Samsung bets on a memory-plus-foundry dual engine as Tesla and Nvidia knock
Seoul bilateral yields HBM4E samples and LP40 foundry talks; Tesla A16 to be built at Taylor, Texas; Samsung pins HBM5 comeback on an in-house 2nm base die
Summary
Samsung Electronics is building a "dual-engine" model, memory and foundry both driving earnings, to recover ground lost to SK Hynix in the AI cycle. A June 8 Seoul bilateral with Nvidia put HBM4E samples in Nvidia's hands, sketched a roadmap through HBM5, and opened talks for Samsung Foundry to build Nvidia's next-gen Groq LP40 chip. On the foundry side, overflow from TSMC's strained capacity has brought Nvidia, Tesla and Qualcomm: Tesla's A15 is dual-sourced (Samsung + TSMC) and the A16 reportedly goes exclusively to Samsung's Taylor, Texas fab. Samsung's HBM3e struggled through 2025, a base-die redesign won qual only for select China-market accelerators, and it pins an HBM5 comeback on being first to demo the architecture with an in-house 2nm base die.
By the numbers
- June 8, Seoul Samsung-Nvidia bilateral; HBM4E samples delivered, HBM5 roadmap discussed.
- ~25-30%, Samsung's estimated share of Nvidia Rubin HBM4 (behind SK Hynix's ~60-70%).
- A16, Tesla's next-gen chip, reportedly exclusive to Samsung's Taylor, Texas fab.
- 2nm, node for Samsung's planned in-house HBM5 base die.
- ~$73B, reported scale of Samsung's 2026 semiconductor investment.
Why it matters
Samsung is the only company that is both a top-three memory maker and a leading-edge foundry, the dual engine is its structural advantage if it can execute. Winning Nvidia foundry work and an exclusive Tesla node would offset the HBM share it ceded to SK Hynix. The HBM5 base-die bet decides whether the Rubin-generation gap is cyclical or permanent.
What to watch
- Whether HBM4E qualifies for Rubin and lifts Samsung's allocation.
- The Nvidia LP40 foundry decision and Tesla A16 ramp at Taylor.
- HBM5 first-mover claims and 2nm base-die yields.