Art of Speculation
Chinese-Language AI Fundamentals & Macro, Concentrated Investing
Memory pricing power now sits with makers like Micron, as even Apple cannot absorb the costs. Micron rose 14% after 16 long-term agreements locked in prices for 3-5 years, which the market reads as possibly ending memory's boom-bust cycle; management sees shortages past 2027. Per JPMorgan, memory rises from under 20% of big-tech capex in 2022 to about 52% in 2026. If AI capex holds, $2,000 by year-end is reasonable; next year, once 2027 earnings are priced in, take care. x.comÂ
Micron's results show memory is no longer just a cyclical trade but a core asset with real long-term contracts and extraordinary margins. Fiscal Q3 revenue was $41.46 billion and adjusted EPS $25.11 versus $20.78 expected, with gross margin at 84.9%; Q4 is guided to $49.1-51 billion at about 86%. Sixteen supply agreements through 2030 are conservatively worth $100 billion, with $22 billion of cash prepayments. HBM4 ships in volume, HBM4E arrives in 2027, and tightness should last beyond 2027. x.comÂ
Micron and SK hynix pricing in a memory shortage past 2030 extends the memory-cycle case beyond the Street's 2028 expectation. SK hynix's record quarter adds evidence: per Wall St Engine, revenue rose 328% to $78.9B, operating profit rose 632% to $62.8B, and operating margin hit 80%. Agentic and physical AI could add demand as they roll out. x.com x.comÂ
Rising memory contract prices strengthen the case for memory stocks still far below their highs. JPMorgan forecasts Q4 DRAM prices up 13% and NAND up 18%; Micron is 28% below its 52-week high, Seagate 44%, Sandisk 47% and Western Digital 55%. x.comÂ
Networking, power and compute look more durable than cyclical memory in the AI buildout. Lumentum's revenue compounded year after year across the 2010s and 2020s, unlike Micron's; AWS revenue also compounded. Lumentum, Nebius, Bloom Energy and Credo should heavily outperform SanDisk and Micron from here, though AI itself remains heavily doubted. x.com x.comÂ
Each new AI wave is lifting demand across Micron's HBM, DRAM and storage products. Micron said training, inference, agentic AI and physical AI raise memory needs; the stock still trades at about 5x next-twelve-month earnings, leaving its growth story just getting started. x.comÂ
Micron's ordinary memory could see tighter supply because HBM consumes far more factory capacity than conventional DRAM. Meeting AI demand could thus extend Micron's opportunity beyond AI memory. x.comÂ
Inflation, rising rates, Chinese competition and rising AI-related financial costs are pressuring the memory market. Micron, Samsung and SK hynix look exposed; waiting for the cycle to turn. x.com x.comÂ
Agentic AI's repeated context favors cache memory and storage demand for Micron, SK hynix and Sandisk over fresh GPU computation. Per @demian_ai, about 95% of input tokens for DeepSeek V4.1 Flash on OpenRouter were cache hits; storage matters unless routing and caching keep almost everything in GPU memory. New products such as 3D DRAM and HBM-on-logic could be the final trigger for a full memory-stock re-rating. x.com x.comÂ
Apple is a victim of the AI memory boom: it is paying more for the same memory and storage chips that made Micron's quarter a blowout, and reportedly raised iPad and MacBook prices as a result. The shares fell 6.1% on the reports. In this supply-chain repricing, the picks-and-shovels names win and the device makers absorb the cost. x.comÂ
China's domestic market lets YMTC and CXMT threaten Micron, Samsung Electronics and SK hynix without entering US products. China receives 15-25% of smartphones and PCs and is on a path to at least 20-30% memory share over the next several years; current leaders are much too optimistic. History suggests at least a near-term semiconductor rally even if the AI trade has topped. It likely has not: agentic AI should give the trade room to run into 2027. x.com x.com x.comÂ
Improving earnings and more attractive valuations favor AI infrastructure names Nvidia, Micron, Broadcom and TSMC. Goldman Sachs charts put memory stocks at about 4 times two-year forward earnings versus an 8 times median, and large fabless chip stocks at about 13 times versus 18 times. x.comÂ
