Q3 FY2026 10-Q (filed 25 Jun 2026); Q3 FY2026 earnings call (24 Jun 2026); FY2025, FY2023 and FY2018 10-Ks; KeyBanc Technology Leadership Forum (10 Aug 2026) and Six Five Summit (27 Aug 2026) transcripts, both with Sumit Sadana
latest period
Q3 FY2026 (quarter ended 28 May 2026), plus FQ4 guidance and management commentary to 27 Aug 2026
missing or stale data
FQ4 FY2026 results and the FY2026 10-K are not yet in sources, so SCA deposits received and the ~$100bn floor RPO are unverified. The company does not quantify the floor-price gross margin, and it does not disclose HBM revenue or share. No Form 4 data. Industry supply/demand figures are management estimates. No market or valuation data in sources.
Micron Technology — Bull & Bear Memo
Prepared 11 September 2026 from company sources only.
1. Business in one line
Micron makes DRAM (76% of Q3 FY2026 revenue) and NAND (24%) in its own fabs. It sells bits at prices set by industry supply and demand, so profit is the gap between that price and a per-bit cost it lowers with each node.
CyclicalCyclical, and management is arguing that multi-year contracts have made it something else. The whole memo turns on that claim. Q3 FY2026 revenue was $41.5bn, up 346% y/y, but cost of goods sold rose only to $6.4bn from $5.8bn. Nearly all of the extra profit came from price, not from volume or cost [Q3 FY2026 10-Q].
2. Bull case — Peter Lynch pitch
The simple reason this stock could work is that AI has made memory the bottleneck in computing systems, and new supply physically can't arrive before 2028. Micron is also using the shortage to lock customers into five-year take-or-pay contracts whose floor prices, management says, still earn more than any past peak. If that holds, a business the market prices as a cycle has a higher trough than its old peaks.
A1Supply can't catch demand before 2028▾
What must happen
Industry bit supply keeps growing slower than AI-led demand through CY2027, so prices stay high while Micron's new fabs are built.
Why Micron
The constraint is physical and self-reinforcing. HBM takes more wafers per bit than conventional DRAM: about 3:1 for HBM3E, moving toward ~4:1 at HBM4E (Sadana, Aug 2026). Every HBM bit therefore removes three to four DDR bits from supply. New nodes add fewer bits per wafer, so growth needs greenfield fabs. Micron's own ID1 fab outputs first wafers in mid-CY2027 and ID2 in late CY2028.
Evidence
Q3 DRAM bit shipments grew only low-single-digit % q/q while ASPs rose low-60s%. DRAM inventory is below 120 days [Q3 10-Q; call]. Management expects CY2026 industry DRAM bit growth of low-to-mid 20s%, and said on 10 Aug that CY2027 will be tighter than 2026.
What to monitor
DRAM ASP q/q, bit shipments q/q, inventory days.
A2The contracts are real cash and hold at the floor▾
What must happen
Strategic Customer Agreements (SCAs) are honoured through CY2030. Floors hold, and the committed deposits actually arrive.
Why Micron
Micron says it pioneered binding SCAs and has signed the most in the industry. The old one-year LTAs had no binding terms. The new ones are take-or-pay with "no contractual outs" and an evergreen structure (Sadana).
Evidence
16 SCAs signed, most with five-year terms. The floor-price RPO is ~$100bn on 14 of them. Customers have committed $22bn of deposits and financial commitments, ~$18bn of it cash, with ~$10bn expected in FQ4. Management says floor-price margins are "well above our peak quarterly margins in any past cycle" [Q3 10-Q; call]. For scale, FY2018 peak annual gross margin was 59% [FY2018 10-K].
What to monitor
FQ4 cash deposits, RPO and next-12-month RPO in the FY2026 10-K.
A3Micron keeps technology parity and its share of the scarcity▾
What must happen
Micron stays qualified at the leading edge (HBM4/4E, 1γ DRAM, G9 NAND) so it captures its share of the shortage profits.
Why Micron
Its execution record is the best in its history. It is also the only memory maker building front-end fabs in the US, which matters to customers who want supply resilience.
Evidence
1γ and G9 are on track to be Micron's highest-volume nodes. HBM4 has shipped more than $1bn, and its 12-high ramp is running twice as fast as HBM3E 12-high. Next-generation nodes reach volume in 2H CY2027 [call]. Micron targets an HBM share close to its DRAM share.
What to monitor
HBM4 yields and qualifications, and CMBU gross margin (83% in Q3).
A4The capex is pre-sold, and the peak cash is banked▾
What must happen
New capacity is filled under contract rather than built on speculation, and excess cash goes to shareholders instead of the next glut.
Why Micron
SCAs give Micron multi-year volume visibility before it places equipment orders. Management says it builds cleanrooms but installs tools "consistent with our view of medium-term demand" (Sadana).
Evidence
Net cash was $24.4bn and debt $5.7bn after $4.4bn of repayment in Q3. Micron plans to raise capital returns from 9 Dec 2026 and to return 100% of excess cash over time [call].
What to monitor
Net capex versus bit growth, and the pace of buybacks after December.
Why the market might be missing it: every past Micron peak ended the same way, so the market reasonably prices FY2026 as another one. What's new is a contract floor that is disclosed in a 10-Q and backed by ~$18bn of customer cash. If the floor margin really sits above 59%, the trough has moved up, and a model that applies the FY2023 pattern understates through-cycle earnings.
A1 — Supply can't catch demand before 2028
Industry bit supply keeps growing slower than AI-led demand through CY2027, so prices stay high while Micron's new fabs are built.
Evidence: Q3 DRAM bit shipments grew only low-single-digit % q/q while ASPs rose low-60s%. DRAM inventory is below 120 days [Q3 10-Q; call]. Management expects CY2026 industry DRAM bit growth of low-to-mid 20s%, and said on 10 Aug that CY2027 will be tighter than 2026.
vs
Bear risk 1 — The supply response is already being funded, and some of it needs no new fab
Micron's net capex goes from ~$13.9bn in FY2025 ($15.86bn gross less $2.01bn of incentives) to ~$27bn in FY2026 and "above $45bn" in FY2027 [FY2025 10-K; call; Sadana]. Peers and state-backed CXMT and YMTC have the same incentive. There is also hidden supply in the HBM trade ratio. Micron's own 10-K warns that if HBM demand weakens and suppliers shift capacity back to conventional DRAM, conventional supply could rise significantly. The 3–4 DDR bits given up for each HBM bit can come back without building anything.
Confirms: A negative DRAM ASP quarter, HBM pricing softening, or competitors' capex running ahead of bit demand. Management already expects a "meaningful moderation" in the rate of price increases in FQ4.
Permanence: Permanent once capacity is installed: it depreciates whether or not prices hold.
Bear risk 3 — Demand is more elastic than the story allows
Customers are already buying fewer bits at these prices. MCBU revenue rose 49% q/q on lower bit shipments [Q3 10-Q]. Management cut its assumption for server DRAM content growth to allow for higher unit growth, and PC and phone unit volumes are declining [call]. Part of 2026 demand may also be inventory building ahead of contracted supply.
Confirms: Bit shipments falling while ASPs rise, spreading from MCBU into CMBU/CDBU (61% of Q3 revenue), or pauses in hyperscaler capex.
Permanence: Fixable if it is only allocation-driven de-specing, which management says will reverse once supply is available. Permanent if customers re-architect around less memory.
A2 — The contracts are real cash and hold at the floor
Strategic Customer Agreements (SCAs) are honoured through CY2030. Floors hold, and the committed deposits actually arrive.
Evidence: 16 SCAs signed, most with five-year terms. The floor-price RPO is ~$100bn on 14 of them. Customers have committed $22bn of deposits and financial commitments, ~$18bn of it cash, with ~$10bn expected in FQ4. Management says floor-price margins are "well above our peak quarterly margins in any past cycle" [Q3 10-Q; call]. For scale, FY2018 peak annual gross margin was 59% [FY2018 10-K].
vs
Bear risk 2 — The contracts cap more upside than they protect
The floor RPO is ~$100bn over a term that runs to end-CY2030. By my arithmetic that is roughly $20–25bn a year, against FQ4 guidance of $50bn for a single quarter. That puts roughly a tenth of the current run-rate under a floor. Meanwhile, ~40% of revenue is expected to end up at fixed prices or with ceilings at CQ2-2026 levels, so the upside is capped on a much larger share than the downside is protected. The deposits are refundable financing inflows, returned in the back half of the term, not revenue.
Confirms: Realised prices converging to floors, SCA disputes (the 10-Q flags the risk of litigation to enforce purchase commitments), FQ4 deposits well below ~$10bn, or next-12-month RPO small relative to revenue.
Permanence: Permanent to the thesis. The contract terms are fixed through 2030.
A3 — Micron keeps technology parity and its share of the scarcity
Micron stays qualified at the leading edge (HBM4/4E, 1γ DRAM, G9 NAND) so it captures its share of the shortage profits.
Evidence: 1γ and G9 are on track to be Micron's highest-volume nodes. HBM4 has shipped more than $1bn, and its 12-high ramp is running twice as fast as HBM3E 12-high. Next-generation nodes reach volume in 2H CY2027 [call]. Micron targets an HBM share close to its DRAM share.
vs
Bear risk 4 — The one cost lever is weakening
Management now projects blended DRAM cost per bit to rise, driven by LP6, DDR6 and HBM complexity and by greenfield start-up costs [call]. Node-driven cost-down was the only margin lever Micron fully controls.
Confirms: Cost of goods sold per bit rising faster than prices can be renegotiated for new products under the SCAs.
Permanence: Fixable, through premiums on new products that the SCAs allow to be negotiated.
A4 — The capex is pre-sold, and the peak cash is banked
New capacity is filled under contract rather than built on speculation, and excess cash goes to shareholders instead of the next glut.
Evidence: Net cash was $24.4bn and debt $5.7bn after $4.4bn of repayment in Q3. Micron plans to raise capital returns from 9 Dec 2026 and to return 100% of excess cash over time [call].
vs
Bear risk 1 — The supply response is already being funded, and some of it needs no new fab
Micron's net capex goes from ~$13.9bn in FY2025 ($15.86bn gross less $2.01bn of incentives) to ~$27bn in FY2026 and "above $45bn" in FY2027 [FY2025 10-K; call; Sadana]. Peers and state-backed CXMT and YMTC have the same incentive. There is also hidden supply in the HBM trade ratio. Micron's own 10-K warns that if HBM demand weakens and suppliers shift capacity back to conventional DRAM, conventional supply could rise significantly. The 3–4 DDR bits given up for each HBM bit can come back without building anything.
Confirms: A negative DRAM ASP quarter, HBM pricing softening, or competitors' capex running ahead of bit demand. Management already expects a "meaningful moderation" in the rate of price increases in FQ4.
Permanence: Permanent once capacity is installed: it depreciates whether or not prices hold.
3. Bear case — Munger invert
The most likely way I lose money is that the 2026 price spike funds an industry-wide capacity wave that lands in 2028 and later, while the contract floors protect only a small slice of today's revenue. Earnings then revert as they did in FY2023, just from a higher peak.
Attacks A1 + A4The supply response is already being funded, and some of it needs no new fab▾
How it could fail
Micron's net capex goes from ~$13.9bn in FY2025 ($15.86bn gross less $2.01bn of incentives) to ~$27bn in FY2026 and "above $45bn" in FY2027 [FY2025 10-K; call; Sadana]. Peers and state-backed CXMT and YMTC have the same incentive. There is also hidden supply in the HBM trade ratio. Micron's own 10-K warns that if HBM demand weakens and suppliers shift capacity back to conventional DRAM, conventional supply could rise significantly. The 3–4 DDR bits given up for each HBM bit can come back without building anything.
What would confirm failure
A negative DRAM ASP quarter, HBM pricing softening, or competitors' capex running ahead of bit demand. Management already expects a "meaningful moderation" in the rate of price increases in FQ4.
Damage to economics
Nearly all current profit is price, the most reversible kind. Depreciation from the capex wave lands in cost of goods sold just as price falls. FY2022 revenue of $30.8bn at 45% gross margin became $15.5bn at −9% in FY2023 [FY2023 10-K].
Permanent or fixable
Permanent once capacity is installed: it depreciates whether or not prices hold.
Attacks A2The contracts cap more upside than they protect▾
How it could fail
The floor RPO is ~$100bn over a term that runs to end-CY2030. By my arithmetic that is roughly $20–25bn a year, against FQ4 guidance of $50bn for a single quarter. That puts roughly a tenth of the current run-rate under a floor. Meanwhile, ~40% of revenue is expected to end up at fixed prices or with ceilings at CQ2-2026 levels, so the upside is capped on a much larger share than the downside is protected. The deposits are refundable financing inflows, returned in the back half of the term, not revenue.
What would confirm failure
Realised prices converging to floors, SCA disputes (the 10-Q flags the risk of litigation to enforce purchase commitments), FQ4 deposits well below ~$10bn, or next-12-month RPO small relative to revenue.
Damage to economics
Removes the durability premium. The business reverts to spot-cyclical economics with a thin floor, and the capped upside means the next spike pays less.
Permanent or fixable
Permanent to the thesis. The contract terms are fixed through 2030.
Attacks A1 (demand)Demand is more elastic than the story allows▾
How it could fail
Customers are already buying fewer bits at these prices. MCBU revenue rose 49% q/q on lower bit shipments [Q3 10-Q]. Management cut its assumption for server DRAM content growth to allow for higher unit growth, and PC and phone unit volumes are declining [call]. Part of 2026 demand may also be inventory building ahead of contracted supply.
What would confirm failure
Bit shipments falling while ASPs rise, spreading from MCBU into CMBU/CDBU (61% of Q3 revenue), or pauses in hyperscaler capex.
Damage to economics
Lower bit growth means the new fabs start up at low utilisation, with fixed cost and no volume to absorb it.
Permanent or fixable
Fixable if it is only allocation-driven de-specing, which management says will reverse once supply is available. Permanent if customers re-architect around less memory.
Attacks A3The one cost lever is weakening▾
How it could fail
Management now projects blended DRAM cost per bit to rise, driven by LP6, DDR6 and HBM complexity and by greenfield start-up costs [call]. Node-driven cost-down was the only margin lever Micron fully controls.
What would confirm failure
Cost of goods sold per bit rising faster than prices can be renegotiated for new products under the SCAs.
Damage to economics
Floor margins erode from the cost side even if floor prices hold.
Permanent or fixable
Fixable, through premiums on new products that the SCAs allow to be negotiated.
Any negative quarter confirms bear #1; flat with strong bits supports A1
A1 / Bear 1Q3 10-Q; call
DRAM bit shipments q/q
+low-single-digit % (Q3 FY26)
Bits falling while price rises = demand destruction; bits >10% with stable price = A1 working
A1 / Bear 3Q3 10-Q
SCA cash deposits received
$422mn contract liabilities; ~$10bn expected FQ4; ~$18bn total
Shortfall versus ~$10bn in FQ4 weakens A2
A2 / Bear 2Q3 10-Q; call
Floor RPO and next-12-month portion
~$5bn at Q3 (≈⅓ within 12m); ~$100bn expected for 14 SCAs
Next-12-month RPO as % of revenue measures how much is really floored
A2 / Bear 2Q3 10-Q; call
Consideration payable to customers
$3.32bn (vs $1.19bn Aug 2025)
Growing faster than revenue = hidden price concessions
Bear 2Q3 10-Q
Net capex
FY25 ~$13.9bn; FY26 ~$27bn; FY27 "above $45bn"
Rising faster than contracted bit demand = supply wave
A4 / Bear 1FY2025 10-K; call; Sadana
Gross margin
84.9% Q3; ~86% FQ4 guide
A decline toward the 59% prior annual peak would test the floor claim
A2 / Bear 1Q3 10-Q; call
MCBU bit shipments
Lower q/q (Q3 FY26)
Spreading to data-centre units = elastic demand
Bear 3Q3 10-Q
5. External challenge notes
Not run for this version. Write "external challenge" to add outside-source challenges (competitors, CXMT/YMTC, short-seller arguments) here without changing the rest of the memo.
6. Bottom line
The stock could work because AI has made memory a physical bottleneck that new fabs can't relieve before 2028, and Micron is turning that scarcity into five-year take-or-pay contracts backed by ~$18bn of customer cash.
Supply has to stay tight through CY2027, the contracts have to be honoured at floors that management says still beat every prior peak margin, and the capex ramp to above $45bn has to be filled under contract rather than on speculation.
The most important way it breaks is that floors cover only about a tenth of the current revenue run-rate while ceilings cap the upside, so an industry capacity wave in 2028 and later, including capacity released back from HBM, drives prices and margins down the way FY2023 did.
I would change my mind on a negative DRAM ASP quarter, FQ4 deposits well short of ~$10bn, or a FY2026 10-K next-12-month RPO that shows how thin the floor really is; the opposite readings on all three would strengthen the case.
Sources
Filings 4
Form 10-Q, Q3 FY2026 (quarter ended 28 May 2026), filed 25 Jun 2026: financials, SCA terms, RPO, deposits, ASP/bit commentary, risk factors
Form 10-K FY2025, filed 3 Oct 2025: FY2023–25 financials, net capex, HBM wafer-intensity risk, customer concentration
Form 10-K FY2023: FY2022–23 cycle (revenue, gross margin, net loss)
Form 10-K FY2018: FY2018 peak (59% gross margin, $14.1bn net income)
Transcripts 3
Q3 FY2026 earnings call, 24 Jun 2026 (Mehrotra, Murphy): SCAs, $100bn floor RPO, $22bn/$18bn deposits, FQ4 guidance, cost-per-bit outlook
KeyBanc Technology Leadership Forum, 10 Aug 2026 (Sadana): CY2027 tighter, HBM trade ratio 3:1→~4:1, no contractual outs
Six Five Summit, 27 Aug 2026 (Sadana): FY2027 capex above $45bn, US investment raised to $250bn
Not in sources 5
FQ4 FY2026 results and FY2026 10-K
Floor-price gross margin (not quantified by company)