AMD is seeking what it calls a "broader partnership" with Samsung as it works to secure its memory supply, and Samsung reportedly hopes to convert that memory relationship into foundry orders for AMD's logic chips, according to Tom's Hardware.

The talks reflect two pressures converging at once. AMD, which designs CPUs, GPUs and data-center accelerators but owns no fabrication plants, needs reliable access to advanced memory — particularly HBM-class DRAM for its AI accelerators and GDDR for graphics products — at a time when leading memory makers are allocating nearly all of their high-bandwidth output to the largest buyers. Samsung, meanwhile, operates one of the world's largest logic foundries alongside its memory business, and wants deeper engagement with a major fabless customer.

What does AMD want from Samsung?

For AMD, the primary objective is supply security. The company already buys memory from multiple vendors, and a broader agreement with Samsung would add a second pillar to its sourcing strategy for high-bandwidth memory stacked on AI accelerators — a segment where capacity has been tight and allocation decisions by memory suppliers directly shape how many GPUs and accelerators a designer can actually ship.

A closer relationship with Samsung gives AMD a counterweight in negotiations with its existing suppliers and a hedge against allocation shortfalls. Tom's Hardware reports that AMD's framing is explicitly about securing memory supply rather than any single product line.

What does Samsung want in return?

Samsung's ambitions reportedly run beyond DRAM sales. The company, per the report, "hopes to turn its memory supply relationship with AMD into foundry" orders for logic chips.

That would be a significant win for Samsung Foundry, which has struggled to land large fabless customers for its most advanced process nodes and has watched TSMC capture the overwhelming majority of leading-edge logic demand, including AMD's CPU and GPU shipments. Winning AMD as a foundry customer — even for a subset of products — would validate Samsung's advanced-node roadmap and diversify its customer base beyond its internal System LSI division.

For AMD, any foundry diversification carries risk. Moving high-performance chip designs to a second manufacturer requires re-qualification, adds engineering cost, and could complicate the tightly coordinated product cadence AMD has built around its primary foundry partner.

Why now?

The memory market's center of gravity has shifted toward AI accelerators. High-bandwidth memory attached to GPUs and custom accelerators has become the scarcest component in the AI supply chain, and fabless designers that lack long-term supply agreements risk building silicon they cannot fully populate with memory. AMD's MI-series accelerator roadmap depends on steadily increasing HBM content per package, which makes memory partnerships strategically equivalent to foundry partnerships in importance.

Samsung, for its part, has been working to close the gap with SK Hynix in HBM qualification for major AI customers, and pairing memory supply with a broader strategic relationship strengthens its pitch.

What happens next?

Neither company has confirmed the scope or timeline of the discussions, and it remains unclear whether any foundry engagement would cover leading-edge compute silicon or adjacent products. The direction, however, is clear: memory supply has become bargaining leverage, and Samsung appears willing to trade allocation priority for a shot at AMD's logic business — a dynamic that could reshape both the memory and foundry competitive pictures if a deal materializes.