AMD CEO Lisa Su confirmed the company is collaborating with Samsung Electronics and SK hynix on HBM4, the next generation of high-bandwidth memory, as part of what she calls a semiconductor industry that runs on partnership rather than solo effort.

"Semiconductors are a team sport," Su said, framing AMD's roadmap as dependent on a broad coalition of suppliers, foundries, and memory makers spanning several countries. The HBM4 engagement with the two Korean memory leaders sits at the center of that approach.

Why does HBM4 matter to AMD?

High-bandwidth memory is the critical companion to AI accelerators, and AMD competes directly with Nvidia in a market where memory supply and specifications can decide product competitiveness. By working with both Samsung and SK hynix — alongside incumbent supplier relationships in the HBM market — AMD aims to secure supply and influence the HBM4 specification for its next-generation data center GPUs.

For Samsung and SK hynix, design collaboration with a top AI chip customer early in the HBM4 cycle carries commercial weight. SK hynix has led the HBM segment in recent generations, while Samsung has pushed to win back share; an AMD engagement strengthens both suppliers' positions as the industry transitions beyond HBM3E.

How large is the investment?

Su pointed to tens of billions of dollars in global semiconductor investment tied to building out this ecosystem, spanning fabs, packaging, and memory capacity across multiple regions. She did not break the figure down by company or project in her remarks.

The scale reflects the capital intensity of AI compute: advanced logic nodes, CoWoS-class packaging, and HBM production each require multi-billion-dollar commitments, and no single company controls the full stack.

What does this mean for supply dynamics?

Dual-sourcing HBM4 across Samsung and SK hynix gives AMD leverage on pricing and volume in a memory market that has been supply-constrained since the AI boom began. It also deepens the Korea–US technology alignment in advanced memory, at a time when governments on both sides are pushing domestic capacity expansion.

Su's comments signal that AMD intends to compete in flagship AI accelerators with full memory roadmap alignment, not just chip design. As HBM4 moves toward production, the depth of these supplier engagements will help determine whether AMD can close the gap with Nvidia in the high-end AI GPU market.