Samsung in Talks with Nvidia on Next-Gen Foundry and HBM Collaboration
0xBroomberg
Samsung co-CEO Jeon Young-hyun met Nvidia's Jensen Huang in Seoul on Monday to discuss next-generation chip foundry work and HBM4E/HBM5 memory, signaling Samsung's push to lock in a deeper, longer-term role in Nvidia's supply chain.
What did the two sides actually discuss?
The talks ran along two tracks: securing near-term HBM4 supply and co-developing next-gen high-bandwidth memory — HBM (high-speed stacked memory designed for AI chips) — including HBM4E and HBM5.
On foundry, Jeon said Samsung is in talks to manufacture Nvidia's autonomous-driving chips on 4 nm and 8 nm processes.
This means → the meeting was not ceremonial — it covered a full-scope partnership negotiation spanning both memory and foundry.
Why is the Groq chip a key bargaining chip?
Nvidia acquired Groq for roughly $20 billion last December and launched the third-gen LPU — language processing unit — LP30, confirmed to be built on Samsung's 4 nm process, with shipments planned for the second half of this year.
In plain terms = Samsung already holds the manufacturing order for Nvidia's latest AI inference chip — that is real foundry revenue, not a memorandum of understanding.
Market watchers had expected the next-gen LP40 might go to TSMC, but Jeon stated clearly that Samsung foundry intends to keep the contract.
Where does Samsung sit in Nvidia's supply chain?
Current collaboration spans two domains: autonomous-driving chip foundry and Groq AI accelerator foundry, layered on top of HBM memory supply.
Samsung pledged adequate supply of HBM4 and the low-power memory module SOCAMM this year, with HBM4E and HBM5 cooperation extending from next year onward.
This reflects a clear strategy: bundle foundry and memory into a dual-track offering to upgrade Samsung's role from secondary supplier to deep integration partner in Nvidia's chain.
Content is for reference only, not financial advice.