Jensen Huang, the chief executive of Nvidia, has moved decisively to quell market anxieties regarding the company’s appetite for high-bandwidth memory (HBM). During a high-profile visit to South Korea on June 5, Huang addressed rumors that the AI chip giant was planning to scale back its reliance on HBM due to persistent supply constraints. His message was clear: Nvidia is not retreating, but rather refining how it utilizes these critical components for its next generation of silicon.
The rumors were sparked by an analysis suggesting that Nvidia’s upcoming 'Vera Rubin' architecture might feature reduced memory capacities. This sparked a temporary panic in the semiconductor sector, wiping out over $100 billion in market value for Micron and causing sharp declines for South Korean giants SK Hynix and Samsung Electronics. Huang countered this narrative by confirming that all three major memory suppliers—Samsung, SK Hynix, and Micron—have been certified to provide HBM4 for Nvidia’s latest AI chips.
Industry analysts have since clarified that the perceived 'reduction' in memory may be a misunderstanding of system architecture rather than a pivot in strategy. While the CPU-side memory configurations in the Vera Rubin NVL72 rack might see initial adjustments, the core GPU memory capacity—the engine of AI performance—remains unchanged at 288GB of HBM4 per chip. Huang emphasized that while supply is indeed tight, the goal is to use memory 'wisely and skillfully' rather than using less of it.
Huang’s presence in Seoul underscores the strategic importance of the South Korean supply chain to Nvidia’s dominance in the AI era. Beyond memory, the CEO is reportedly meeting with leaders from Hyundai, LG, and NAVER to ensure that every link in the production chain is synchronized with Nvidia’s aggressive release cycle. With the Vera Rubin chips already in full production and slated for delivery in the third quarter of this year, the pressure to secure every available gigabyte of high-end memory is higher than ever.