Version 12

What

Why it matters

Open questions

• Will the White House grant Apple an exemption to buy CXMT DRAM? [18][17] The outcome determines whether Pentagon blacklisting is a durable competitive moat for Micron or porous to lobbying by major customers.

• CXMT reportedly won a $3 billion RAM deal from an unnamed customer [19] — is that deal in consumer DRAM or HBM, and does it demonstrate CXMT can compete in the AI data center segment Micron targets?

• BofA raised its Micron price target after Q3 [15], and retail bulls call the selloff a buying opportunity [23] — but Nillo argues the stock is already priced as if de-commoditization lasts indefinitely [26]. At what valuation does the bull case require perfection?

• Samsung and SK Hynix are pursuing their own 3-5 year supply agreements [20] — do Micron's specific contract terms (pricing floors, take-or-pay enforcement) still differentiate, or does the structure become table stakes across the industry?

Narrative

Timeline

• 2026-05-26: Micron crosses the $1 trillion market cap threshold driven by HBM demand from AI workloads. [47]
• 2026-06-08: Amazon announces a multibillion-dollar optical fiber deal with Corning; Milk Road AI publishes 'Save This' on Corning as an underappreciated AI infrastructure beneficiary. [1][48][2]
• 2026-06-08: Jensen Huang describes Qualcomm positively; Milk Road AI publishes 'Save This' on Qualcomm as an on-device inference play. [3][49]
• 2026-06-14: Milk Road AI publishes 'Save This' on Micron with a $4,000 price target, framing it as America's strategic monopoly on AI memory. [4]
• 2026-06-15: CXMT aligns with Huawei on domestic HBM3 production targeting end-2026; analysts warn expansion risks global price pressure. [39][40][50]
• 2026-06-23: Chanos argues the AI energy bull case rests on a temporary power constraint and investors should own AI models rather than infrastructure builders; position amplified across Business Insider, Yahoo Finance, CNBC, and YouTube. [29][28][32][33][34]
• 2026-06-24: Micron Q3 FY2026 earnings: CEO Mehrotra discloses 16 Strategic Customer Agreements covering $100B over five years with $22B in deposits, pricing floors, and take-or-pay terms; AI-driven shortages forecast beyond 2027. [10][30][31][11][12]
• 2026-06-24: Milk Road AI adds GE Vernova to the 'Save This' series, arguing turbine capacity is a structural AI asset because data centers require 24/7 baseload power the grid cannot supply fast enough. [5]
• 2026-06-26: Milk Road AI publishes CoreWeave defense citing $2.08B Q1 revenue, $99.4B contracted backlog, and Nvidia's $2B co-investment alongside a joint 5-gigawatt AI factory expansion. [6]
• 2026-06-29: Financial Times and Reuters report Apple is actively lobbying the White House for an exemption to buy DRAM from Pentagon-blacklisted CXMT, driven by AI-induced memory price inflation; reports cause a Micron stock selloff. [17][41][18][42][43]
• 2026-06-29: 24/7 Wall St argues Micron is largely insulated from Apple/CXMT risk because Apple's demand targets consumer-device DRAM not the HBM Micron sells into AI data centers. [21]
• 2026-06-29: Citi ranks DRAM supply as the single biggest bottleneck across all AI compute semiconductor categories; Samsung and SK Hynix pursue their own 3-5 year supply agreements with major tech firms. [14][20]
• 2026-06-29: Milk Road AI adds Marvell to the 'Save This' series: 18 active custom chip programs at the four largest hyperscalers, CXL market-share leader, $10B+ custom silicon target by FY2029. [7]
• 2026-06-30: Milk Road AI adds Meta and Nebius to the 'Save This' series; Nebius framed as the strategic beneficiary of Nvidia's roughly $40B in ecosystem investments designed to sustain GPU demand across a fragmented compute landscape. [8][9]
• 2026-07-01: Milk Road AI reiterates Micron $4,000 price target with an inference-memory-bound argument: GPUs sit idle more than 95% of the time during inference decode; Micron trades at 8x forward earnings on projected FY2027 EPS of $112 with HBM4 ramping at twice the prior generation's pace. [13]
• 2026-07-01: Reports indicate CXMT won a $3 billion RAM deal from an unnamed customer; Apple is described as a potential next customer for the blacklisted Chinese supplier. [19]
• 2026-07-01: BofA raises Micron price target following Q3 earnings; retail investors split between framing the Apple/CXMT-driven selloff as a buying opportunity and arguing the stock is already priced as if de-commoditization will last indefinitely. [23][26][15]

Perspectives

Tensions

• Milk Road AI and Mehrotra argue 16 five-year take-or-pay contracts de-commoditize memory pricing [10][12]; Apple is actively lobbying the White House to buy DRAM from blacklisted CXMT [17][18], showing a named customer seeking supply outside that structure. [10][12][17][18]
• 24/7 Wall St argues Apple's CXMT demand targets consumer-device DRAM (iPhone, Mac), leaving Micron's HBM and AI data center position intact [21]; ollobrains argues the AI buildout has broken the old consumer-memory pricing dynamic and Apple's historical leverage as a large buyer is gone [22]. [21][22]
• Milk Road AI calls Micron 'America's monopoly' on AI memory [4]; SK Hynix holds 58% of the HBM submarket as of Q1 2026 with Samsung and Micron each at 21%, and CXMT reportedly won a $3 billion RAM deal [12][19]. [4][12][19]
• Micron forecasts AI-driven memory shortages beyond 2027 [31]; CXMT targets HBM3 production by end-2026 [40] and reportedly already won a $3B RAM deal with Apple named as a potential next customer [19] — a supply timeline Micron's shortage forecast does not address. [31][40][19]
• Retail bulls call the Apple/CXMT-driven Micron selloff a buying opportunity [23][24]; Nillo argues the stock is already priced as if the de-commoditization change will hold permanently [26], and Briefing Block argues the ceiling may come from customer demand rather than production capacity [27]. [23][24][26][27]
• Chanos argues investors should own AI models rather than infrastructure builders [28]; Milk Road AI's eight-investment series now spans memory, fiber, power, compute, custom silicon, on-device inference, and neocloud — with Nebius framed as Nvidia engineering a compute ecosystem to sustain GPU demand indefinitely [9]. [28][9]