More memory CapEx may yield proportionally fewer bits as HBM consumes more wafer capacity, supporting SK hynix and Samsung's pricing. Both reported double-digit quarterly DRAM and NAND price gains; Samsung targets 60-70% of mid- to long-term capacity under LTAs. SK hynix has the stronger execution record, but Samsung could benefit enormously from conventional memory without winning HBM4. If trends persist, normalized margins could rise, though memory remains cyclical. x.com 

Samsung Electronics' HBM4 pricing looks set to lead the memory Big Three, helping make next year's earnings phenomenal. Samsung memory EVP Jaejune Kim said next year's HBM supply is sold out at target prices, with more customer requests coming. Korean media had reported quotes in the mid-to-high $4/Gb range; Nvidia appears to have accepted Samsung's asking price. Samsung-equipped Rubin variants should sell as separate premium-priced SKUs against SK hynix or Micron versions. x.com 

Nvidia's proposed common NVHBM base die would make Micron, SK hynix and Samsung more interchangeable, weakening their pricing power. Seoul Economic Daily reports Nvidia wants all three to fabricate it at TSMC; Samsung reportedly agreed. Standardization would ease validation and order shifts. The risk lessens if SK hynix or Micron retain their own logic or NVHBM stays a small share of Nvidia's purchases. x.com 

Samsung's HBM4 ramp makes SK hynix's ability to defend its HBM lead the key question in the memory race. Chosun Biz reports SK hynix's HBM4 shipments may have missed expectations for a second straight quarter amid Nvidia Rubin delays and chip tuning. Counterpoint puts Samsung's Q2 HBM share at 33%, up from 21%; TrendForce expects strong SK hynix Q3 profits. Samsung is the preferred memory stock right now, though there may be no single clear winner. SK hynix's October 28 report is the next check. … Show more

Shorter HBM stacks are not reducing demand for SK hynix's HBM, because customers are redesigning AI accelerators around limited supply. On its call, management said lower stacks let customers build more accelerators; supplying the same bit volume then requires more packages, base dies and back-end processing. Denser dies can offset some lost layers, and SK hynix says customer volume requests exceed twice its current supply. x.com 

HBM stack height should leave excess DRAM wafers, driving the current memory short. The short this time is specifically about HBM, not all memory. x.com 

TSMC gains leverage in custom HBM because Nvidia reportedly standardized its NVHBM base die on TSMC's 3nm process, even for Samsung's stacks. Seoul Economic Daily reports Samsung accepted while retaining its own 4nm base dies for standard HBM4/HBM4E and AMD custom HBM, and validating a fallback for Nvidia. That weakens Samsung's all-in-one memory, foundry, packaging and photonics pitch to Nvidia, though its bundle can still serve other customers. x.com 

HBM suppliers could earn more revenue and higher margins as yields improve at higher pin speeds. Nvidia wants more HBM and needs its suppliers to improve yield; the reported pin-speed reduction was most likely a yield issue, not evidence that SK hynix and Micron failed. x.com 

Samsung's record quarter looks closer to a peak than a new base: conventional DRAM shipments are flat year on year, while higher prices per bit drove profits despite a revenue miss. Its phone unit is cutting output because memory costs too much, and AI chip designers are seeking to use less HBM. Buyers' response threatens the price-led margin surge. The October 29 divisional earnings breakdown may show whether memory profit grew much faster than group profit. x.com 

Tight memory supply keeps the AI memory supplier trade in bill-of-materials names if it persists. The investor aleabitoreddit reports an SK hynix shortage outlook through 2030 and an operating-margin result near 80%. HBM allocation and average-selling-price guidance are the next tests; Korea has already priced the earnings print, so the next move depends on who locks capacity. x.com