independent semiconductor analysis
حسب الانحياز · 2 takes across the edition
Reconstructs the 2022-2026 HBM qualification race stack by stack, arguing the Rubin qual is less a level playing field than it looks: SK Hynix's yield and prior-gen track record translate into a structural ~two-thirds allocation, with Samsung and Micron splitting the remainder rather than displacing it.
“Passing qual is not the same as winning volume; SK Hynix's prior-gen yield converts into a structural two-thirds of Rubin HBM4.”
Argues the Ascend bottleneck has shifted from SMIC logic wafers to HBM and advanced packaging: Huawei has die banks of Ascend silicon waiting on memory. Notes continued reliance on stockpiled and externally-sourced inputs despite the self-sufficiency narrative.
“Huawei's Ascend ramp is gated by HBM and packaging, not SMIC logic, die banks sit waiting on memory.”