Micron Technology — FQ3 FY2026 Earnings Review

BUY
NASDAQ: MU  | Record AI-memory super-cycle: revenue +346% YoY to $41.5B, clean double beat, FQ4 guided to ~$50B / ~$31 EPS, with $100B of non-cancelable take-or-pay backlog reframing the franchise.
FQ3 FY2026 = quarter ended May 29, 2026  ·  Reported Wednesday, June 24, 2026  ·  FYE late August  ·  Fiscal Q3 (calendar Q2 2026)  ·  All figures non-GAAP unless flagged
Revenue beat
+15.4%
$41.5B vs $35.9B cons; +346% YoY
Non-GAAP EPS beat
+19.7%
$25.11 vs $20.98 cons; +1,215% YoY
Growth trajectory
Accelerating
Rev YoY +196% to +346%; peak accel at latest qtr
FQ4 guide vs Street
+17% / +22%
$50B rev / $31 EPS -- guide low end clears cons
Executive summary -- what is new
Micron delivered the most important memory print in a decade. Fiscal Q3 revenue was a record $41.5B (+346% YoY, +74% QoQ), a fifth consecutive record, beating consensus by +15.4%; non-GAAP EPS of $25.11 beat by +19.7% and is up +1,215% YoY. Non-GAAP gross margin hit a company-record 84.9% -- more than double a year ago -- with all three guided metrics (revenue, gross margin, EPS) clearing the high end of guidance. This is an AI-memory ASP super-cycle: DRAM prices rose low-60s% and NAND mid-80s% sequentially on structurally short supply, with bit shipments nearly flat. Data-center revenue exceeded $25B (>$100B annualized run-rate).
New guidance (FQ4 FY2026): Record revenue ~$50.0B ±$1B (+17% vs ~$42.7B Street; ~+342% YoY), gross margin ~86%, opex ~$1,650M, EPS ~$31 ±$1 (+22% vs ~$25.4 Street). Even the low end ($49B / $30) clears consensus. Implied FY2026 revenue ~$127B vs ~$113.9B Street (+11%), EPS ~$73 vs ~$63 Street (+15%).
Tone -- materially more structural: Strategic Customer Agreements jumped from 1 to 16 (~$100B minimum contracted revenue, ~$22B deposits/commitments with ~$18B cash, five-year take-or-pay, non-cancelable). Tightness window extended from "beyond 2026" to "beyond calendar 2027." HBM TAM now expected to "easily cross $100B in 2027" (pulled from 2028). Capital return committed from Dec 9, 2026 (CHIPS anniversary), with 100% of excess cash targeted via share repurchase. The narrative shifted from "great cyclical up-quarter" to "structurally transformed business model."
Watch items: (1) CEO Sanjay Mehrotra was absent from the most transformative call in company history -- no explanation provided; a genuine management flag. (2) SCA disclosure posture reversed entirely: management refused all specifics in FQ1/FQ2 on confidentiality grounds, then disclosed granular mechanics in FQ3. (3) Gross-margin expansion is decelerating: +110bps guided into FQ4 vs +1,000bps just reported; management flagged "meaningful moderation in the rate of price increases." (4) CapEx guidance has escalated ~50% from initial framework ($18B to $27B) across three quarters. (5) Acceleration this extreme is mathematically unsustainable as FY2027 laps the blow-out -- the real question is durability of the new revenue level.
Upcoming catalysts: FQ4 print (~late Sep 2026) with first full 10-K RPO disclosure; HBM4 12-high volume ramp (~2x faster than HBM3E, >$1B shipped, certified for NVIDIA Vera Rubin); capital-return inflection from Dec 9, 2026; greenfield fabs (Idaho ID1 mid-2027); emerging auto/robotics/edge-AI content as under-modeled optionality. Catalyst calendar is overwhelmingly upside-skewed.

Key metrics and trends (11 quarters)
Standalone fiscal quarters. Revenue and GAAP figures from Daloopa (company_id 136); GM% = gross profit / revenue; YoY = same fiscal quarter prior year. Non-GAAP EPS is the cleaner earnings series. EBITDA = GAAP operating income + D&A.
Fiscal Qtr Revenue ($M) Rev YoY GM% GM YoY bps DRAM ($M) NAND ($M) NG EPS
FY23Q3 3,752 -- -17.8% -- -- -- -1.43
FY23Q4 4,010 -- -10.8% -- -- -- -1.07
FY24Q1 4,726 -- -0.7% -- -- -- -0.95
FY24Q2 5,824 -- 18.5% -- 4,158 1,567 0.42
FY24Q3 6,811 +82% 26.9% -- 4,692 2,065 0.62
FY24Q4 7,750 +93% 35.3% -- 5,326 2,365 1.18
FY25Q1 8,709 +84% 38.4% +3,906 6,400 2,241 1.79
FY25Q2 8,053 +38% 36.8% +1,834 6,123 1,855 1.56
FY25Q3 9,301 +37% 37.7% +1,089 7,071 2,155 1.91
FY25Q4 11,315 +46% 44.7% +938 8,984 2,252 3.03
FY26Q1 13,643 +57% 56.0% +1,768 10,812 2,743 4.78
FY26Q2 23,860 +196% 74.4% +3,761 18,768 4,997 12.20
FY26Q3 ★ 41,456 +346% 84.6% +4,684 31,328 9,943 25.11
FQ4 guide ~50,000 ~+342% ~86% ~+3,450 -- -- ~31
Revenue by business unit (FQ3 FY2026, all records): Cloud Memory $13,769M  ·  Core Data Center $11,524M  ·  Mobile & Client $11,521M  ·  Auto & Embedded $4,634M. Data-center (Cloud + Core) reached >$25B (~61% of revenue). Enterprise SSDs hit $5B in the quarter, more than doubling sequentially. Growth is overwhelmingly price/ASP-driven, not volume.
Revenue ($B) -- last 11 quarters + guide
FY23Q3
$3.8B
FY23Q4
$4.0B
FY24Q1
$4.7B
FY24Q2
$5.8B
FY24Q3
$6.8B
FY24Q4
$7.8B
FY25Q1
$8.7B
FY25Q2
$8.1B
FY25Q3
$9.3B
FY25Q4
$11.3B
FY26Q1
$13.6B
FY26Q2
$23.9B
FY26Q3
$41.5B
FQ4 guide
$50.0B
YoY Revenue Growth % -- 8 quarters with YoY available
FY24Q3
+82%
FY24Q4
+93%
FY25Q1
+84%
FY25Q2
+38%
FY25Q3
+37%
FY25Q4
+46%
FY26Q1
+57%
FY26Q2
+196%
FY26Q3
+346%
Growth troughed at +37% (FY25Q3), then re-accelerated: +46% → +57% → +196% → +346%. Five consecutive quarters of acceleration.
Data sourced from Daloopa (company_id 136). FQ4 figures are management guidance. EBITDA calculated as GAAP op inc $33,318M + D&A $2,364M = $35,682M.

Beat / miss -- heatmap (8 quarters, this quarter highlighted)
Measured on non-GAAP EPS and revenue vs Street consensus (FMP). Green = beat, red = miss. MU has beaten EPS in 12/12 of the last 12 quarters (100%); revenue 10/12 (83%), 4/4 over the last 4. Beat magnitude is accelerating -- the Street chronically lags the AI-memory pricing curve.
Metric FY24Q4FY25Q1FY25Q2FY25Q3FY25Q4FY26Q1FY26Q2 FY26Q3 ★
EPS +5.4%+2.3%+9.1%+19.4%+5.9%+20.7%+32.8%+19.7%
Revenue +1.3%-0.1%+2.0%+4.9%+0.9%+5.7%+19.5%+15.4%
FQ3 FY2026 vs consensus Consensus Actual Variance
Revenue $35,912M $41,456M +$5,544M / +15.4%
Non-GAAP EPS $20.98 $25.11 +$4.13 / +19.7%
Non-GAAP gross margin -- 84.9% Above high end of guide
Beat rates: L12Q EPS 12/12 (100%), Revenue 10/12 (83%). L4Q both 4/4 (100%). Last 3 EPS surprises: +20.7%, +32.8%, +19.7% vs low-single-digit beats in FY24/early FY25. Revenue surprise expanded from ~+1-2% to +19.5% and +15.4% in the two most recent quarters. The Street has been chronically slow to model the AI-memory pricing curve.
Revenue actuals cross-checked to Daloopa net sales (exact match). Consensus from FMP Street estimates; Daloopa does not carry sell-side consensus.

Guidance deep dive -- FQ4 FY2026
Metric New guide (mid) Range Consensus vs Cons Implied YoY
Revenue $50.0B $49--51B ~$42.7B +17% ~+342%
Gross margin ~86% ~85--87% ~83% +300bps ~+3,450bps
Non-GAAP EPS $31.00 $30--32 ~$25.4 +22% ~+923%
Non-GAAP OpEx $1,650M $1,650M -- -- --
CapEx ~$10.0B ~$10.0B -- -- FY26 total ~$27B
Revenue waterfall ($B) -- FQ4
Prior anchor (FQ3 guide)
$33.5
FQ3 actual
$41.5
Street consensus
$42.7
New guide LOW
$49.0
New guide MID
$50.0
New guide HIGH
$51.0
Even the low end ($49B) sits ~$6B above Street. FQ3 actual beat FQ3 guide by +24%.
EPS waterfall ($) -- FQ4
Prior anchor (FQ3 guide)
$19.1
FQ3 actual
$25.1
Street consensus
$25.4
New guide LOW
$30.0
New guide MID
$31.0
New guide HIGH
$32.0
MID is +$5.60 above consensus. Three consecutive massive beats suggest guidance remains sandbagged.

Guidance trajectory -- sequential progression

Qtr reporting Fwd qtr guided Rev guide GM guide EPS guide
FQ2 FY25 FQ3 FY25 $10.7B 42.0% $2.50
FQ3 FY25 FQ4 FY25 $12.5B 51.5% $3.75
FQ4 FY25 FQ1 FY26 $18.7B 68.0% $8.42
FQ1 FY26 FQ2 FY26 $33.5B 81.0% $19.15
FQ2 FY26 FQ3 FY26 (actual) Guide $33.5B; delivered $41.5B (+24%) 81% → 84.9% $19.15 → $25.11
FQ3 FY26 FQ4 FY26 (NEW) $50.0B 86.0% $31.00
The margin tell. Sequential gross-margin expansion is flattening -- the bps step ran +1,768 (FQ1) → +1,000 (FQ3) → +110 guided (FQ4). Management explicitly flagged "a meaningful moderation in the rate of price increases." Revenue/EPS guidance stays on the super-cycle trajectory, but margin is approaching a ceiling at unprecedented levels. The guide remains conservative versus the +24%/+31% revenue/EPS beats just delivered.
FY2026 full-year implied: Revenue ~$127B vs ~$113.9B consensus (+11%), EPS ~$73 vs ~$63 consensus (+15%). Both exceed consensus by double-digit percentages -- further evidence of systematic consensus undershoot.
Guidance from MU FQ3 FY2026 call (2026-06-24); revenue/GM guide tagged in Daloopa. Prior guide from FQ2 call. Consensus from FMP (approximate).

Historical performance -- inflection points (8-quarter trajectory)
Metric FY24Q4FY25Q1FY25Q2FY25Q3FY25Q4FY26Q1FY26Q2FY26Q3
Revenue YoY % +93.3%+84.3%+38.3%+36.6%+46.0%+56.7%+196.3%+345.7%
Rev accel (bps QoQ) -900-4,600-170+940+1,070+13,960+14,940
EPS YoY % NMNM+271%+208%+157%+167%+682%+1,215%
EPS accel (bps QoQ) NM-6,330-5,130+1,020+51,510+53,260
A textbook memory super-cycle in two phases. Phase 1 -- Decelerating recovery (FY24Q4 to FY25Q3): Revenue growth decelerated from +93% to +37% as easy comps burned off and commodity pricing normalized. Even at the trough, growth was solidly positive -- no double-dip. Phase 2 -- Vertical re-acceleration (FY25Q4 to FY26Q3): Five consecutive quarters of accelerating revenue growth, culminating in +346% YoY. EPS amplifies revenue dramatically (+1,215% vs +346%) via operating leverage on a fixed-cost fab base and record 84.9% gross margins. The acceleration itself is accelerating -- Rev Accel went from +940 bps to +14,940 bps QoQ -- the hallmark of a demand shock outstripping capacity.
Marked inflections: (1) FY25Q3 trough -- revenue YoY bottoms at +36.6%, deceleration narrows to -170 bps, signaling exhaustion. (2) FY25Q4 the turn -- first positive acceleration in five quarters (+940 bps). (3) FY26Q2 breakout -- revenue vaults to +196% as HBM/data-center pricing inflects. (4) FY26Q3 peak acceleration -- both growth rates and their acceleration at window highs simultaneously.
The critical forward question: FY2027 will lap the FY2026 blow-out quarters, so YoY rates must mathematically compress. The relevant question shifts from "is it accelerating?" to "how durable is the new revenue level?" Management's $100B take-or-pay backlog and "no line of sight" on when supply catches demand both argue the level is structural, not cyclical.
Data sourced from Daloopa (company_id 136). NM = not meaningful (laps negative prior-year EPS). Acceleration = QoQ change in YoY growth rate, basis points.

Key catalysts
Tier Catalyst Timing Management signal Skew
1 FQ4 guidance execution (~$50B rev) Aug 2026 Three consecutive massive beats suggest guidance sandbagged UPSIDE
1 HBM4 volume ramp / Vera Rubin H2 CY2026 Demand far exceeds supply; HBM4 ramping ~2x faster than HBM3E; >$1B shipped UPSIDE
1 SCAs expanding toward 50% rev coverage Through CY2030 16 signed; $22B+ deposits; non-cancelable take-or-pay; ~$100B minimum UPSIDE
1 Capital return initiation Dec 9, 2026 100% excess cash via buyback; ~10% annual buyback potential at current FCF UPSIDE
2 HBM4E volume ramp CY2027 Well underway; customizable base dies on TSMC advanced node UPSIDE
2 Structural supply-demand gap Beyond CY2028 No intercept visible; supply growth is the binding constraint UPSIDE
2 Idaho Fab 1 wafer output Mid-CY2027 Construction well underway; greenfield bits contribute CY2028 NEUTRAL
2 Enterprise SSD share gains Ongoing $5B/qtr; record share 4th consecutive year; QLC + Gen6 leadership UPSIDE
3 Agentic AI / robotics demand CY2026+ 20-year growth vector; 30-40% CAGR in CPU server demand; not yet modeled UPSIDE
3 FY2027 CapEx intensity FY2027 Above mid-40% of revenue; majority construction; near-term FCF pressure MIXED
3 U.S.-China trade dynamics Ongoing CXMT/YMTC contained within China; tariff surcharges imposed; policy wildcard MIXED
Catalyst calendar overwhelmingly upside-skewed. The combination of (1) structural supply shortage with no visible intercept, (2) $100B in non-cancelable SCAs, (3) imminent capital return unlock, and (4) a product roadmap aligned with the two highest-growth vectors in semiconductors (HBM for AI training/inference and LPDRAM for agentic AI) creates a multi-year compounding thesis. Primary risk is execution on greenfield capacity and trade policy disruption, not demand destruction.
Sources: MU FQ3 & FQ2 FY2026 earnings call transcripts. SCA/HBM4 dollar figures are transcript-sourced. Fundamental data from Daloopa.

Street Q&A -- 14 questions, 4 deflections
Seven analysts, 14 questions. Management was exceptionally detailed on structural topics (SCA mechanics, HBM demand, competitive positioning) but systematically avoided pinning specific numbers on capital allocation and supply/demand gap metrics -- preserving optionality at the cost of Street modeling precision.
Analyst Topic Verdict
Reitzes (Melius) Capital return -- can buyback reach ~10% of market cap? DEFLECT
Reitzes (Melius) DRAM mix -- appetite to skew toward commodity DRAM? WELL ANS
Sur (JPMorgan) HBM bookings -- booked through CY2027 on HBM3E/HBM4? WELL ANS
Sur (JPMorgan) Mid-term DRAM/NAND bit demand CAGR outlook DEFLECT
O'Malley (Barclays) SCA cancellation provisions and financial hooks WELL ANS
O'Malley (Barclays) Cash deposit mechanics -- why returned vs. recognized? WELL ANS
Weathers (Deutsche) SOCAMM / LPDRAM demand trends in data center WELL ANS
Weathers (Deutsche) Greenfield fab startup cost impact on cost per bit WELL ANS
Rakesh (Mizuho) Do SCAs include HBM and major CSPs? WELL ANS
Rakesh (Mizuho) Cash deposit return schedule mechanics WELL ANS
Schneider (Goldman) What % of forecast demand can MU supply through CY2028? DEFLECT
Schneider (Goldman) Could FY2027 CapEx reach 55-60% of revenue? DEFLECT
Rakers (Wells Fargo) NAND strategic advantage within SCA bundling WELL ANS
Rakers (Wells Fargo) China competition -- CXMT/YMTC evolution WELL ANS

Deflection pattern

Theme What was asked What was missing
Capital return magnitude Buyback as % of market cap No quantitative framework for sizing the program
Demand growth Mid-term DRAM/NAND bit demand CAGRs No numbers; reframed as supply-constrained (true but incomplete)
Supply fulfillment % of demand MU can serve through CY2028 No aggregate range; segment heterogeneity cited
CapEx precision FY2027 CapEx intensity band Refused to narrow beyond 'above mid-40%'
All four deflections involve forward quantitative commitments. Management was exceptionally forthcoming on structural topics but consistently avoided pinning specific numbers on capital allocation and supply/demand gap metrics. This is a deliberate strategy to preserve optionality -- but it leaves the modeling community guessing on four of the most important inputs for FY2027-2028 estimates.
Source: MU FQ3 FY2026 earnings call transcript (2026-06-24).

Contradictions -- 6 identified across transcript series
Contradiction 1 -- HBM TAM $100B timeline: constantly moving target
Severity: Moderate
The $100B milestone migrated from 2030 to 2028 to 2027 across three consecutive earnings cycles (9 months). Each revision was presented as a definitive forecast. While upward revisions are positive, the magnitude of change raises questions about forecasting rigor.
Contradiction 2 -- CEO absent from landmark quarter (HIGH SEVERITY)
Severity: High
Sanjay Mehrotra led every call from FY25Q2 through FY26Q2 with prepared remarks and Q&A. On FQ3 -- the most transformative quarter in company history ($41.5B revenue, 85% GM, $100B SCA backlog) -- the CEO was absent with no explanation. A genuine management red flag per investing principles (C-suite stability, transparency).
Contradiction 3 -- SCA confidentiality reversal
Severity: Moderate
Management categorically refused SCA specifics in FQ1/FQ2 on 'confidentiality' grounds (Mehrotra: 'we are certainly not getting into the specifics'). In FQ3, Sadana disclosed granular mechanics: 16 agreements, $22B deposits, take-or-pay structure, price bands, five-year terms. Either prior over-caution or a deliberate staged reveal -- both affect weighting of management commentary.
Contradiction 4 -- CapEx guidance escalated ~50%
Severity: Moderate
FY2026 CapEx went from ~$18B baseline (implied $4.5B/qtr 'reasonable baseline') to $20B to 'above $25B' to ~$27B in three quarters. Each revision presented matter-of-factly. Undermines individual forecast credibility, though driven by genuine demand.
Contradiction 5 -- Customer fulfillment language shifted
Severity: Low-Moderate
FQ1/FQ2: quantified 50-67% fulfillment rate. FQ3: vaguer 'supply numbers are a fraction of what they want.' Escalation or obfuscation? Either way, the tighter language supports the scarcity narrative without the prior precision.
Contradiction 6 -- Non-HBM margins exceed HBM margins
Severity: Low
FQ2 explicitly acknowledged non-HBM margins exceed HBM -- reframing the bull case from 'HBM as margin engine' to 'supply scarcity across all DRAM as margin engine.' This matters because the entire HBM premium narrative is partially misplaced.
Contradictions 2 (CEO absence) and 3 (SCA disclosure flip) are the most notable from an investing-principles perspective. The CEO's unexplained absence from the most transformative quarter in company history is a genuine management flag. The SCA disclosure reversal suggests either prior over-caution or a deliberate staged reveal -- both affect how investors should weight management commentary on other "confidential" matters. Contradictions 1 and 4 (TAM and CapEx serial revision) are directionally positive but erode the predictive value of any individual forecast.
Source: MU FQ2 through FQ3 FY2026 earnings-call transcripts. No fundamental figures fabricated.

Indirect read-throughs
Macro / sector: MU frames the memory shortage as structural and lasting "beyond calendar 2027" -- a bullish AI-capex super-cycle signal confirming hyperscaler spending intentions remain unbroken. Record DRAM (+low-60s% seq) and NAND (+mid-80s% seq) pricing is inflationary for downstream server/PC/smartphone/auto BOMs. Greenfield fab costs rising industry-wide structurally supports pricing -- cost floors are rising for all producers. 16 non-cancelable SCAs with $22B+ deposits represent a permanent shift in the supplier-customer power dynamic. The biggest indirect read-through is for hyperscale CSPs: their AI ambitions require committing to multi-year, non-cancelable memory contracts at structurally higher prices.
Company / sector Read-through Direction
Apple (AAPL) Analyst noted Apple willing to pay full price for DRAM and pass through to customers -- signals broad OEM acceptance of structurally higher memory costs BETTER
NVIDIA (NVDA) HBM demand driven by GPU/accelerator ramps; HBM TAM pulled to $100B+ in 2027 is a direct positive for GPU attach rates; CPU ambitions validated BETTER
Qualcomm (QCOM) Validated as real data center CPU contender alongside x86; more CPU suppliers = more DRAM attach, especially LPDRAM/SOCAMM BETTER
Intel/AMD (x86) Implied via agentic AI driving 30-40% CAGR forecasts for server CPUs; benefits if agentic AI truly drives growth at those rates BETTER
Hyperscale CSPs Locked into multi-year non-cancelable memory purchases at floor-to-ceiling price bands; memory becoming larger, less flexible cost line MIXED
Samsung / SK Hynix Same structural tailwinds -- rising ASPs, constrained supply, pricing power for entire oligopoly; MU's SCA model may pressure competitors to adopt similar structures BETTER
CXMT / YMTC (China) Capabilities growing but sales overwhelmingly within China; no evidence of share theft outside China NEUTRAL
Automotive OEMs Memory allocation constraints could bottleneck vehicle production, especially ADAS/autonomous; rising memory costs and potential supply shortfalls through CY2028+ WORSE
Fab construction / power infra Accelerating greenfield build (Idaho, NY, Taiwan, Singapore) means sustained construction and power-equipment spend BETTER
PC / smartphone OEMs Memory price inflation hits BOMs into soft units; premium-mix OEMs best positioned to pass through WORSE
Source: MU FQ3 FY2026 earnings transcript (2026-06-24). Apple comment from analyst (Reitzes, Melius). Peer names illustrative, not MU statements. Data from Daloopa.

Coverage flags: Visible Alpha, Bloomberg, S&P/Kensho, AlphaSense, Snowflake and internal SharePoint/OneNote/Outlook sources were not connected this run -- consensus is FMP-sourced (approximate at the quarterly level); SCA/HBM4 dollar figures and competitor/fab timelines are transcript- or web-sourced and flagged accordingly. No stock price, market cap, or multiple fabricated.
Data sourced from Daloopa.