🧠 Food for thought

Implications, context, and why it matters.

The AI memory boom is built on an unusually severe supply crunch

• SemiAnalysis calls the current memory shortage a “once-in-four-decades” event 1.
• Physics sits behind it. DRAM density gains have slowed for about a decade, so cost per bit no longer drops and output now leans more on new fabs plus process transitions 1.
• The squeeze has intensified. One recent report says SK hynix has roughly four weeks of DRAM and NAND inventory remaining 2.
• Another piece, citing internal documents, says the company expects tight availability for consumer-grade memory through at least 2028. It also says SK hynix has not confirmed this publicly 3.

AI’s memory appetite is tightening supply for other markets

• AI memory profits carry a trade-off for other industries. Rising AI demand and the industry shift toward high-bandwidth memory (HBM) can leave fewer conventional memory parts for other buyers 4.
• HBM takes more capacity to make, so as SK hynix commits more factory lines to it, conventional DRAM and NAND volumes can tighten 4.
• That backdrop boosts pricing leverage. Citigroup analysts forecast the average selling price of DRAM could rise 120% this year 4.
• Fresh plants are coming online, yet reports say much of the added output will go to AI-related demand, which limits quick relief for non-AI customers 5.