Samsung's Most Advanced Foundry Capacity Is Fully Booked Through 2027 — and It's Pitching 5nm as the Overflow Option

Samsung's more advanced 4nm foundry capacity is fully booked through 2027, driven by HBM4 base-die production and orders from Nvidia. With its most advanced available manufacturing node effectively sold out more than a year in advance, Samsung has been promoting its 5nm process as a lower-cost alternative for customers who can't secure 4nm slots — pushing demand onto an older node rather than turning customers away entirely.

The capacity crunch comes as Samsung's 4nm process yield has improved to approximately 80%, a substantial gain in process stability and manufacturing efficiency that supports running the already fully-booked capacity at higher effective output. Improving yield on a sold-out node is one of the few levers a foundry has to extract more usable output without adding physical capacity, which typically takes years to bring online.

Samsung's position sits within a broader HBM supply picture where it is the clear second supplier behind SK hynix. Industry estimates put SK hynix at roughly 60% to 70% of HBM4 volume allocated to Nvidia's Vera Rubin platform, with Samsung around 25% to 30% and Micron supplying the remainder — meaning Samsung's foundry capacity crunch is happening even as it holds a meaningfully smaller share of the highest-value HBM allocation than its larger rival.

A leading foundry's most advanced available node being fully booked more than a year in advance is a signal worth reading carefully: it means physical manufacturing capacity, not just memory chip supply, is a binding constraint on how fast the broader AI hardware buildout can scale — a different bottleneck than the memory-pricing story playing out separately at SK hynix and Micron.