Built by Grok 4.6 High

Micron Technology reported fiscal Q4 FY2026 after the close on Wednesday, September 30, 2026. Analysis prepared October 10, 2026. Quarter ended September 3, 2026. Fiscal year-end late August.

MU | Earnings Review — FQ4 FY2026

Verdict: DOLLARS STILL COMPOUNDING — the YoY rate has peaked. Fiscal Q4 net sales were $54,229 million (+379.3% YoY), a sixth consecutive revenue record and still a window high. Acceleration collapsed to +3,360 bps from +14,940 bps last quarter. Non-GAAP EPS $33.42 (+1,003.0% YoY) is the first EPS deceleration of the AI cycle (−21,170 bps). That is a comp wall, not a demand break: FQ4 now laps $11.3 billion and $3.03, not the sub-$10 billion / sub-$2 prints that manufactured the vertical steps.

Print (clean double beat, price not bits). Versus FMP $51.334B / $31.77 = +$2.9B / +5.6% revenue and +$1.65 / +5.2% EPS. Versus the June box: revenue cleared the $51.0 billion high by $3.2B; non-GAAP GM printed 87.0% ($47,204 million / sales) vs ~86%; EPS cleared the $32.00 high by $1.42. DRAM bits mid-single-digits with prices high-teens; NAND bits ~+10% with prices ~30%. Data center (CMBU+CDBU) is 63% of the print. GAAP EPS $32.87 is $0.55 under non-GAAP — not a quality issue. Opex $2,568 million missed the $1,650 million box on a company-wide incentive raise plus $300M of community investments; most manufacturing incentive sat in inventory and is the FQ1 GM tax.

New guidance (FQ1 FY2027): record $61.5 billion ± $1.5 billion, non-GAAP GM 86.25% (the stated floor for FY2027), opex ~$2,060 million, EPS $38.15 ± $1.00. FMP already sits at $61.54B / $38.20. Implied YoY +351% / +698% — still explosive, the first deceleration of the rate. The actual raise versus June is FY2027 opex +$2.5B (was +$1.0B), 1H capex $25 billion (2H higher, implying ≥$50B), and tightness through 2028.

Tone — more duration, less launch-event. “FY2026 outstanding; FY2027 even better.” Tightness “much tighter in 2027 and 2028 than in 2026”; “no line of sight to balance.” 26 SCAs (from 16); RPO ~$150B at floor; $32B customer commitments; >75% of FY2027 shipments already committed. HBM CY2027 mostly sold at “significant” price increases; custom NVHBM with NVIDIA. China mix single-digit in FY2027. Capital return still gated to December 9.

Contradictions (3 genuine). SCA math still does not close: $150B floor RPO vs “over 35% of revenue through 2030” is the same ~2× gap as Q3. CQ2 ceilings on the old book cannot coexist with FQ4 high-teens blended DRAM ASP. The CY2028 HBM TAM of ~$100B is three quarters stale after a disclosed CY27 price reset.

Upcoming catalysts. FQ1 print 2026-12-23 (Street on the midpoint — in-range can screen as a miss). Whether 86.25% is the GM floor. December 9 CHIPS buyback start. SK Hynix ~Oct 27 / Samsung Oct 29 / NVIDIA Nov 17 as the external test of 2028 tightness.

Fiscal 2021–2026 annuals (the cycle, not just the upswing). FY2026 is 4.3× FY2022 sales.

Memory is units × ASP. Drivers sit above the consolidated total. YoY is the same fiscal quarter one year prior — never mixed-quarter sequential as a growth rate. Non-GAAP EPS is the Street line.

Revenue drivers — DRAM / NAND / business units

BU disclosure starts FY2025Q4. Core Data Center is now the single largest unit.

Driver read. Growth is overwhelmingly price, not bits. NAND YoY re-accelerated to +526% because data-center SSD was “nearly $10 billion, more than 10 times the year-ago quarter, and over two-thirds of NAND.” CDBU — conventional-server DDR plus SSD, not HBM — is the mix surprise and the 90% GM engine. CMBU GM is stuck at 83% because higher HBM mix offset price; that is the FQ1 tell on whether CY27 HBM price resets close the gap to conventional DRAM. MCBU bits fell for a second consecutive quarter even as dollars and 90% GM printed a record — harvesting the low end, not losing the premium attach.

Consolidated P&L

FY2026 closed at $133,188 million of sales (+256.3% YoY) and non-GAAP EPS $75.52 (+811%). DRAM “surpassed $100 billion” at $100,679 million. That is 4.3× FY2022’s prior-cycle peak, not a recovery year.

FMP scores net sales and non-GAAP diluted EPS. The company also guides GM and opex. Q(N) guide applies to Q(N+1) actuals.

This quarter vs consensus / own box

Street was already above the company revenue high ($51.3B vs $51.0B) and inside the EPS box ($31.77 vs $30–$32). Beating Street and the high end of a guide the Street had already overtaken is a pricing/volume shock, not a sandbag. Mehrotra: “each exceeding the high end of our guidance.”

Beat rates

Pattern: consistent beater. Revenue’s only L12 misses are FY24Q1 (−1.7%) and FY25Q1 (−0.1%) — trough-cycle noise. This print’s +5.2% / +5.6% is the smallest percent surprise of the last four quarters. If FQ1 prints inside the new $61.5B / $38.15 box (FMP already on the midpoint), “consistent beater” starts to mean guide-accurate rather than Street-wrong.

Micron guides one quarter out (revenue, GM, opex, EPS, capex, tax). It does not issue a full-year revenue or EPS number. Daloopa stores the quarterly guide on the issuing print.

New FQ1 FY2027 box (issued 30 Sep 2026)

Implied FQ1 YoY at the midpoint vs FY2026Q1 actuals: revenue +350.8% vs $13,643 million; EPS +698.1% vs $4.78; GM +2,942 bps vs 56.8%. The sequential GM dip (87.0% → 86.25%, −75 bps) is FQ4 incentive-comp manufacturing costs absorbed into inventory. Murphy: FQ1 is the floor; “higher gross margins beyond fiscal Q1 … with a more moderate rate of price increases.” FQ1 factors in single-digit sequential bit growth and double-digit cost for both DRAM and NAND — still a price/mix print.

FY2027 color — the actual guide change

No company FY revenue/EPS/GM range. Street FY2027 (FMP, 31/24 analysts): $276.33B / $173.88, +107.5% / +130.2% vs FY2026 actuals. Annualizing FQ1 ($61.5B × 4 = $246B) is below Street, which is the right read if sequential growth each quarter holds and FQ1 is the low-water mark.

Majority of the FY2027 capex increase is construction for clean rooms that come online late calendar 2028+. It does not add 2027 bits. That is bullish for 2027 ASPs and FCF-negative for 2027/28 conversion. FY2027 opex of ~$6,841 million + $2.5B ≈ $9.3B. Operating leverage still works if Street’s $276B prints. If tightness fades, the opex raise does not flex.

Classification. FQ1 P&L is an in-line print after a year of sandbagged beat-and-raises. Do not treat $61.5B as conservative the way $50B was in June. The shock is the cost of keeping tightness going.

Revenue YoY is still an 8-quarter high at +379.3%. The second derivative has rolled. EPS YoY inflected down to +1,003.0%. Trajectory over absolutes: the level is a super-cycle record; the rate of change of the rate is no longer vertical.

Marked inflections.

1. FY2025Q3 — trough. Revenue YoY bottoms at +36.6%. Easy comps ran out; consumer/mobile still soft.
2. FY2025Q4 — the turn. Revenue YoY +46.0% (+940 bps). First AI-memory signal overtaking the comp drag.
3. FY2026Q2 — breakout. Revenue +196.3% (+13,960 bps). HBM / data-center DRAM volume and pricing inflect together.
4. FY2026Q3 — peak acceleration. Revenue +345.7% (+14,940 bps); EPS +1,214.7%. Both the level and the second derivative peaked here, not this quarter.
5. FY2026Q4 — this print. Revenue still accelerates, but only +3,360 bps. EPS decelerates 21,170 bps. FQ1 guide extends the rollover: +350.8% / +698.1% YoY.

The bull case does not need +14,940 bps of acceleration forever. It needs the new revenue level to be durable: tightness through 2027–28, HBM contracted, SCAs locking volume. The data this quarter does not show demand rolling over. It shows the growth rate digesting its own base.

Street has re-anchored FQ1 to the company midpoint. The live debate is whether 86.25% is the GM floor, whether Dec-9 buybacks start, and whether SK Hynix / Samsung / NVIDIA confirm tightness through 2028.

Bull confirmation, in order: FQ1 GM ≥ 86.25% with the floor restated; sales at/above $61.5B plus a FQ2 sequential-up comment; Dec 9 buybacks actually start without a new capex raise; SK Hynix / Samsung / NVIDIA all describe CY2027 as still short; SCA count or 2030 coverage moving up from 26 / >35%.

Bear confirmation: FQ1 GM < 86% and the floor language dropped; sales miss FMP and “sequential growth each quarter” is retired; peers guide CY2027 DRAM bits back into the mid-20s+; Dec 9 arrives with no authorization while FQ1 capex prints above $11.5B; MCBU bits down a third straight quarter.

What Street is most likely to get wrong. FMP crawled onto the exact midpoint of a guide that was a $4.5B beat versus pre-print LSEG. The easy estimate revision is done. A $61.0B / 86.0% / $37.50 quarter would be a fundamental in-range print that screens as a miss.

Post-earnings analyst call, 2026-09-30. Satya Kumar (IR), Manish Bhatia (COO), Scott DeBoer (CTO), Mark Murphy (CFO). Sanjay Mehrotra was not on this call. Eight sell-side names, dual-part questions scored separately. 13 well answered / 3 deflected.

The Street used the session to pressure-test one idea: the FQ4 beat and FQ1 guide are not the debate — whether tightness, HBM mix, and SCA coverage can hold 2027–28 margins once construction capex that does not produce bits until late 2028 is the growth vector.

Well answered on why. 2028 tighter than 2026 because >75% of FY27 bits are already committed, HBM trade ratios rise, and clean rooms take “a few quarters” after first wafers. DRAM — not logic or power — is the binding AI constraint. China is two nodes behind (1γ already majority of bits, EUV-gated; 1δ 2H 2027). HBM4E is NVIDIA-foundry custom, not an in-house base die the way HBM4 is. Construction ≠ 2027 bits.

Three deflections, all commercial sizing.

1. HBM TAM (Weathers). Explicit “we are not updating that TAM outlook right now.” CY27 HBM price-up and “narrow the profitability gap with conventional DRAM” is the only new handle. Highest-priority open sizing question.
2. Hyperscaler identity / whether old bands reset (Schneider). “Not commenting specifically.” The 26 SCAs / >35% remain a coverage statistic, not a named book. New deals price off current tightness; the old CQ2 book was not described as reset.
3. Company NAND bits vs industry 2027–28 (Hosseini). Explicit “not commenting out that far” on company DRAM or NAND supply. 2026 undergrowth explained (Singapore clean room to NAND R&D + HBM pilot). Share vs industry in the two tight years is the missing NAND input.

Versus FQ3 (3/10 well answered, almost all SCA floor-price refusals), this session was structurally more forthcoming. Remaining dodges are commercial, not “we will not discuss 2028 at all.” Until TAM, logos, and company bit-share are disclosed, 2028 tightness is a management assertion with a strong operating-mechanism overlay, not a modeled duration.

Murphy’s unprompted close: FQ1 embeds single-digit sequential bits and double-digit cost for both DRAM and NAND; FY2027 R&D up more than $1B even ex incentive.

Three genuine contradictions. The reported P&L arithmetic is internally clean. The credibility items are on SCA conversion math and a stale HBM TAM, not on the print. Labeled outlook updates (tightness horizon, bit-growth ranges, FY27 opex/R&D, capex) are excluded.

Checked tensions that are not contradictions: 16→26 SCAs and $100B→$150B RPO (labeled adds); FY27 opex +$1B→+$2.5B (Mark said the words “more than the $1 billion we said last quarter”); tightness horizon extensions (“we now expect”); FQ4 opex miss vs $1,650M (in-quarter incentive + $300M community); “fourfold” data center (3.9×); DRAM “surpassed $100 billion” (ties $100,679 million); FY GM 81.1% / “40 percentage points above FY2025” (ties).

MU extends tightness through 2028 and names NVIDIA custom HBM4E and DRAM (not power) as the AI bottleneck — better for the memory oligopoly and WFE, BOM inflation for hyperscalers and device OEMs.

Top three. (1) Tightness is being extended, not faded — June said “beyond calendar 2027”; this print adds 2028, says both years are tighter than 2026, and refuses line of sight to balance. Unambiguously better for MU / Samsung / SK Hynix on price; multi-year BOM inflation for every memory consumer. (2) NVIDIA custom HBM4E (NVHBM) is the named customer event — better for the NVDA–MU–TSMC rack, closed door for late ASIC HBM upside. (3) DRAM, not power or logic, is now MU’s stated bottleneck, with Meta Muse as the consumer-agentic proof that CPU servers are a second demand vector.