AMD CEO Lisa Su plans a return trip to South Korea in October, according to a Digitimes report, with HBM4 memory supply and neural processing unit (NPU) cooperation at the top of the agenda.

The report offers no itinerary details, no confirmed meeting partners and no signed agreements. What it signals is continuity: Su has visited South Korea before, and October would mark a second round of face-to-face engagement with the country's memory and chipmaking giants within a short window.

Why does HBM4 dominate the agenda?

AMD's data center accelerator roadmap depends on high-bandwidth memory, and the next generation — HBM4 — is now the critical sourcing question for every AI silicon vendor. South Korea is home to the two largest HBM producers, SK Hynix and Samsung Electronics, and both are racing to qualify HBM4 for customer programs.

For AMD, the stakes are direct. Its Instinct-class GPUs compete against Nvidia's roadmap, and memory bandwidth per stack is one of the hard constraints on accelerator performance. An October visit by the CEO herself, rather than a supply-chain delegation, points to negotiations that only the top of the company can close: allocation volumes, qualification timing and pricing for a memory type that is still ramping.

The report does not specify which Korean suppliers Su would meet. It frames the trip around "deepening" existing ties, which implies follow-up on discussions from her earlier visit rather than the opening of a new front.

What role do NPUs play in the talks?

The second stated focus is NPU cooperation. Neural processing units have moved from a PC-market differentiator to a strategic product category across client, edge and server segments, and Korean semiconductor and systems companies sit on multiple sides of that value chain.

Chip packaging is one obvious intersection. HBM4 stacks and NPU silicon both push advanced packaging capacity — through-silicon vias and substrate supply — and Korea's OSAT and materials base matters to both. Device-side integration is another: Korean electronics makers build the PCs, servers and handsets where NPUs land.

Digitimes does not name specific NPU partners or projects. The report's language ties the topic to Su's broader Korean agenda rather than to a single announced collaboration.

What remains unconfirmed?

Readers should treat the timing itself as provisional. The story rests on a single outlet's sourcing, and executive travel plans of this kind frequently shift. AMD has not issued a public statement on the trip, and neither SK Hynix nor Samsung has commented on any scheduled meetings.

There is also no indication, in the report, of:

• Contract volumes or value for HBM4 supply
• Qualification dates tied to specific AMD products
• Named Korean counterparties
• Any memorandum, LOI or formal partnership covering NPUs

What does the visit mean for the market?

If the October trip proceeds as reported, it will reinforce two trends already visible in the supply chain. First, AI accelerator vendors are locking in next-generation memory capacity earlier and at higher executive levels than in previous HBM generations. Second, Korea's memory producers are becoming counterparties in strategic silicon planning, not commodity suppliers reacting to orders.

For competitors, the signal is competitive. Deepening AMD–Korea engagement on HBM4 tightens the field of memory available to other accelerator programs during a period when advanced packaging and HBM capacity remain the industry's tightest bottlenecks.

Die Signal will track confirmations from AMD, SK Hynix and Samsung as they emerge.

This article is based on a Digitimes report; the visit details it describes remain unconfirmed by the companies involved.