"This is difficult, but I believe in you all!" Huang Renxun hosted a dinner for SK Hynix engineers last week, personally toasting and urging "no delay in delivering HBM4."

NVIDIA CEO Jensen Huang personally hosted a dinner last week to entertain the engineering team of a partner company. This rare move highlights the strategic importance of the next-generation high-bandwidth memory HBM4 for this AI chip giant. As Samsung Electronics takes the lead in shipping HBM4, whether SK Hynix can deliver high-performance products on time will directly impact the market performance of NVIDIA’s new AI accelerator Vera Rubin set to launch later this year.

On the evening of February 14, Jensen Huang appeared at a Korean fried chicken restaurant near NVIDIA headquarters, spending about two hours serving烧啤 (barbecue beer) and toasting to over 30 SK Hynix and NVIDIA engineers. He repeatedly emphasized “We are a team” and “I am proud of you,” urging engineers to “deliver extraordinary results through relentless challenge and effort,” especially mentioning SK Hynix’s commitment to supply the sixth-generation HBM4 products.

This impromptu gathering sends a clear signal: HBM4 is regarded by NVIDIA as a key differentiating component for the Vera Rubin accelerator. NVIDIA has set specifications for HBM4 suppliers requiring “operating speeds above 11 Gbps” and “bandwidth over 3.0 TB/s,” exceeding competitors like AMD by more than 30%. Industry analysts believe that Huang’s personal appearance increases the likelihood that SK Hynix will maintain its position as the largest HBM4 supplier.

Industry sources reveal that NVIDIA initially allocated this year’s HBM supply share in early December last year, with SK Hynix receiving over 55%, Samsung Electronics accounting for more than 20% up to nearly 30%, and Micron about 20%. Although Samsung’s progress in the technical race may cause share fluctuations, industry consensus suggests SK Hynix is expected to secure the maximum quota after quality optimization in the first quarter.

Rare “Engineer Diplomacy”

The semiconductor industry views Jensen Huang personally hosting a dinner for partner engineers as highly unusual. This event was quickly arranged after Huang’s last-week directive for NVIDIA staff to “organize a dinner to encourage SK Hynix HBM engineers,” highlighting the critical role of SK Hynix’s HBM4 in NVIDIA’s future business.

Huang arrived at the 99 Chicken restaurant in Santa Clara around 5:20 PM. During the final toast near the end of the dinner, he stated, “AI accelerators and HBM4 represent extraordinary, some of the most challenging technologies in the world. I am proud of all of you working tirelessly night and day, and I believe you will deliver outstanding results.” He added:

“I know the schedule for HBM4 and Vera Rubin is tight, but I believe in you. Now is the moment for SK Hynix and NVIDIA to jointly showcase greatness to the world.”

SK Hynix officially entered NVIDIA’s supply chain in July 2020 with its HBM2E (third-generation) products, subsequently becoming the de facto exclusive supplier of HBM3 (fourth-generation) and HBM3E (fifth-generation), forming a close relationship with TSMC known as the “AI semiconductor tripartite alliance.”

Technical Barriers and Time Pressure

HBM4 is seen as a critical component determining the performance of NVIDIA’s next-generation AI accelerator Vera Rubin. Scheduled for release in the second half of this year, HBM4, made from stacking 12 layers of advanced DRAM chips, is responsible for providing massive data to the GPU for processing in a timely manner.

NVIDIA’s specifications for HBM4 far surpass those of competitors. Its required operating speeds and bandwidth are more than 30% higher than AMD’s requirements, positioning HBM4 as a key differentiator for Vera Rubin.

Unlike the HBM3E market, which is largely dominated by SK Hynix, the HBM4 market opening in the second half of the year will feature different competitive dynamics. Samsung shipped the first official HBM4 products to NVIDIA on February 12, becoming the industry’s first, with products reaching 11.7 Gbps (up to 13 Gbps) and bandwidth of 3.3 TB/s.

SK Hynix has already ensured HBM4 performance reaching 11.7 Gbps or higher, is supplying paid samples to NVIDIA in bulk, and is optimizing performance. Industry expects SK Hynix to soon receive NVIDIA’s formal approval for “mass production” supply.

At the dinner, Huang urged SK Hynix engineers to “deliver top-tier HBM4 without delay.” This statement is interpreted as a clear demand for meeting supply timelines.

Intensified HBM Share Competition

As Samsung advances in HBM4 technology, the share distribution this year may shift. However, industry consensus holds that SK Hynix still has a good chance to secure the largest quota after completing quality optimization in the first quarter.

A semiconductor industry insider explained to the media:

“The final optimization work for the three memory manufacturers’ HBM4 products will likely be completed around March. Recently, there’s also been a trend to prioritize increasing general-purpose DRAM production to improve profitability rather than competing for HBM market share.”

According to a previous Wall Street Journal article, Evercore’s research pointed out differences in product generation evolution, emphasizing the distinction between “minor” (small incremental) and “major” (significant) upgrades. The transition from H100 to B100/B200 is seen as a minor evolution, while the move to GB-Series rack-level systems is a major evolution. The transition from GB-Series to VR-Series is minor, but the leap to Rubin Ultra, with cluster scale expanding from 144 to 576, is a major upgrade.

This dinner hosted personally by Huang Huang is both a recognition of long-term partners and a prompt for key supply chain links. For SK Hynix, whether it can maintain its leading position amid fierce competition will depend heavily on execution in the coming months.

Risk Warning and Disclaimer

Market risks exist, and investments should be cautious. This article does not constitute personal investment advice and does not consider individual users’ specific investment goals, financial situations, or needs. Users should consider whether any opinions, viewpoints, or conclusions herein are suitable for their particular circumstances. Investment is at your own risk.