AMD CEO Lisa Su traveled to Samsung to negotiate high-bandwidth memory (HBM) supply, and Samsung is asking for foundry orders in return, TechPowerUp reports.

The visit places two of the semiconductor industry's most consequential procurement questions in a single room: AMD's need for additional HBM capacity to feed its data center accelerator roadmap, and Samsung's effort to convert its memory business relationships into foundry business.

What does AMD want from Samsung?

AMD depends on HBM for the memory subsystems of its Instinct accelerators, the product line competing with NVIDIA's dominant AI GPUs. Each accelerator integrates HBM stacks mounted alongside the compute die, and demand for that memory has tightened sharply as AI infrastructure spending has grown.

Securing a second major HBM supplier matters for AMD because it reduces exposure to allocation constraints at existing memory partners. For Samsung, winning an AMD HBM contract would strengthen its position in a market where SK hynix has led on HBM supply to the AI segment and Micron has also secured qualified positions.

What is Samsung asking in return?

Samsung's condition, according to the report, is reciprocity: it wants AMD to place chip manufacturing orders with Samsung Foundry. The company has invested heavily in advanced logic process capacity and needs high-volume customers to fill it. Landing AMD as a foundry client would serve that goal directly.

AMD currently manufactures its leading-edge CPUs and GPUs at TSMC. Any shift of production, or any addition of Samsung as a second foundry source, would represent a significant change in AMD's supply structure.

Why bundle memory and foundry in one negotiation?

The proposed linkage reflects how leverage works across the memory and logic segments. Samsung can offer HBM allocation as an incentive; AMD can offer wafer orders as payment. For Samsung, a bundled deal converts one business line's strength into another line's order book.

For AMD, a package deal could secure memory supply on favorable terms, but it comes with a trade-off: moving any production away from TSMC, or splitting production across two foundries, carries engineering and yield considerations that AMD would have to weigh against the memory supply benefit.

What happens next?

Neither company has publicly disclosed contract terms, volumes, or timelines tied to the discussions. The key signals to watch:

• Whether Samsung appears on AMD's qualified HBM supplier list for current or next-generation accelerators
• Whether Samsung Foundry discloses AMD among its advanced-node customers
• How AMD balances any Samsung foundry engagement against its existing TSMC relationship

The meeting itself confirms that AMD treats HBM supply as a board-level procurement issue, and that Samsung sees an opening to pair its memory portfolio with its foundry ambitions while both companies chase the AI infrastructure build-out.