepped from "deliberately open" to ~$18B net FY26 (vs $13.8B), with a $4.5B quarterly baseline — a ~30% increase, funded against $3.7B FY25 FCF.
- Segment reporting arrived and is flattering: CMBU at 59% GM / 48% op margin vs MCBU 36% and AEBU 31% — the mix story is now quantified, and comparability to the old CNBU/SBU structure is broken as expected.
- Mobile managed NAND development ceased — a new portfolio-rationalization data point with no prior warning in the packet.
- Tax tailwind formally ending: 16.5% guided for FY26 vs the 12–13% actuals that assisted the last three quarters.
- FCF compressed: $803M vs >$1.9B, on the planned CapEx catch-up; FY26 framed as "significantly higher annual FCF" despite higher CapEx.
Bull case- A genuine pricing upcycle is now in the numbers: DRAM price +low double digits with bits +low teens; GM guided to 51.5% and "up in FQ2"; management itemizing structural supply constraints (D4/LP4 EOL constraining node migration, long lead times, expensive cleanroom, HBM silicon intensity) that limit CY26 supply response regardless of demand.
- HBM is de-risked through CY26: ~$8B run rate, six customers, share parity achieved, CY26 HBM3E largely priced and agreed, HBM4 specs raised to industry-leading (>2.8 TB/s, >11 Gbps) with production shipments CQ2 2026 and sell-out agreements expected within months. The single biggest bear argument from last quarter (CY26 unlocked) is substantially retired.
- Mix shift is now quantified and dramatic: data center at 56% of FY25 revenue at 52% GM; CMBU alone at 59% GM. Every point of mix shift from mobile/client (36% GM) to cloud is mechanically accretive, and management is steering bits that way.
- Demand breadth is widening: AI servers, traditional servers (AI-agent thesis), AI PCs + Windows 10 EOL, AI smartphones (one-third of CQ2 flagships at 12GB+), auto/industrial above forecast, HDD shortages helping NAND. Multiple independent demand vectors reduce single-point-of-failure risk.
- Technology execution is compounding: 1-gamma mature yields 50% faster than prior gen, first hyperscaler revenue, sole-source data center LPDRAM, first-to-market PCIe Gen6 SSDs, G9 QLC enterprise-qualified — cost and product leadership arriving simultaneously.
- Cash and balance sheet support the spend: $11.9B cash, $15.4B liquidity, debt reduced $900M, WAM 2033, FY26 FCF guided significantly higher even at ~$18B CapEx.
Bear case- CY26 HBM3E pricing direction is deliberately hidden: agreements are done, but management refused to say whether pricing is up or down. Locking "vast majority" of supply without disclosing price direction leaves open the possibility that CY26 HBM economics are flat-to-down versus current — the announcement optimizes for volume certainty, not necessarily margin.
- The guide embeds an extraordinary amount of good news: record revenue, 51.5% GM, record EPS, plus a verbal "up in FQ2" — expectations are now compounding quarterly, and the last two beats leaned on pricing better than management itself forecast. Forecasting gaps cut both ways.
- FCF is thin relative to the spending trajectory: $803M in FQ4 against a ~$18B FY26 CapEx plan; the "significantly higher FY26 FCF" promise requires the margin guide to hold while CapEx rises ~30%. Any pricing wobble compresses FCF fast at this spending rate.
- EPS quality: third straight quarter of below-guide tax (12% vs 13%); the 16.5% FY26 rate is a real headwind to EPS growth that models must absorb.
- NAND remains the weak leg: revenue -5% YoY, bits -MSD, price up only on mix; the mobile managed NAND exit is an implicit admission of inadequate returns in part of the portfolio. NAND's recovery is asserted ("improving substantially") more than evidenced.
- HBM4 execution risk is still live: qualifications are ahead, specs were raised mid-cycle (customer requirements increased), and the HBM4E base die moved to TSMC — a dependency shift whose rationale (capacity? capability? customer preference?) is unexplained. Custom HBM4E margin upside is a claim, not a delivered number.
- TAM plateau signal persists: $100B-by-2030 reaffirmed rather than raised despite hyperscaler capex escalation — management either sees attach-rate limits (Arcuri's original question, still unanswered) or is sandbagging; either way, no TAM support for multiple expansion from this call.
- Tariff risk remains unquantified and excluded from guidance.
Next-quarter watchlist- FQ1 delivery vs $12.5B / 51.5% / $3.75 — and whether the "GM up in FQ2" verbal commitment survives into formal guidance; any walk-back would break a multi-quarter credibility streak.
- CY26 HBM sell-out completion: HBM4 agreements "in the coming months" — confirm they close, at what pricing, and whether the residual uncommitted HBM3E slice fills. Any disclosure (or leak) on CY26 HBM3E pricing direction is the single most valuable missing data point.
- HBM4 qualification milestones through CQ4/CQ1 against the CQ2 2026 first-production-shipment commitment; customer count and platform breadth for HBM4 vs the six-customer HBM3E base.
- DRAM price trajectory: does +low double digits repeat, decelerate, or accelerate in FQ1; whether NAND price turns positive on fundamentals rather than mix; D4/LP4 shortage monetization in AEBU/MCBU prints.
- Inventory and lead times: DIO "at or better" than 124 with DRAM below target — check for lead-time extension disclosures or customer order-visibility commentary as evidence of tightening vs pull-in risk.
- FCF recovery: FQ1 FCF "strengthen" promise against ~$4.5B CapEx; FY26 "significantly higher FCF" math (needs OCF well above $20B annualized).
- Segment economics: second quarter of CMBU/CDBU/MCBU/AEBU disclosure — does CMBU's 59% GM hold as HBM4 ramps (trade-ratio cost headwind), and does MCBU's +12-pt GM jump prove durable or one-off?
- Tax normalization: 16.5% rate flowing into FY26 EPS models; any further discrete-item surprises.
- Mobile managed NAND exit: revenue impact quantification and whether the freed resources show up in data center SSD share gains; G9 QLC enterprise ramp evidence.
- HBM4E/TSMC structure: standard vs custom mix, margin claims, and whether the TSMC dependency introduces cost or capacity constraints for the 2027 product.
- Still unaddressed: China exposure, tariff quantification, buyback pace, HBM attach-rate ceiling, and the long-run GM terminal level — now a live question with 51.5% guided and no stated ceiling.
|
| Jun 25, 2025 | -0.98% | Q3 FY2025 | Read transcript briefingQuarter in one view- FQ3 print: revenue $9.3B (+15% QoQ, +37% YoY, record), GM 39.0% (+110 bps QoQ, +250 bps vs guide midpoint), EPS $1.91 (+22% QoQ, >200% YoY) — all three above the high end of guidance ($8.8B ±$200M / 36.5% ±100 bps / $1.57 ±$0.10). Murphy attributes the GM beat primarily to better-than-expected pricing in both DRAM and NAND, partially offset by higher consumer mix; prices still fell QoQ (DRAM low-single-digit, NAND high-single-digit) but less than planned.
- FQ4 guide is the real story: revenue $10.7B ±$300M (+15% QoQ, record), GM 42% ±100 bps, EPS $2.50 ±$0.15. Last quarter's verbal "up somewhat" on FQ4 GM converted into a ~300 bps sequential inflection — revenue weighted toward DRAM on "robust pricing execution, favorable product mix and continued cost improvements."
- HBM: ~+50% QoQ again, now at a >$6B annualized run rate (implies ~$1.5B in FQ3 vs >$1B in FQ2). 12-high yield ramp is running *faster* than 8-high did; 12-high shipment crossover expected in FQ4; share parity with DRAM share pulled in from "CQ4 run rate" to "sometime in H2 CY25," with Mehrotra allowing it could come earlier than previously stated.
- Inventory inflected hard: $8.7B, 139 DIO (-19 days QoQ, -$280M) — the promised FQ3 decline delivered in size. Company now expects to exit FY25 with tight DRAM inventory, significantly reduced NAND inventory, and DIO near target.
- FCF >$1.9B, highest in over six years; record $12.2B cash; net debt down to $3B. Announced a $200B, 20-plus-year U.S. investment plan ($150B manufacturing / $50B R&D), including a second Boise fab (ID2), Manassas expansion, and eventual U.S. HBM advanced packaging.
- CY25 DRAM bit demand raised again to high teens (from mid-to-high teens); NAND held at low double digits. HBM TAM *reaffirmed* at ~$35B CY25 (vs ~$18B CY24) — the first quarter in four without a TAM raise.
What management is focused on- The margin inflection narrative, now with numbers: Murphy walked the bridge explicitly — FQ3 beat on better pricing, FQ4 adds favorable mix (more DRAM than NAND, more data center than consumer) on top of a constructive market. Arya's "is 42% the new baseline" got a constructive non-answer: no FQ1 guide, but "gross margins can be up" despite leading-edge bit constraints.
- HBM execution credibility: repeated emphasis that 12-high yields are ahead of plan and ramping faster than 8-high, output ahead of plan, four customers in high-volume shipments across GPU and ASIC, and the AMD MI355X design-in announced at AMD's event. The share-parity timeline was voluntarily pulled forward.
- HBM4 positioning: samples delivered to multiple customers; >2 TB/s per stack, >60% bandwidth uplift, 20% lower power than HBM3E 12-high; built on "well-proven, cost-effective" 1-beta with an internally developed/manufactured CMOS logic base die; volume ramp CY26 "aligned with customers' plans." Trade ratio >3 for HBM4, rising toward 4 at HBM4E.
- D4/LP4 end-of-life as a pricing lever: EOL notices sent months ago, final shipments in 2–3 quarters, D4 now on allocation with "increasing shortages," LP4 shortages expected to follow. D4 revenue is a low-single-digit percentage of revenue in H2 FY25. Management is actively "driving price improvements" against constrained legacy supply and low distributor inventory.
- The $200B U.S. investment announcement: framed as customer- and administration-endorsed; ID1 first wafers H2 CY27, ID2 before the first New York fab, NY ground prep later this year. FY25 CapEx unchanged at ~$14B despite the headline number.
- Reorganization: June reorg into market-segment business units; from FQ4 Micron will report revenue, gross margin, and operating margin by the new segments — a disclosure upgrade that will also change comparability.
- Tariff pull-ins acknowledged for the first time: "may have been some tariff-related pull-ins by certain customers," impact "relatively modest" — a direct, if soft, answer to the question dodged last quarter.
Key numbers and quarter mechanics- FQ3: revenue $9.3B; DRAM $7.1B (76% of revenue, +51% YoY, +15% QoQ; bits +20%+, price down LSD on consumer mix); NAND $2.2B (23%, +4% YoY, +16% QoQ; bits up mid-20s%, price down HSD).
- Business units (old structure, last report under it): CNBU $5.1B (+11% QoQ, record, HBM ~+50%); SBU $1.5B (+4%, consumer-driven); Mobile $1.6B (+45%, customer inventories reduced + DRAM content growth); EBU $1.2B (+20%, industrial/consumer embedded).
- Margins/EPS: GM 39.0% (+110 bps); OpEx $1.1B (+$87M QoQ, in line; R&D and labor); operating income $2.5B, op margin 26.8% (+190 bps QoQ, +13 pts YoY); tax $306M at 12.3% — below the ~14% guide again on one-time discrete items (second straight quarter of tax-assisted EPS); EPS $1.91.
- Cash/balance sheet: OCF >$4.6B; CapEx $2.7B (below plan — Murphy flagged FQ4 catch-up to hold ~$14B FY25); FCF >$1.9B; cash/investments record $12.2B; liquidity $15.7B; debt $15.5B after refinancing $900M 2027 notes into $1.7B of FY33/FY36 paper; net debt $3B; weighted average maturity 2032.
- Inventory: $8.7B, 139 DIO (-19 days) on strong bit shipments in both DRAM and NAND. FQ4 exit: tight DRAM, significantly reduced NAND, DIO "near target levels."
- FQ4 guide: revenue $10.7B ±$300M; GM 42% ±100 bps; OpEx ~$1.2B ±$20M (R&D on future nodes + HBM); tax ~13%; EPS $2.50 ±$0.15 on ~1.15B shares. FY26 tax rate guided to high teens (Singapore global minimum tax). Tariff impacts excluded from guidance.
- Demand frames: CY25 DRAM bit demand high teens (raised); NAND low double digits (unchanged); medium-term mid-teens CAGR for both; Micron bit supply growth below industry demand growth for non-HBM DRAM and NAND. Server units +MSD CY25; PC and smartphone units LSD.
- NAND mechanics: 10% structural wafer capacity reduction exiting FY25 reaffirmed; underutilization has fallen as capacity came out, but part of NAND remains underutilized; leading-edge NAND fully utilized; G9 conversions paced to demand.
- Notable absence: no underload/under-absorption quantification this quarter — last quarter's flagged FQ4 inventory-flush headwind was not mentioned in the GM bridge. Whether it silently cleared or is embedded in the 42% guide is not stated.
Product and launch scorecard- HBM3E 12-high: yield ramp faster than 8-high, output ahead of plan, shipment crossover (majority of HBM shipments) in FQ4 — the "vast majority in H2 CY25" commitment is tracking. Designed into AMD Instinct MI355X (36GB), adding a named second GPU platform beyond NVIDIA GB300.
- HBM customer base: now four customers in high-volume shipments, spanning GPU and ASIC platforms (up from three last quarter). CY25 sold out; CY26 supply/pricing negotiations explicitly *not* yet committed — Sur's question on whether CY26 demand exceeds supply capability got no direct answer.
- HBM4: sampled to multiple customers; >2 TB/s, >60% bandwidth uplift, 20% lower power than 12-high HBM3E; 1-beta die + in-house CMOS base die; volume ramp CY26; qualifications still ahead. Trade ratio >3 (vs ~3 for HBM3E), toward 4 at HBM4E — slightly softer framing than last quarter's ">4:1 at HBM4E."
- High-cap DIMM + LP server: "multiple billions" of FY25 revenue, 5x YoY; sole-source position in server LP maintained. Still no split between DIMM and LP, and no SOCAMM-specific update this quarter.
- 1-gamma DRAM: yields ramping ahead of 1-beta's record pace; first qualification samples of 1-gamma LP5X shipped (industry-first claim; 25%+ faster AI recommendations, 20% lower power; 2026 flagship phones). 30% bit density, >20% power, up to 15% performance vs 1-beta.
- Data center SSD: record share for the third consecutive quarter (CQ1), now the #2 brand in DC SSD per third-party data; 9550 completed additional OEM qualifications (on NVIDIA GB200 NVL72 recommended vendor list); 6550 ION 60TB ramping. Management expects H2 CY25 DC SSD demand better than H1 after the digestion period.
- Client SSD: record client SSD share in CQ1; G9 QLC 2TB SSD with "Adaptive Write Technology" (4x faster writes, TLC-equivalent for most consumer use cases) to be announced the day after the call.
- Mobile: G9-based UFS 4 design win ramped to high-volume production; quality awards from 7 smartphone OEMs; AI content story (12GB+ vs 8GB average) reiterated.
- Auto/industrial: industry-first 1-beta dual-channel LP5 at 9.6Gbps production-ready; industrial growth resuming on AI/factory automation; legacy D4/LP4 (1-alpha) to be supported for longevity segments for years while high-volume segments EOL in 2–3 quarters.
Sell-side read-through- Nine questioners; tone congratulatory-to-constructive, centered on HBM durability and the margin path. No questions on China, buybacks (Moore got a passing "opportunistic buyback" mention), or the new segment structure.
- Arcuri (UBS) asked the strategic question — does HBM TAM scale with the accelerator TAM or hit an attach-rate asymptote? Mehrotra answered with the value roadmap (12-high → HBM4 → HBM4E → customization in '28) and CY26 HBM bit growth "significantly exceeding" DRAM, but did not address attach-rate limits directly.
- Arya (BofA) extracted the cleanest GM disclosure: the FQ3 beat was "defining factor: better-than-expected pricing," and FQ4's step-up is mix (DRAM > NAND, data center > consumer) layered on a better market. His "new baseline?" probe got the first FQ1 color: leading-edge bit constraints, mix toward higher-value DRAM/NAND, "gross margins can be up."
- Muse (Cantor) pinned the HBM run rate (>$6B annualized) and got Mehrotra to volunteer that share parity could arrive *earlier* than the prior end-of-year framing — the most incremental HBM data point of the call.
- O'Malley (Barclays) pressed CY26 normalized HBM share — declined; got instead the "fungible 1-beta capacity" flexibility argument. His NAND utilization question produced the useful disclosure that underutilization has fallen with the structural cut, part of NAND remains underutilized, and leading edge is fully utilized.
- Sur (JPMorgan) asked the sharpest HBM question: is CY26 supply fully committed, and does customers' CY26 forecasted demand exceed Micron's supply capability? Mehrotra reaffirmed CY25 sold out but gave no commitment or demand-vs-supply answer for CY26 — "we are still in the middle of '25." Notable evasion given last year's "sold out with pricing locked" posture at the same point in the cycle.
- Moore (Morgan Stanley) tested the ~35%-of-revenue CapEx frame for FY26 — Murphy declined to confirm a ratio, citing greenfield needs, lumpy grant/construction timing, and FCF-positive FQ4. FY26 CapEx magnitude left open against a $200B headline plan.
- Sankar (TD Cowen) got the HBM4 trade ratio (>3, toward 4 at 4E) and a refusal to differentiate GPU vs ASIC HBM margins.
- Caso (Wolfe) forced the tariff pull-in clarification: modest impact, healthy customer inventories, constructive H2 signals. This closes last quarter's open evasion, though "relatively modest" is unquantified.
- Rakesh (Mizuho) confirmed HBM4 quals are still ahead (early units only) and got the H2-weighted DC SSD recovery framing.
Management credibility- Delivered a clean beat-and-raise with the margin inflection arriving on schedule: FQ3 GM 39.0% vs 36.5% guide; FQ4 guided 42% — the "up somewhat" verbal commitment from last quarter not only survived but came in far above what the language implied. The FQ3 trough call (36.5% guided, 39.0% actual) is validated.
- HBM commitments keep being met or pulled forward: >$1B FQ2 → ~$1.5B FQ3 (>$6B run rate), 12-high yields ahead of plan and faster than 8-high, fourth customer added, share parity moved from "CQ4 run rate" to "sometime in H2 CY25" with upside language. This is the strongest credibility thread on the call.
- Inventory promise kept: DIO guided down in FQ3; delivered -19 days to 139. The "near target by FY25 exit" language is now backed by two quarters of trajectory.
- Pricing narrative is now evidenced, not just asserted: last quarter's CQ2 price-action intent showed up as better-than-expected DRAM and NAND pricing in FQ3 and "robust pricing execution" in the FQ4 guide. The D4 allocation/shortage disclosure adds a concrete, checkable legacy-pricing data point.
- EPS quality flag persists: tax came in at 12.3% vs ~14% guided on one-time discrete items — the second consecutive quarter tax helped EPS. FQ4 guides ~13%, and FY26 steps to high teens; the tax tailwind is ending.
- Language slippage to monitor: Mehrotra said the HBM share goal would be achieved "in second half of '26" in the O'Malley answer — context (H2 CY25 stated twice elsewhere) indicates a verbal slip, not a guidance change, but it is worth confirming. HBM4E trade ratio framed as "toward 4" versus last quarter's ">4:1" — slightly softer. HBM TAM held at ~$35B after three straight raises — appropriately conservative or a plateau signal; unverifiable from the transcript.
- Evasions logged: CY26 HBM supply commitment and demand-vs-supply balance (Sur, unanswered); CY26 normalized HBM share (O'Malley, declined); FY26 CapEx level/ratio (Moore, declined); GPU vs ASIC HBM margin split (Sankar, declined); tariff pull-in magnitude ("relatively modest," unquantified); no underload quantification despite last quarter's flagged FQ4 headwind.
What changed versus the prior quarter- The margin trough is behind, with a vengeance: GM went 37.9% → 39.0% actual → 42% guided, versus the prior path of 37.9% → 36.5% guided → "up somewhat." The FQ4 inventory-flush underload and start-up costs that were supposed to weigh on FQ4 were not mentioned — either absorbed in the guide or overtaken by pricing.
- HBM stepped from >$1B to a >$6B run rate (~$1.5B/quarter), a fourth customer entered high-volume shipments, AMD joined NVIDIA as a named platform customer, and the share-parity timeline moved from "CQ4 2025 run rate" to "sometime in H2 CY25" with earlier-than-that upside.
- HBM TAM was reaffirmed (~$35B CY25, ~$18B CY24) rather than raised — the first pause after three consecutive raises; CY26 framing shifted to bit growth "significantly exceeding" DRAM industry growth.
- CY25 DRAM bit demand raised to high teens (from mid-to-high teens); NAND unchanged at low double digits.
- Tariff pull-ins moved from dodged question to acknowledged factor — "may have been some," "relatively modest."
- D4/LP4 EOL became a disclosed pricing event: EOL notices out, final shipments in 2–3 quarters, D4 on allocation, D4 revenue now low-single-digit percent in H2 FY25 (prior quarter framed DDR4+LP4 at ~10% of revenue for the rest of FY25 — the bases differ, but the legacy exposure is shrinking fast).
- Inventory inflected: 158 → 139 DIO, with FY25 exit now guided near target — the <120-day DRAM target language was replaced by "tight DRAM inventories" and "near target" overall DIO.
- Capital story escalated: $200B 20-year U.S. plan announced ($30B incremental), ID2 added, U.S. HBM packaging flagged post-wafer-scale; FY25 CapEx still $14B, but FY26 left deliberately open.
- Reporting structure changes in FQ4: new market-segment BUs with revenue, GM, and op margin disclosure — better visibility, broken comparability.
- Tax tailwind ending flagged: FY26 high-teens rate (Singapore global minimum tax) now formally on the script.
Bull case- The margin inflection is real and steep: 39.0% → 42% guided with the bridge itemized (pricing execution, DRAM/data center mix, cost improvement), and management signaling FQ1 GM "can be up" even with leading-edge bit constraints — scarcity is now working in Micron's favor.
- HBM is compounding ahead of every marker: ~50% QoQ growth for a second straight quarter, >$6B run rate, 12-high yields beating the 8-high learning curve, four customers across GPU and ASIC, AMD MI355X design-in, share parity pulled forward, and HBM4 already sampled with a cost-advantaged 1-beta + in-house base-die architecture.
- Pricing power is broadening beyond HBM: better-than-expected DRAM and NAND pricing in FQ3, "robust pricing execution" guided for FQ4, D4 on allocation with shortages increasing, LP4 shortages expected, and distributor inventories low — the legacy EOL is creating a second pricing lever.
- Demand raised again (DRAM high teens) with supply discipline intact: Micron bit growth below industry demand in non-HBM DRAM and NAND; HBM trade ratios rising toward 4 structurally tighten leading-edge supply; NAND's 10% structural capacity cut is holding.
- Cash generation is now substantial: >$1.9B FCF (best in 6+ years), record $12.2B cash, net debt $3B, maturities termed to 2032 — the $14B CapEx program and the $200B long-term plan are fundable without balance-sheet strain.
- Mix shift to data center is durable: data center revenue more than doubled YoY, DC DRAM at a fourth consecutive record, high-cap DIMM + LP at 5x YoY and sole-source LP position, #2 DC SSD brand — the consumer drag that defined the FQ2/FQ3 trough is fading as a mix factor.
Bear case- The beat leaned on pricing that management didn't fully predict: the FQ3 GM upside was "better-than-expected pricing" — good news, but it means the guide missed the market turn, and the same forecasting gap cuts both ways if pricing rolls over. Prices still fell QoQ in both DRAM (LSD) and NAND (HSD).
- CY26 HBM is not locked: unlike a year ago ("sold out with pricing locked"), management gave no commitment on CY26 supply or pricing and declined to say whether customer demand exceeds supply capability — with HBM4 qualifications still ahead and competitors' qualification troubles (referenced by O'Malley) a variable Micron can't control.
- Consumer mix is still dilutive and NAND still weak: consumer drove the FQ3 price declines and the SBU/Mobile revenue beats; NAND price fell high-single-digits even in a record-revenue quarter, and part of NAND remains underutilized. The 42% guide assumes mix improvement continues.
- Tariff pull-ins are acknowledged but unquantified: "relatively modest" is an assertion; if H2 demand includes pulled-forward volume, the FQ1 "bit constraints" could resolve into an air pocket rather than scarcity pricing.
- FQ4 guide embeds a lot of good news at once: +15% sequential revenue, +300 bps GM, and an EPS guide that still carries a below-statutory 13% tax rate before the FY26 high-teens step-up. Execution room for disappointment is thin.
- CapEx escalation risk is unpriced in guidance: a $200B headline plan, ID2, Manassas, U.S. packaging, and New York — with FY26 CapEx deliberately left open and Murphy refusing the 35%-of-revenue frame. FCF is strong now, but the spending trajectory is one-directional.
- Disclosure transition risk: the FQ4 segment reorg will reset comparability at exactly the moment margins inflect — watch for reduced visibility into the old CNBU/SBU economics during the transition.
Next-quarter watchlist- FQ4 delivery vs the $10.7B / 42% / $2.50 guide — and the first FQ1 FY26 framework: does "gross margins can be up" hold, and how do leading-edge bit constraints resolve (scarcity pricing vs volume shortfall)?
- HBM slope and share math: FQ4 HBM revenue vs the >$6B run rate; confirmation of 12-high shipment crossover; whether share parity is declared achieved in H2 CY25; any CY26 LTA signings or pricing commitments — the single biggest open item after Sur's unanswered question.
- HBM4 qualification progress: customer qual milestones through H2 CY25 against the CY26 volume ramp; any update on the trade-ratio framing ("toward 4" vs prior ">4:1 at HBM4E").
- Pricing evidence: whether DRAM and NAND contract pricing keeps improving into CQ3/CQ4; whether D4 allocation and LP4 shortages translate into reported legacy pricing; NAND price trajectory after two quarters of declines.
- Inventory exit: FY25 exit DIO "near target," tight DRAM, significantly reduced NAND — check the actual exit numbers and whether the unmentioned underload/start-up costs resurface in the FQ4 GM bridge or FQ1 commentary.
- New segment reporting: first revenue/GM/op-margin disclosure under the reorganized BUs — what it reveals (or obscures) about data center vs consumer profitability.
- Tariff pull-in payback: any quantification of pull-in magnitude and whether H2 customer demand holds after the acknowledged pre-buying.
- CapEx and FY26 setup: FQ4 CapEx catch-up to ~$14B FY25; first FY26 CapEx signal against the $200B plan; FY26 tax step-up to high teens flowing into models.
- DC SSD recovery: H2 CY25 "better than H1" claim — check bit growth resumption and whether the #2 brand share holds; G9 QLC client SSD launch reception.
- Still unaddressed: China exposure, buyback pace, HBM capacity sizing for CY26, GPU-vs-ASIC HBM economics, and the long-run GM ceiling question — now partially answered by the 42% guide but with no stated terminal view.
|
| Mar 20, 2025 | -8.04% | Q2 FY2025 | Read transcript briefingQuarter in one view- FQ2 print: revenue $8.1B (-8% QoQ, +38% YoY), GM 37.9% (-160 bps QoQ), EPS $1.56 (above guidance high end). The guided air pocket landed as advertised: DRAM revenue -4% QoQ (bits down high-single-digit, price up mid-single-digit on mix), NAND revenue -17% QoQ (bits modestly higher, price down high-teens).
- HBM crossed $1B quarterly revenue, +50% QoQ, ahead of plan — the first hard HBM number management has ever given, and it lands above the "a little bit high" $800–900M FQ1 estimate they steered down last quarter. High-cap DIMM + LP data center portfolio also crossed $1B combined.
- FQ3 guide: revenue $8.8B ±$200M (record at midpoint), GM 36.5% ±100 bps (down again), EPS $1.57 ±$0.10. Bits grow in both DRAM and NAND; margin declines on consumer mix and NAND underutilization. FQ4 GM guided "up somewhat" verbally, not numerically.
- HBM TAM raised again: >$35B CY25 (from >$30B last quarter, >$25B before that) — third raise in three quarters. Share parity with DRAM share now pinned to a CQ4 2025 run rate; CY25 HBM sold out; CY26 agreements in discussion.
- Inventory went the wrong way as pre-flagged: $9.0B, 158 DIO (+9 days QoQ) — back to the FQ4 FY24 level. DIO guided down in FQ3; DRAM below the 120-day target by FQ4 reiterated.
- NAND structural action deepened: wafer output down mid-teens from prior levels, >10% structural capacity reduction exiting FY25 vs FY24, equipment reuse for node conversions, and an explicit intent to take price action in CQ2 on both NAND and DRAM.
What management is focused on- Reframing the margin trough as mix, not deterioration: Murphy and Mehrotra repeatedly attribute the FQ2/FQ3 margin declines to consumer-oriented volume mix and NAND pricing, partially offset by HBM — with the message that conditions have *improved* since the mid-February conference comments and the guide reflects that.
- HBM leadership and the 12-high transition: volume production begun, "vast majority" of HBM shipments to be 12-high in H2 CY25, 8-high designed into NVIDIA GB200 and 12-high into GB300 (first time NVIDIA is named on the call), third large HBM3E customer began volume shipments in FQ2, HBM4 (>60% bandwidth uplift) ramping CY26 with trade ratio exceeding 4:1 at HBM4E.
- NAND supply discipline as an industry project: Micron underutilizing fabs, cutting capacity structurally, slowing Gen9 ramp deliberately, and publicly counting on "supply actions announced by NAND companies" plus a CQ2 pricing inflection.
- Consumer recovery confirmed, not just predicted: Mehrotra stated the spring inventory normalization "is turning out to be the way we had projected," with smartphone bit growth resuming in FQ3 and PC growth H2-weighted (Windows 10 EOL October 2025, AI PC content 16GB vs 12GB average).
- Capital program milestones: $14B FY25 CapEx unchanged; Singapore HBM advanced packaging groundbreaking (capacity from CY27); Idaho fab construction milestone triggered first CHIPS grant disbursement, output from FY27; Taiwan DRAM test facility named.
- Tariffs pre-empted: minimal direct exposure as importer of record; costs to be passed through; excluded from guidance.
Key numbers and quarter mechanics- FQ2: revenue $8.1B; DRAM $6.1B (76% of revenue, +47% YoY, -4% QoQ; bits -HSD%, price +MSD% on mix); NAND $1.9B (23%, +18% YoY, -17% QoQ; bits modestly up, price down high-teens). NAND consumer bits came in *above* plan — the mix problem.
- Business units: CNBU $4.6B (+4% QoQ, 57% of revenue, third straight record, driven by HBM +50%); SBU $1.4B (-20%, DC customer storage digestion + NAND pricing); Mobile $1.1B (-30%, customer inventory work-down); EBU $1.0B (-3%, auto inventory initiatives).
- Margins/EPS: GM 37.9% (-160 bps); OpEx $1.0B flat (R&D below plan on qualification timing); operating margin 24.9% (-260 bps); EBITDA $4.1B (50.7% margin); tax 10.7% (one-time items); EPS $1.56 vs $1.79 FQ1.
- Cash/balance sheet: operating cash flow >$3.9B; CapEx $3.1B net of incentives; FCF $857M; cash $9.6B; debt $14.4B (avg maturity 2032 after $1B 10-yr note + $1.7B term loan refi); revolver upsized to $3.5B post-quarter.
- FQ3 guide: revenue $8.8B ±$200M; GM 36.5% ±100 bps; OpEx ~$1.13B; tax ~14% (FQ3 and FQ4); EPS $1.57 ±$0.10; CapEx >$3B. FY25 OpEx +10%+; FY25 CapEx ~$14B unchanged.
- Underload mechanics shifted: less FQ3 period cost than originally projected because capacity was structurally reduced — more of the charge now flows through inventory and hits FQ4 as it clears; under-absorption weighs on GM into FY26. FQ4 also sees the start of DRAM new-node/Idaho construction start-up costs, growing through FY26.
- Cost frames given: FY25 all-in DRAM cost flattish; NAND front-end cost down low-double-digits.
- Demand: CY24 DRAM bit demand high-teens (in line); CY24 NAND ~10% (below prior low-double-digits); CY25 DRAM raised to mid-to-high teens (from mid-teens), NAND low-double-digits; Micron supply growth below industry demand growth in both; bit share maintained in both.
- Inventory: 158 DIO (+9 days); down in FQ3; DRAM below 120-day target by FQ4; DIO declining through CY25.
Product and launch scorecard- HBM3E 8-high: >$1B revenue, +50% QoQ, shipments ahead of plan; designed into NVIDIA GB200; 30% power reduction vs competition claimed. Now a quantified, delivered product line.
- HBM3E 12-high: volume production started; 20% power advantage over competing *8-high* with 50% more capacity; designed into GB300; vast majority of HBM shipments in H2 CY25; premium pricing over 8-high and "accretive to DRAM margins" per Mehrotra; early yield ramp acknowledged but not quantified.
- HBM4: >60% bandwidth vs HBM3E, volume ramp CY26, timing "aligned to customer requirements"; trade ratio >4:1 at HBM4E — a new disclosure tightening the leading-edge supply-tightness thesis.
- HBM customer base: third large customer in volume shipments in FQ2; CY25 sold out; CY26 LTAs in discussion.
- LP/SOCAMM: SOCAMM developed with NVIDIA for GB300 — first explicit NVIDIA collaboration on LP; LP cuts memory power >2/3 vs D5 in AI servers; high-cap DIMM + LP combined >$1B in FQ2, "multibillion" FY25 reiterated. Still no split between the two.
- 1-gamma DRAM: first EUV node; industry-first 1-gamma D5 shipments last month; 20% lower power, 15% better performance, >30% bit density vs 1-beta; 16Gb 1-gamma D5 sampled to PC clients.
- Data center SSD: demand moderated in FQ2 (customer inventory), bit growth returning "in the months ahead"; record DC SSD share in CQ4 2024 per industry reports; 9550 on NVIDIA GB200 NVL72 approved vendor list, qualified at multiple customers; G8 QLC NAND qualified in Pure Storage's 150TB DirectFlash module — a notable third-party design win.
- Client/mobile/auto: Gen9-based 4600 client SSD launched ("fastest in the world" claim); 2650 mainstream SSD qualified at multiple OEMs; LP5X + UFS 4.0 in Samsung Galaxy S25 high end; industry-first 1TB G9 UFS 4.1 sampling; first automotive LP5X at 9.6Gbps production-ready; 4150 first automotive-qualified enterprise SSD, sampling.
Sell-side read-through- Seven questioners; tone constructive, focused on the margin path. No questions on China, buybacks, or capital returns again.
- Sur (JPMorgan) got the key forward margin commitment: FQ4 GM "up somewhat" from FQ3, with tailwinds (market improvement, HBM mix) and headwinds (NAND under-absorption shifting into FQ4 inventory clear-out, start-up costs) itemized. Also extracted the CY25 DRAM bit demand raise rationale (consumer inventory normalization + AI content + HBM).
- Arcuri (UBS) tried to split FQ3 revenue/bits between DRAM and NAND — refused; got only "bias of growth" toward DRAM. His "when do we get a clean gross margin" question produced the FY25 cost frames (DRAM flat, NAND down low-double-digits) but no clean-margin timeline.
- Sankar (TD Cowen) asked whether recent price improvement is real demand or tariff pull-ins — Mehrotra answered with the demand narrative and did not address pull-ins directly, a soft evasion. His 12-high yield question got a candid "early stages, yield ramp" acknowledgment plus the premium/accretive framing.
- Moore (Morgan Stanley) mis-stated DIO as 153 (corrected: 158) and pressed the feasibility of reaching <120 days on DRAM by FQ4 — Murphy confirmed the target without giving the volume mechanics.
- Muse (Cantor) extracted the most useful new mechanics: underload shifts from FQ3 period cost into FQ4 inventory flush; under-absorption persists into FY26; Idaho start-up costs grow through FY26. Also got HBM TAM phasing: H2-weighted on 12-high premium and customer-base expansion.
- Caso (Wolfe) confirmed DDR4/LP4 ~10% of revenue for the rest of FY25 (restated, not new) and pressed HBM capacity scalability into CY26 — Mehrotra declined to project CY26 share but asserted capacity keeps growing.
- Danely (Citi) asked the sharpest question: same revenue as three years ago but GM ~10 points lower — is 50%+ GM dead? Mehrotra blamed NAND industry imbalance and pointed to mix shift and supply discipline, offering no numerical path back — the structural-margin question remains open.
Management credibility- Delivered on the pre-announced bad quarter: FQ2 GM -160 bps vs the ~-100 bps callback framing — worse than the February signal, though EPS beat on lower R&D and tax one-timers. The miss-vs-callback delta is worth noting: mix and NAND pricing deteriorated more than flagged.
- HBM disclosure finally arrived and validates the ramp: >$1B, +50% QoQ, ahead of plan — consistent with last quarter's "a little bit high" steer on $800–900M for FQ1 (implies FQ1 was ~$650–700M). The negation-based disclosure regime is over; this is a credibility positive.
- Spring normalization claim now asserted as observed: "turning out to be the way we had projected" — supported by FQ3 bit growth in both DRAM and NAND and mobile resumption, but consumer *pricing* evidence is still pending (CQ2 price inflection is a promise, not a fact).
- Underload guidance moved: last quarter Murphy framed FQ3 underload as "closer to 100 bps" of period cost; this quarter less FQ3 period cost than projected, with the burden shifted to FQ4 via inventory. The total cost didn't disappear — it moved. FQ3 GM guide of 36.5% (-140 bps) is nonetheless below what "improved conditions" language might imply.
- Share-parity language tightened again: "run rate basis in calendar Q4 2025" — more precise than last quarter's "H2 CY25," and now against a >$35B TAM. Three TAM raises in three quarters with the share target maintained raises the execution bar each time.
- Evasions logged: FQ3 DRAM/NAND revenue split (refused), tariff pull-in question (unanswered), path back to 50% GM (no numbers), CY26 HBM share (declined), HBM capacity sizing (still none), 12-high yield specifics (acknowledged qualitatively only).
What changed versus the prior quarter- HBM went from unquantified to >$1B/quarter, +50% QoQ, ahead of plan — the single biggest change; the ramp-slope question ("flattens from here") is resolved upward.
- HBM TAM raised to >$35B (from >$30B); share parity now a CQ4 2025 run-rate commitment; NVIDIA named for the first time (GB200 8-high, GB300 12-high, SOCAMM collaboration).
- CY25 DRAM bit demand raised to mid-to-high teens (from mid-teens); CY24 NAND revised down to ~10%.
- NAND moved from wafer-start cuts to structural capacity reduction: >10% lower wafer capacity exiting FY25 vs FY24, equipment reuse, delayed node transitions — and an explicit CQ2 price-increase intent on both NAND and DRAM, which is new language.
- Underload timing re-phased: less FQ3 period cost, more FQ4 inventory flush; under-absorption now acknowledged into FY26; Idaho/node start-up costs begin FQ4.
- FQ3 guided to a record $8.8B with GM down again to 36.5% — revenue recovers before margin does; FQ4 "up somewhat" is the first forward margin inflection language.
- Inventory: 158 DIO as pre-flagged, with the DRAM <120-day FQ4 target now explicitly confirmed under questioning.
- Tariffs entered the script for the first time: minimal direct exposure, pass-through intent, excluded from guidance.
- Consumer recovery shifted from forecast to claimed observation, with smartphone bit growth resuming in FQ3.
Bull case- HBM is now a proven, quantified growth engine: >$1B and +50% QoQ against a >$35B TAM, sold out for CY25, CY26 LTAs in negotiation, third customer ramping, 12-high premium and GB300 design-in setting up H2 — share parity at a CQ4 run rate implies a steep second-half revenue trajectory.
- Record FQ3 revenue guide with bits up in both DRAM and NAND confirms the FQ2 decline was the consumer/NAND air pocket, not a demand break; CNBU at 57% of revenue and three straight records anchors the mix story.
- Leading-edge DRAM tightness is structural and intensifying: HBM 3:1 trade ratio rising to >4:1 at HBM4E constrains non-HBM supply; management is confident enough to telegraph CQ2 price increases in both DRAM and NAND.
- NAND self-help is now structural, not cyclical: >10% permanent capacity reduction, equipment reuse, delayed transitions, CapEx discipline — plus industry-wide supply actions management says will improve dynamics.
- Margin inflection is mapped: FQ3 trough at 36.5%, FQ4 "up somewhat," with HBM/12-high premium, high-value mix, and consumer normalization as tailwinds; DRAM cost flat and NAND cost down low-double-digits in FY25.
- Balance sheet supports the build: $857M FCF in a down quarter, $12.1B liquidity, debt termed out to 2032, CHIPS funding flowing, $14B CapEx fully funded.
Bear case- Margins keep falling even as revenue recovers: 37.9% → 36.5% guided, with under-absorption now pushed into FQ4 and start-up costs layering in through FY26 — Danely's "same revenue, 10 points lower GM" question got no numerical answer, leaving the structural margin ceiling genuinely open.
- The CQ2 pricing inflection is asserted, not evidenced: NAND price fell high-teens in FQ2; the plan to "take price action" depends on competitor discipline Micron cannot control and refused to speculate on.
- Consumer mix is dilutive at the worst time: FQ2 NAND consumer bits came in *above* plan, dragging mix; FQ3 GM embeds higher consumer volumes again — the recovery is arriving first in the lowest-margin volume.
- Inventory is at 158 days with $9B on the balance sheet and the <120-day DRAM target requires a sharp two-quarter drawdown that Murphy declined to mechanize; NAND inventory health is materially worse and unquantified.
- 12-high execution risk is real: early yield ramp acknowledged, "vast majority" of H2 shipments must transition, and the CQ4 share-parity commitment now sits against a TAM that has been raised three times — any yield or qualification slip compresses the window.
- Tariff pull-in question went unanswered: if recent price firmness is partly pull-in rather than end demand, the CQ2 inflection could disappoint; guidance excludes tariff effects entirely.
- EPS quality note: the FQ2 beat leaned on below-plan R&D timing and one-time tax items; the tax rate steps up to ~14% in FQ3/FQ4.
Next-quarter watchlist- FQ3 delivery vs guide: $8.8B revenue, 36.5% GM, and the promised bit growth in both DRAM and NAND — plus the first FQ4 framework: does "up somewhat" survive contact with the inventory-flush underload and start-up costs?
- CQ2 pricing inflection: evidence that DRAM and NAND price increases actually stick — contract pricing commentary, spot trends, and whether competitors' NAND supply actions hold.
- HBM slope and 12-high transition: FQ3 HBM revenue vs the >$1B base; 12-high yield/ramp progress; progress on CY26 LTA signings; any fourth-customer disclosure; whether the CQ4 share-parity target is reiterated.
- Inventory drawdown mechanics: DIO decline in FQ3 as guided, and the path to <120 days DRAM by FQ4 — watch whether NAND DIO improvement lags as flagged.
- NAND structural reduction: confirmation the >10% capacity cut is on schedule, Gen9 ramp pacing, and DC SSD bit growth resumption "in the months ahead" as promised.
- Consumer recovery quality: smartphone bit growth in FQ3, PC order patterns into H2 (Windows 10 EOL, AI PC attach), and whether consumer pricing follows volume.
- Tariff developments: any quantification of exposure, pull-in behavior, or pass-through mechanics now that the topic is on the script.
- SOCAMM/LP revenue evidence: the >$1B combined high-cap DIMM + LP milestone is new — watch for a split or growth trajectory as GB300 ramps.
- Still unaddressed: China exposure, buyback pace vs the $14B CapEx program, DRAM-vs-NAND margin split, and the long-run GM ceiling question Danely raised.
|
| Dec 18, 2024 | -16.18% | Q1 FY2025 | Read transcript briefingQuarter in one view- This is the post-earnings analyst callback (Sadana, Bhatia, Murphy — no Mehrotra), so detail is incremental to the main call; the headline facts referenced: FQ2 gross margin guided down ~100 bps vs FQ1, "almost all NAND related" (mix, pricing, lower shipments); FQ2 bits: DRAM down sequentially, NAND down meaningfully, with NAND the majority of the sequential revenue decline.
- FQ3 headwind quantified: underload charges from NAND wafer-start cuts begin hitting in FQ3 as period costs plus higher-cost inventory — Murphy framed the impact as "closer to 100 bps" (when offered a 100–200 bps range). NAND market challenges extend into CQ1, overlapping FQ3.
- Data center now >50% of revenue, +40% QoQ and +400% YoY in FQ1 — the offset to consumer weakness. Data center SSD had a near-term moderation ("lumpiness") after several record quarters; bit growth resumes FQ3.
- HBM TAM raised again: >$30B for CY25 (from >$25B last quarter, >$20B before that). HBM share = DRAM share (low-20s) now pinned to second half of calendar 2025. Sadana said Hari's $800–900M FQ1 HBM revenue estimate is "a little bit high" — the closest thing to an HBM quantification this quarter, and it came via negation.
- Inventory: FQ1 DIO improved to below 150 days (from 158), but FQ2 DIO and absolute dollars rise again on the volume decline; DRAM inventories end FY25 below target levels; NAND improves FQ2→FQ3→FQ4 but stays less healthy than DRAM.
- Demand cuts: CY24 PC unit forecast cut to flattish YoY (refresh cycle pushed into 2025); CY25 DRAM bit demand held at mid-teens, NAND at low-double-digits; Micron expects to ship in line with industry / stable bit share in both for CY24–25.
What management is focused on- Isolating the problem as NAND + consumer, protecting the data center narrative: Sadana repeatedly stressed "most of the impact is limited to consumer-oriented segments" and that the data center revenue trajectory through FY25/CY25 "remains very robust." The FQ2 margin delta is framed as only 100 bps *despite* the headwinds because of mix (HBM, high-cap DIMMs, LP) and data center growth.
- Decisive NAND supply action: wafer-start reductions taken "immediately" to align Micron's supply with demand; NAND CapEx already cut — NAND's share of CapEx is now below its share of revenue. Management refused to speculate on competitor supply discipline (Chin's question).
- HBM ramp mechanics and credibility: emphasis on the 8-high ramp exceeding prior expectations, a "scalable process" transitioning to 12-high through CY25, capacity added "week-over-week," and the reminder that Micron started from essentially zero in HBM3 — framing the ramp slope as a build-out story, not a demand story.
- LP-in-data-center leadership claim: Sadana claimed Micron is "pretty much" the entire market for LP DRAM in data center today ("pioneering memory supplier," "pole position"), tied to RAS innovation and the NVIDIA GB200/Blackwell cycle (Rakers' framing; Sadana did not name NVIDIA). LP + high-cap DIMMs together = "multi-billions," separate from HBM's multiple billions and DC SSD's multiple billions.
- Consumer recovery timing unchanged: customer inventories "much healthier by spring," shipments currently below consumption, resumption of bit growth in FQ3 and H2 FY25; PC refresh delayed, not canceled, with content growth (future-proofing specs) as the 2025 driver.
Key numbers and quarter mechanics- FQ2 GM guide: ~-100 bps vs FQ1 (FQ1 was guided at 39.5% ± 100 bps last quarter; the callback confirms the decline is NAND-driven: mix, pricing, lower shipments). DRAM pricing guidance explicitly refused ("we don't really guide for pricing... competitive reasons").
- FQ2 bits: DRAM down sequentially; NAND down "meaningfully"; NAND = majority of the sequential revenue decline (Bolton's "down 30% plus" framing was not confirmed — Murphy repeated only "majority of the revenue decline").
- FQ3 underload: ~100 bps GM impact (period costs + higher-cost inventory), partially offset by continued mix benefits; Murphy declined to guide FQ3 GM but said favorable drivers (data center growth, HBM/high-cap/LP mix, DC SSD volume resumption) support margin expansion beyond FQ3.
- FQ1 data center: >50% of revenue, +40% QoQ, +400% YoY. Sur's estimate of $1.1–1.2B FQ1 data center SSD revenue (~65% of NAND) was not confirmed or denied; share trajectory cited by Sur (3–5% in CY21 → 10–12% exiting last year → ~14% in Q3, up again in FQ1) was not disputed.
- Inventory: FQ1 DIO <150 days (progress confirmed); FQ2 DIO and dollars up (volume-driven); DRAM below target by end of FY25; NAND improving sequentially FQ2→FQ4.
- CapEx mix: HBM is the largest portion of FY25 CapEx (front-end + assembly/test + cleanroom + the wafer-start replacement from the 3:1 trade ratio); facilities/construction and back-end (India named) next; NAND CapEx cut, below its revenue share.
- DDR4/LP4 ≈ 10% of DRAM revenue for the remainder of FY25 — a structural mix shift to leading-edge, not inventory holding (per Sadana); long-lifecycle industrial/auto/defense keeps a contained tail.
- Demand framework: CY25 DRAM bit demand mid-teens, NAND low-double-digits; Micron ships in line with industry in both for CY25 (stable bit share CY24–25), with a calendar-vs-fiscal four-month skew caveat from Murphy.
- HBM agreements: bit/cube volumes by quarter (8-high vs 12-high) with pricing attached; firm POs inside "fairly lengthy" lead times; cancelable only outside the firm window — first real disclosure of LTA mechanics, extracted by Orabi (TD Cowen, filling in for Sankar).
Product and launch scorecard- HBM3E 8-high: ramp "exceeded our expectations" last quarter (Bhatia); strong continued ramp; sold against a >$30B CY25 TAM with share parity (low-20s) targeted for H2 CY25 — note this is now explicitly second-half, slightly more precise than last quarter's "sometime in calendar 2025."
- HBM3E 12-high: capacity transition through CY25, becoming a larger mix portion through the year; Bhatia asserts the 8-high process is scalable to 12-high. No new qualification milestone beyond last quarter's "production-capable units shipped."
- HBM revenue scale: Sadana said Hari's $800–900M FQ1 estimate is "a little bit high" — implying FQ1 HBM is below ~$800M but well above the prior "several hundred million" FY24 aggregate; the ramp "flattens somewhat but still grows" per Hari's framing, which Sadana/Bhatia did not dispute, attributing slope to gradual capacity adds, not demand.
- Data center SSD: near-term moderation into CQ1 after "several quarters" of heavy buying; bit growth restarts FQ3 and continues; product claims: 30TB and new 60TB SSD — first PCIe Gen5 60TB on the market — plus the 9550 ("fastest SSD in the world"). Sur's #2/#3 share position framing unchallenged.
- LP5X in data center: Micron claims to be effectively the whole current market, with RAS-feature innovation; positioned for the Blackwell/GB200 cycle; combined LP + high-cap DIMM data center opportunity reiterated as "multi-billions."
- High-cap DIMMs: first-mover claim on mono-die 128GB DIMMs reiterated; still no discrete revenue figure.
- NAND supply actions: wafer starts cut (quantified only via the ~100 bps FQ3 underload impact); decisive and immediate, per Sadana.
Sell-side read-through- Eight questioners; tone probing but not hostile. The callback format (no CEO) produced more mechanical than strategic questioning. Notably absent again: China, buybacks, tax.
- Sur (JPMorgan) tried to extract FQ2 blended DRAM pricing direction — flatly refused. His DC SSD share math ($1.1–1.2B, ~65% of NAND, ~14% share) went uncorrected, a soft confirmation; got the FQ3 bit-growth resumption commitment.
- Rakers (Wells Fargo) got the quarter's most useful new number: FQ3 underload impact "closer to 100 bps" and confirmation that favorable mix drivers persist through FQ3 with margin expansion resuming beyond. His LP5X/GB200 question produced the "we are the market" LP claim.
- Caso (Wolfe) asked the key "what changed since September" question — answer: PC refresh pushout (CY24 units now flattish), customer inventory digestion running into CQ1 seasonality, and DC SSD lumpiness; consumer-only framing. His follow-up extracted the clean FQ2 bit split: DRAM down, NAND down meaningfully, NAND the majority of revenue decline.
- Rakesh (Mizuho) got the 3:1 trade-ratio math restated and confirmation HBM is the largest FY25 CapEx bucket; no HBM capacity sizing given.
- Chin (Stifel) competitor NAND discipline question deflected; his DDR4 ~10% question produced a genuine disclosure: the mix shift is structural production, not inventory management.
- Bolton (Needham) pressed customer-inventory quantification (weeks/percent burned through) — refused; got only "better place in spring" and shipping-below-consumption restated. His "NAND down 30%+" inference was left unconfirmed.
- Arcuri (UBS) pinned down that Micron ships in line with industry bit growth in both DRAM and NAND for CY25 (stable share) — closing the under-ship question.
- Orabi (TD Cowen, for Sankar) extracted the first LTA mechanics disclosure: quarterly bit/cube commitments with pricing, firm inside long lead-time PO windows, cancelable outside them.
- Hari (Goldman) got the most consequential soft disclosure: his $800–900M FQ1 HBM estimate is "a little bit high," and his "ramp flattens from here" framing was not pushed back — constraints are capacity/equipment qualification pace, not demand or space specifically; Taiwan facility linkage not confirmed.
Management credibility- Willingness to give negative specificity is a credibility positive: the ~100 bps FQ3 underload quantification, the "DRAM down / NAND down meaningfully" FQ2 bit split, and the FQ2 inventory re-build admission were all volunteered or confirmed cleanly. Murphy was careful with calendar-vs-fiscal skew caveats.
- HBM opacity continues but with a soft correction: still no HBM revenue disclosure ("we may choose to provide some data points in the future"), but Sadana did steer Hari's $800–900M estimate down ("a little bit high") — the first time management has bounded the number from above. This partially offsets last quarter's disclosure regression, though the refusal to quantify persists even as HBM becomes the largest CapEx consumer.
- The "what changed" answer was candid: Caso's question got a direct admission that the PC refresh slipped and CY24 PC units are now flattish — a forecast cut acknowledged rather than buried. However, last quarter's "inventories normalize by spring 2025" is now doing heavier lifting: FQ2 is worse before it gets better, and the spring normalization is asserted, not evidenced.
- Consistency check on inventory: last quarter DIO rose to 158 with drawdown "second-half-weighted"; this quarter FQ1 DIO did fall below 150 (delivered), but FQ2 goes back up — the drawdown path is now explicitly non-linear, and "DRAM below target by end of FY25" is a new, more specific commitment.
- Share-parity language tightened slightly: "second half of calendar '25" (Sadana, twice) vs "sometime in calendar 2025" last quarter — marginally more precise, and now explicitly maintained *despite* the TAM being raised to >$30B, which raises the revenue bar.
- Evasions logged: DRAM pricing direction (refused), customer inventory quantification (refused), HBM capacity sizing (refused), competitor NAND behavior (declined to speculate — reasonable), NAND FQ2 magnitude (Bolton's 30%+ unconfirmed).
What changed versus the prior quarter- Guidance inflected negative: FQ2 GM down ~100 bps (vs FQ1's guided +300 bps step up), with FQ2 bits down in both DRAM and NAND — the first sequential revenue/bit decline guided since the recovery began. Last quarter's narrative was "record FQ1 revenue"; this quarter's is "consumer digestion + NAND correction."
- HBM TAM raised again: >$30B CY25 (from >$25B), the second raise in two quarters; share-parity timing pinned to H2 CY25.
- FQ1 HBM bounded from above: below Hari's $800–900M — implying a very steep FQ1 ramp from FY24's "several hundred million" total, but short of the most aggressive sell-side marks.
- Inventory path re-shaped: FQ1 DIO <150 (improvement delivered) but FQ2 guided back up; DRAM below target by end of FY25 is new specificity.
- NAND moved from "structural improvement" story to active correction: wafer-start cuts, underload charges, CapEx below revenue share — a material change from last quarter's "rational NAND industry" framing.
- PC forecast cut: CY24 units now flattish YoY; refresh pushed into 2025 — last quarter the spring-2025 normalization was the story; now the delay is acknowledged as a driver of the FQ2 miss.
- Data center crossed 50% of revenue with +400% YoY growth — the mix pivot is now numerically dominant.
- LTA mechanics disclosed for the first time (firm windows, cancelability) — a transparency improvement.
- DC SSD went from unqualified momentum to "lumpy": moderation into CQ1 after last quarter's >$1B record; FQ3 resumption promised.
Bull case- The weakness is explicitly scoped and temporary: consumer segments + NAND only; data center (>50% of revenue, +40% QoQ, +400% YoY) untouched; FQ2 GM delta held to ~100 bps despite the headwinds because of mix — evidence the HBM/high-cap/LP mix shift is structurally lifting the margin floor.
- HBM ramp is capacity-limited, not demand-limited: customers "chasing more supply," firm POs under long lead times, TAM raised to >$30B, 8-high ramp ahead of plan, 12-high transition on a "scalable" process, share parity (low-20s) targeted H2 CY25 — the revenue math from low-20s share of a $30B+ TAM supports the "multiple billions" FY25 claim.
- Management is taking NAND medicine early: immediate wafer-start cuts with a quantified ~100 bps FQ3 underload cost, CapEx below revenue share — decisive action that shortens the correction if competitors follow.
- FQ3 re-acceleration is the stated path: DC SSD bit growth restarts, DRAM and NAND volumes both up sequentially from FQ3, margin expansion resumes beyond FQ3, DRAM inventories below target by year-end.
- Structural mix shift is real and disclosed: DDR4/LP4 down to ~10% of DRAM revenue; LP-in-data-center effectively a Micron-only market today; first PCIe Gen5 60TB SSD; leading-edge DRAM supply tight with HBM's 3:1 trade ratio pressuring non-HBM supply — supports DRAM pricing even as commodity weakens.
- Consumer recovery setup intact: shipping below consumption now, inventories healthy by spring, PC refresh delayed-not-canceled with higher content per unit in 2025.
Bear case- The recovery just hit its first guided air pocket: FQ2 revenue, bits, and GM all down sequentially, with NAND down "meaningfully" (possibly 30%+ per Bolton's unchallenged math) and FQ3 carrying ~100 bps of underload — two consecutive quarters of margin pressure before the promised re-acceleration.
- The spring-normalization thesis is now load-bearing and unverifiable: customer inventory burn quantification was refused; the PC refresh already slipped once (CY24 units cut to flattish); if spring slips again, the FQ2 inventory build becomes a pricing liability into H2.
- HBM remains unquantified while becoming the largest CapEx consumer: the only data point this quarter was a downward steer on Hari's estimate, and his "ramp flattens from here" framing went uncorrected — consistent with a ramp that is steep but decelerating, with share parity back-half-loaded into H2 CY25.
- DC SSD "lumpiness" is new: the product line that tripled last year and hit >$1B/quarter is now moderating into CQ1 — a reminder that even the AI-adjacent NAND business is not immune to digestion.
- Inventory goes the wrong way again in FQ2 (DIO and dollars up) right after one quarter of improvement; the end-of-FY25 "below target" commitment for DRAM is the third iteration of a drawdown promise.
- LTA cancelability cuts both ways: the disclosure that customers *can* cancel outside firm windows introduces a theoretical air-pocket risk if AI demand expectations reset, even if current behavior is "chasing supply."
- DRAM pricing opacity at an inflection: refusal to even directionally guide blended FQ2 DRAM pricing, combined with admitted "weak commodity pricing," leaves the mix-vs-price margin bridge unverifiable until the print.
Next-quarter watchlist- FQ2 delivery and the FQ3 setup: does GM land at the guided ~-100 bps, and does FQ3 guidance embed the ~100 bps underload plus the promised volume resumption? Any FQ3 GM framework (flat-to-up vs continued decline) is the key number.
- NAND correction progress: evidence the wafer-start cuts are biting (bit shipments, pricing stabilization post-CQ1); whether competitors followed with cuts (Chin's question, unanswered); DC SSD bit growth actually restarting in FQ3 as promised.
- HBM quantification and slope: whether management finally provides "data points" (Sadana's tease); 12-high mix progression through CY25; capacity adds and any Taiwan-facility linkage; whether the H2 CY25 share-parity target survives another TAM raise.
- Spring inventory normalization: the single most important macro checkpoint — customer inventories healthy by spring, shipments reconverging with consumption, and Micron's own DIO resuming decline after the guided FQ2 uptick.
- DRAM inventory below target by end of FY25: new commitment — track the trajectory and whether leading-edge tightness (HBM trade-ratio pressure) shows up in non-HBM DRAM pricing.
- PC refresh 2025: unit forecasts, AI PC content ramp, and whether the flattish CY24 cut stabilizes or slips again.
- LP-in-data-center revenue evidence: the "we are the market" claim and multi-billions framing (with high-cap DIMMs) still lack any quantified contribution — watch for a first data point as GB200 ramps.
- LTA firmness: any sign of cancellations or renegotiation outside firm windows as CY25 HBM supply loosens industry-wide.
- Still unaddressed: China (fifth straight quarter), buyback pace against the CapEx ramp, tax-line credibility, and the DRAM-vs-NAND margin split.
|
Corbanu