SK Hynix-branded 'HBM Chips' snacks distributed to crowds outside Mapo-gu restaurant; Huang chants 'More HBM,' signaling tight supply
Nvidia CEO Jensen Huang sat down with the heads of South Korea's major conglomerates for a samgyeopsal and soju dinner at "Hyeongnim Jeoyo," a barbecue restaurant near Hongik University in Mapo-gu, Seoul, on Thursday — the first day of his visit to the country.
About 40 minutes into the meal, Huang stepped outside and handed out boxes of doughnuts and snack bags to onlookers and reporters. SK Group Chairman Chey Tae-won joined him in distributing the treats.
The snack the two men gave away was "Honey Banana HBM Chips," a semiconductor chip-shaped snack that SK Hynix launched last November.
The name "HBM Chips" is a double entendre: it stands for the first letters of "Honey Banana Mat Chips" in Korean, while also nodding to SK Hynix's high bandwidth memory — HBM — and the word "chip" in semiconductors. SK Hynix leads the AI memory chip market with its HBM products.
Nvidia has built its meteoric rise in the AI market by pairing SK Hynix's HBM with its own graphics processing units to produce AI accelerators sold worldwide.
The sight of the two companies' chiefs distributing SK Hynix's HBM Chips side by side was widely seen as a public display of their firm AI partnership.
SK Hynix had already drawn attention by handing out the same snack to attendees at GTC 2026, an Nvidia-hosted event held in San Jose in March.
Huang also chanted "More HBM" repeatedly in front of the crowd outside the restaurant — an indirect acknowledgment that current HBM supply is falling short of Nvidia's demand.
!["Honey Banana HBM Chips," a semiconductor chip-shaped snack developed by SK Hynix in partnership with convenience store chain 7-Eleven. [Provided by SK Hynix]](https://wimg.heraldcorp.com/news/cms/2026/06/05/news-p.v1.20251126.0586c30ab640449fa55bbe299ef54e76_P1.jpg)
Nvidia has begun mass-producing its next-generation AI accelerator "Rubin" using sixth-generation HBM4 sourced from SK Hynix, Samsung Electronics and Micron, but supply still falls short of surging demand.
Beyond HBM, Nvidia also requires large volumes of low-power LPDDR DRAM, making close cooperation with SK Hynix and other memory chip makers increasingly critical.
Speaking to reporters outside the restaurant Thursday, Huang said the next-generation AI superchip "Vera Rubin" would use large amounts of HBM, and that the "Vera" CPU would also require significant quantities of LPDDR5 memory.
He added that Nvidia's "RTX Spark" — the company's first AI PC superchip in 40 years — would likewise need large volumes of LPDDR5 memory, indirectly underscoring the urgency of stabilizing the memory supply chain.
joze@heraldcorp.com
jeongwan@heraldcorp.com
