Micron Turned $10,000 Into About $130,000 in 5 Years. Most of It Came in the Past 12 Months.

Micron reported fiscal Q3 2026 revenue of $41.5 billion, up 346% year over year, and net income of $28.2 billion as AI data-center demand drove memory pricing sharply higher; adjusted gross margin reached 84.9% versus 39% a year earlier. DRAM revenue rose 343%, primarily because average selling prices increased about 260%, while shipment volume grew just over 20%. Despite multi-year customer supply agreements, shares at roughly $975 trade near 6x expected fiscal 2027 earnings and remain 22% below their $1,255 52-week high, reflecting investor concern that cyclical memory profits could fall from current exceptional levels.

Analysis

MU is no longer being valued on a conventional trough-to-peak earnings framework; the low multiple is a market-implied probability that current pricing and margin structure mean-revert. The key underwriting variable is not AI memory demand alone, but whether contracted supply includes enforceable volume commitments, price floors, and meaningful cancellation penalties. Without those protections, multi-year agreements may merely defer—not eliminate—the normal inventory correction once customers have secured adequate buffer stock.

The second-order risk is that extraordinary profitability invites a capacity response across DRAM and HBM, including from Samsung Electronics (005930.KS) and SK Hynix (000660.KS). New capacity is unlikely to impair results over the next one to two quarters, given qualification cycles and HBM yield constraints, but it can reset fiscal-2028 earnings power sharply if bit-supply growth exceeds AI-server memory-content growth. MU's earnings sensitivity is therefore asymmetric: modest downside to shipment volume can be overwhelmed by falling ASPs and gross-margin deleveraging.

Near term, a 22% pullback from the peak is not itself a contrarian buy signal; it may reflect institutional reluctance to capitalize peak margins. The more constructive contrarian case is that HBM qualification barriers and customer concentration create a more oligopolistic market than prior commodity cycles, warranting a higher mid-cycle margin floor. That thesis is falsified by sequential ASP declines, elevated customer inventories, weaker HBM mix, or any material fiscal-2027 capex increase by MU, Samsung, or SK Hynix.

NVDA faces a mixed read-through: memory scarcity supports accelerator-system pricing and reinforces supply bottlenecks, but rising memory cost can constrain system volumes or pressure OEM economics. The cleaner beneficiaries of a sustained memory investment cycle are semiconductor-capital-equipment names such as LRCX and AMAT, although they become late-cycle shorts if producers collectively accelerate wafer-fab spending before demand visibility extends beyond contracted allocations.

Key Decisions for Investors

• Do not initiate a directional MU long solely on the headline multiple. Require confirmation in the next earnings release that DRAM and HBM ASPs remain sequentially positive and that contract terms provide price/volume protection; absent that evidence, treat MU as a watch item rather than a recommendation.
• For a 1-3 month tactical hedge against a memory-price reset, buy MU put spreads 10-15% below spot, funded only partially with calls above the prior high. The trade targets a guidance or ASP disappointment while limiting loss if tight supply persists; exit if management raises forward contracted-price coverage and capex remains disciplined.
• For a 6-18 month relative-value position, consider long LRCX or AMAT versus short MU in equal semiconductor-beta terms after evidence of broad DRAM/HBM capacity expansion emerges. The expected payoff is equipment order upside before incremental memory supply compresses MU's earnings multiple; stop out if industry capex remains constrained or MU sustains sequential pricing gains for two additional quarters.
• Maintain NVDA exposure separately from MU rather than using MU as an AI proxy. Reassess NVDA only if memory availability begins delaying rack shipments or if server OEM gross margins show memory-cost pass-through failure; those would be earlier indicators of AI infrastructure demand friction than MU's eventual earnings decline.