Masahiro Wakasugi

AI-driven memory demand is so strong that the industry consensus points to a shortage in 2027 and potentially 2028 unless capex is accelerated now. Micron's strong earnings and 85% gross margin signal high profitability, and SK Hynix is expected to post similar strong results. Samsung and SK Hynix must expand capacity to capture demand, making the memory segment very attractive.

AI-driven memory demand is so strong that the industry consensus points to a shortage in 2027 and potentially 2028 unless capex is accelerated now. Micron's strong earnings and 85% gross margin signal high profitability, and SK Hynix is expected to post similar strong results. Samsung and SK Hynix must expand capacity to capture demand, making the memory segment very attractive.

AI investment is shifting from training to inference, requiring different memory and storage systems, broadening demand across DRAM and NAND. Memory companies like Samsung, SK Hynix and Kioxia are investing heavily to expand capacity, which is strongly positive for the materials and equipment supply chain. The entire semiconductor supply chain is in a 'hot' phase with sustained demand.

1. FACT: Analysts note that Nvidia's shift from AI training to AI inference will drastically increase demand not just for HBM, but for conventional DRAM, NAND flash, and advanced packaging (like hybrid copper bonding). 2. BRIDGE: The entire Asian semiconductor supply chain is being pulled higher by Nvidia's $1T forecast. While Samsung and SK Hynix are direct memory beneficiaries, TSMC is the foundational foundry manufacturing these advanced inference chips and packaging solutions. 3. VERDICT: LONG. TSMC remains the ultimate tollbooth for the exploding AI inference and advanced packaging market. 4. KEY RISK: Geopolitical escalation in the Taiwan Strait or a sudden cyclical downturn in non-AI semiconductor demand (smartphones/PCs).