Nvidia CEO Jensen Huang on June 5 denied rumors that the company would reduce high-bandwidth memory (HBM) usage in its AI chips during an interview in South Korea, confirming instead that three memory suppliers have been qualified for HBM4 production. The denial came one day after research firm SemiAnalysis published a report suggesting memory capacity reductions in Nvidia's next-generation Vera Rubin NVL72 rack systems, which triggered speculation about supply constraints. Huang's response addresses ongoing concerns about AI chip component availability as Nvidia ramps production of its latest architecture, with full-scale manufacturing already underway and third-quarter delivery scheduled.
Nvidia Confirms Three Suppliers Qualified for HBM4 Production
When asked about rumors that Nvidia would reduce HBM usage due to supply constraints, Huang stated: "We will use large amounts of HBM memory. Of course, supply is currently limited. Therefore, we need to use memory more wisely across all systems. We will continue to work with partners here to secure as much supply as possible and use it as cleverly as possible."
Huang confirmed that three memory suppliers—Samsung Electronics, SK Hynix, and Micron Technology—have passed qualification and entered mass production to provide HBM4 for Nvidia's latest AI chips. He stated that all three suppliers "are competing to support the company's Vera Rubin architecture." Huang previously announced that Vera Rubin chips have entered full-scale production and are expected to deliver in the third quarter.
SemiAnalysis Report Details Memory Configuration Changes
On June 4, research firm SemiAnalysis published an article stating that the SOCAMM DRAM capacity (a new memory module standard designed specifically for AI servers) in Nvidia's next-generation flagship supercomputing rack Vera Rubin NVL72 may decrease from the previously expected approximately 55TB to approximately 28TB. The report also indicated that most Rubin systems will adopt 96GB SOCAMM modules rather than the previously market-expected 192GB.
Analysis of the SemiAnalysis report revealed that the primary memory changes originated from CPUs using LPDDR5X memory. Each Vera Rubin NVL72 rack contains 72 Rubin GPUs and 36 Vera CPUs, with each GPU using 288GB of HBM4 memory, totaling approximately 20.7TB for the entire rack—this portion remained unchanged. Since SOCAMM uses a socket design, it can be replaced with higher memory capacity at any time, meaning only the initial memory capacity of CPUs in Vera Rubin NVL72 shipping configurations was reduced.
SemiAnalysis founder Dylan Patel commented on X (formerly Twitter): "I love how people reposting our report left out most of the content. This happens all the time."
Memory Stocks Drop Following Supply Speculation
Following the SemiAnalysis report, some observers interpreted the news as Nvidia beginning to reduce memory usage due to shortages. Combined with custom AI chip (ASIC) giant Broadcom's earnings report on the same day failing to meet elevated investor expectations, global memory concept stocks experienced a pullback.
On June 4, Micron Technology (Nasdaq: MU) dropped over 10% intraday, erasing over $100 billion in market value in a single day. On June 5, South Korea's stock market experienced severe adjustments with semiconductor giant stocks suffering heavy losses. By market close that day, SK Hynix fell 9.92% and Samsung Electronics declined 6.4%.
Huang Meets Korean Partners During Supply Chain Trip
During the interview, Huang discussed his Korea trip itinerary, which included meetings with Hyundai Motor, LG, SK Hynix, Samsung Electronics, and Korean internet giant NAVER. He stated the core purpose was to ensure supply chain partners "stay aligned and fully prepared." Huang said he brought "massive business" to Korea and hinted at additional surprises yet to be announced.
FAQ
What did Nvidia CEO Jensen Huang say about HBM memory usage on June 5?
Huang denied rumors of HBM reduction during a June 5 interview in South Korea, stating Nvidia will use large amounts of HBM memory and will work with partners to secure as much supply as possible while using it wisely across all systems.
Which memory suppliers are qualified for Nvidia's HBM4 production?
Huang confirmed that Samsung Electronics, SK Hynix, and Micron Technology have all passed qualification and entered mass production to provide HBM4 for Nvidia's Vera Rubin architecture chips, which are expected to deliver in the third quarter.
Why did memory stocks drop on June 4 and June 5?
Memory stocks declined after SemiAnalysis published a June 4 report suggesting memory capacity reductions in Nvidia's Vera Rubin systems, leading to speculation about supply constraints. Micron dropped over 10% on June 4, while SK Hynix fell 9.92% and Samsung declined 6.4% on June 5.
