Samsung posted record profits and still watched its stock fall. That contradiction is the entry point for this episode's argument: memory chips, not model weights, are becoming the scarce resource defining the AI buildout. The hosts break down how HBM, DRAM, and NAND are being consumed at a rate that strains fab capacity globally, with Samsung and SK Hynix holding near-duopoly pricing power as a direct result.
The mechanism is worth understanding in detail. HBM production cannibalizes the same fab capacity used for consumer DRAM and NAND, which drives up prices for phones, laptops, and SSDs even as AI data centers absorb supply. The episode walks through why new capacity cannot come online fast enough, the multi-year lead times on fab equipment, and why the oligopoly of suppliers means this is more likely a structural supercycle than a speculative bubble. SK Hynix's planned U.S. listing, which is already seeing outsized IPO demand, is framed as a signal of how seriously institutional capital is taking this thesis.
The sell-the-news dynamic after Samsung's earnings beat is examined through the lens of past memory cycles, and the hosts argue most investors are misreading the timeline. The case for reading the full transcript or listening to the episode is in the granular breakdown of supply constraints starting at the 11:43 mark and the forward-looking discussion of what cheaper, more abundant memory would actually mean for AI capabilities at 16:25.
[WATCH ON YOUTUBE →]