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 

Higher HBM stack requirements favor Samsung Electronics shares, even as the broader semiconductor industry suffers and DRAM is in dire straits. Daishin Securities says plans for 4-high and 6-high HBM in low-end inference appear shelved, while Nvidia's 8-high Rubin option appears dropped amid base-die supply constraints and customers' preference for bandwidth. x.com 

Cerebras is heavily overvalued at 62.7x EV/sales, even though it can re-rate as HBM stays rationed across the GPU world. Its wafer carries 44GB of SRAM, skipping the HBM queue, the latest quarter grew 74% with a $25.4B backlog, and its customer list should broaden beyond two UAE-linked buyers. Adding makes sense only near $186. If G42 and MBZUAI slow their orders, that multiple has nothing under it. x.com 

3D-stacked DRAM could constrain bit output even more than today's HBM, preventing memory oversupply beyond 2030. HBM demand already diverts capacity from conventional DRAM; logic-integrated stacks could add pressure if they scale around 2029-30. That comparison depends on the architecture being understood correctly, and memory makers have not settled the DRAM process nodes. x.com 

Memory remains structurally scarce because SK hynix says conventional DRAM and HBM are practically sold out through 2029, while customers request roughly twice the HBM volume it can produce. Long-term agreements cover more than half of 2027 and 2028 capacity, and the company expects the supply-demand imbalance to persist through 2030. x.com  x.com 

Cerebras is too expensive to buy at current levels, even though its technology is fully legitimate. Wafer-scale SRAM could be a game-changer for AI inference, and the hybrid-bonded DRAM planned for the CS-6 (expected in 2029, per the quoted post) addresses its main bottleneck, memory capacity. The company checks many boxes and offers more optionality, but it is not yet a threat to Nvidia. x.com 

Memory supply looks constrained through at least 2030 because HBM, vertical HBM and vertical edge-inference memory require more wafers per bit than standard DRAM. Micron's roadmap shows that wafer pressure persisting as those products arrive; its illustrative product timing remains subject to change. x.com 

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.com 

Cerebras may lose demand as OpenAI appears to be shifting some workloads away from Nvidia and Cerebras. The concern is that Broadcom's Jalapeño chip is delivering stronger performance, potentially reducing demand for Cerebras, whose stock is back near 52-week lows. x.com 

Micron's bet on 3D DRAM over rivals' 4F2 could give it an edge as AI memory demand rises. Micron expects shortages for at least two more years: HBM per GPU rose 3.6x from H100 to B300, inference memory per server doubled in three years, and agent context grows about 30x annually. It plans further 6F2 nodes, 1-delta DRAM in 2H27 and 400-plus-layer G11 NAND in 2027; processor-stacked vHBM is targeted around 2028. If Micron is right, rivals are spending on the wrong path. Overweight Micron. x.com