Nvidia CEO Addresses Samsung's HBM Chip Challenges and SK hynix's Progress

Samsung to achieve HBM chip supply goal, but needs new design, says Nvidia CEO

Nvidia CEO Jensen Huang speaks during a press Q&A session at Fontainebleau Las Vegas, Tuesday (local time). Korea Times photo by Nam Hyun-woo Jensen Huang credits Samsung, SK hynix as key memory suppliersBy Nam Hyun-woo LAS VEGAS -- Nvidia CEO Jensen Huang expressed optimism, Tuesday (local time), about Samsung Electronics' efforts to supply its most advanced high-bandwidth HBM memory chip to his company. However, Huang stressed the need for the Korean chipmaker to "engineer a new design." During a press Q&A session on the sidelines of CES 2025, the artificial intelligence (AI) guru said Samsung is "working on" Nvidia's qualification tests for HBM3e and there is "no question they will succeed." "I have confidence that Samsung will succeed with HBM. I have confidence like tomorrow is Wednesday," Huang said. The remarks were made in response to a question about why he chose to highlight Micron's G7 memory chip, which is used in Nvidia's new RTX 50 series graphics cards, unveiled during his CES keynote speech. The RTX 50 series is powered by Micron's Blackwell-based AI technologies, and following Huang's comments, the U.S. chipmaker's stock saw a significant rise. "Remember, Samsung created HBM," he said. "Originally, the very first HBM memory that Nvidia has ever used was from Samsung." Nvidia CEO Jensen Huang gives a keynote presentation at CES 2025 in Las Vegas, Monday (local time). UPI-Yonhap Samsung Electronics has been struggling to get its HBM3e chips to pass qualification tests for Nvidia's AI processors, while its rival, SK hynix, has become the dominant supplier, providing 12-layer HBM3e chips to Nvidia. In October, Samsung said in an earnings conference call that it has achieved "meaningful progress" in meeting qualification requirements for a "major customer," assumed to be Nvidia. But the chipmaker has yet to receive an official verification from Nvidia for its eight-layer and 12-layer HBM3e chips. The challenges surrounding HBM have raised concerns about Samsung's competitiveness as the world's largest memory chipmaker. As HBM continues to account for only a small share of Samsung's chipmaking business, a decline in the traditional DRAM market has hurt the company's overall earnings. Against this backdrop, Huang expressed confidence that Samsung will recover and said that Nvidia's testing of its HBM3e chips will be completed soon. However, he emphasized that the company needs to develop a new design. "Korea is very impatient, which is a good thing," Huang said when asked about the delays in Samsung's chip tests. "But they have to engineer a new one, new design. But they can do it. They are working very fast and very committed to do it." So far, industry observers have speculated that the delay in qualifying Samsung chips is due to a lack of optimization with Nvidia's chip specifications, rather than an issue with the chip quality itself. However, Huang's comments suggest that further delays in Samsung's bid to supply Nvidia could occur unless there is a fundamental change in the memory chip design. Huang also noted that Samsung and SK hynix are two of Nvidia's largest suppliers, praising them as "excellent memory companies" which will "continue to succeed" given the high demand for memory chips in Blackwell architecture. During his keynote speech, Huang showcased the GB200 Grace Blackwell Superchip, which is powered by multiple HBM chips. After the Q&A session, Huang mentioned that he will meet SK Group Chairman Chey Tae-won during CES, but did not reveal any further details about the meeting.

SK chairman talks with Nvidia CEO about faster-than-requested HBM development

SK Group Chairman Chey Tae-won responds to questions from reporters at the group's exhibition pavilion for CES 2025 in Las Vegas, Wednesday (local time). Courtesy of SK Group Chey defends Huang's misstatement on Samsung, SKBy Park Jae-hyuk SK Group Chairman Chey Tae-won expressed confidence in SK hynix's bargaining power against Nvidia, saying that the chipmaker's development of high-bandwidth memory (HBM) chips has already outpaced the speed requested by the U.S. computing giant. He made the remarks during a press conference with Korean media at the CES 2025 tech trade show in Las Vegas, Wednesday (local time), confirming his meeting with Nvidia CEO Jensen Huang at the exhibition's venue that day. "Although Nvidia previously asked us to speed up the development of HBM, the recent speed of our development has been surpassing the client's requested pace," the SK Group chairman said. HBM is a crucial component for graphics processing units (GPUs) that power generative artificial intelligence (AI) systems. Amid the increasing use of AI in everyday life, SK hynix has remained a major supplier of HBM to Nvidia, while Samsung Electronics has been struggling to get its HBM chips to pass qualification tests for the U.S. firm's AI processors. A day before Chey's press conference, the Nvidia CEO also stressed Samsung's need to engineer a new design, although he expressed optimism about the Korean firm's efforts to supply its HBM to his company. However, Huang seemed unaware that both Samsung and SK produced graphic memory chips when he erroneously stated that instead of Samsung's graphic memory chip, Micron Technology's G7 memory chip would be used in his company's GeForce RTX 50 series GPU. Although the misstatement triggered concerns among investors about the possibility that the Korean semiconductor companies may lag behind their U.S. rival, the SK Group chairman defended the Nvidia CEO, saying that it is difficult for top executives to memorize all the details about their products. The Nvidia CEO later corrected his remarks, saying that Samsung is among the multiple suppliers of memory chips for his company's new graphic card. "To be honest, Nvidia just needs to be good at understanding computing and developing the most efficient solutions for computing," Chey said. SK Group Chairman Chey Tae-won, front row right, listens to an explanation at Samsung Electronics' exhibition pavilion for CES 2025 in Las Vegas, Wednesday (local time), along with Samsung Electronics CEO and Vice Chairman Han Jong-hee, front row left. Yonhap The chairman added that he also exchanged opinions with Huang regarding physical AI that enables autonomous machines to perceive, understand and perform complex actions in the real world. When asked about his impression of this year's CES, the chairman said that the exhibition showed the soaring use of AI in all types of devices, including robots. He also introduced SK Group's AI and data center business during his visit to the world's largest tech fair for the third straight year. Read MoreHome AI to drive earnings recovery, Samsung Electronics CEO saysSK chief expects Nvidia to maintain dominance in chip sector for at least 3 years In response to a question about the group's strategies toward Donald Trump's return to the White House, Chey maintained a cautious stance, saying that many of the incoming U.S. government's policies remain unknown. He also reiterated that there is no reason for his company to abandon the Chinese market, despite the recent withdrawals of many Korean companies from China, amid the intensifying political and trade disputes between the world's two largest economies. Related Stories Will SK, Nvidia boast stronger AI chip ties at CES 2025? 2024-12-31 01:00 | Tech SKT to launch AI agent service in US in March 2025-01-08 14:31 | Tech

SK hynix's HBM development speed surpasses Nvidia's requested supply pace: SK chairman

SK Group Chairman Chey Tae-won, center, visits SK's pavillion during the CES tech show, Jan. 8, in Las Vegas. Yonhap SK Group Chairman Chey Tae-won said Wednesday that SK hynix's development speed of high bandwidth memory (HBM) has now surpassed the supply pace requested by Nvidia, signaling increased negotiating leverage with the artificial intelligence computing giant. "Until recently, SK hynix's HBM development pace lagged behind Nvidia's requirements, so they urged us for faster progress," Chey told reporters during a press conference at the Las Vegas Convention Center on the sidelines of the ongoing CES 2025, discussing a meeting with Nvidia CEO Jensen Huang earlier in the day. "Currently, our development pace slightly surpasses Nvidia's (requested pace)," he said. "While this could change, we are now developing at a comparable pace." HBM, an advanced, high-performance memory chip, is a crucial component of Nvidia's graphics processing units (GPUs) that drive generative AI systems. SK hynix is a major HBM supplier to Nvidia with its industry-leading fifth-generation HBM3E chips. The Korean chip giant earlier announced that its planned 2025 production of HBM has been sold out, with Nvidia as the biggest customer. "(Huang and I) discussed and confirmed the HBM schedule established at the working level," he added. "The amount of supply for this year has been determined, though I don't remember the exact numbers." Looking beyond HBM, Chey emphasized SK Group's focus on AI data centers as a key future growth driver. This is part of a broader strategy to transform the energy-to-communications conglomerate into a high-tech AI technology firm. "I think energy solutions for AI data centers are critical," he said, citing key energy issues, such as power supply, energy efficiency and cooling. "In this regard, the AI data center business aligns closely with SK's business portfolio." In line with this strategy, SK Group's CES 2025 exhibition highlights its AI data center solutions and related technologies, showcasing its AI data center solutions and cutting-edge AI technologies. (Yonhap)

Nvidia CEO Jensen Huang hints at Samsung's qualification for HBM

Nvidia CEO Jensen Huang delivers a keynote address at the Consumer Electronics Show (CES) in Las Vegas, Nev., Jan. 6. AFP-Yonhap Jensen Huang, CEO of Nvidia on Tuesday hinted that Samsung Electronics will soon be able to supply its high bandwidth memory (HBM) chips to the U.S. AI chip giant. "They are working on it. They're gonna succeed. No question," Huang said in a press event on the sidelines of the ongoing CES 2025 in Las Vegas. "I have confidence that Samsung will succeed with HBM. I have confidence like, tomorrow is Wednesday." Samsung Electronics has been struggling to have its HBM3E chips pass Nvidia's quality tests, while its local chipmaking rival SK hynix recently began mass-producing industry-leading 12-layer HBM3E chips. Huang's comments are in line with the Korean tech giant's announcement in late October that it has achieved "meaningful progress" in meeting quality testing requirements for a "major customer," expected to be Nvidia, whose GPUs are essential for AI computing. Huang also expressed his strong confidence in Samsung Electronics' achievements in memory history, especially in the HBM field. "I have confidence that Samsung has succeeded," he said. "Remember, Samsung created HBM. Originally, the very first HBM memory that Nvidia has ever used was from Samsung." Huang emphasized Samsung's potential to overcome current challenges in HBM3E chip development. "They have to engineer a new design," he said. "But they can do it. They are working very fast. They're very committed to do it." HBM is integral to the future of computing, powering Nvidia's GPUs, which are essential for AI and other advanced applications. Nvidia relies on both Samsung and SK hynix as HBM suppliers. "As you know, SK and Samsung are two of the largest suppliers to Nvidia," he said. "I'm very confident that SK hynix and Samsung are excellent companies and excellent memory companies. So I expect them to continue to succeed." During CES 2025, which opened on Tuesday for a four-day run in Las Vegas, the Nvidia CEO hinted that he has plans to meet SK Group Chairman Chey Tae-won, who is expected to attend the annual tech show. "I think I'm going to (meet him). I'm looking forward to it," he said. (Yonhap)

Nvidia Says Samsung Faces Challenges With AI Memory Chip Design

(Bloomberg) -- Nvidia Corp. head Jensen Huang said that Samsung Electronics Co. has faced difficulties producing a new type of memory chip for AI systems, but he expressed confidence that the partner company would overcome the challenges. The technology in question -- the latest type of high-bandwidth memory, or HBM -- is a vital part of new artificial intelligence systems that feature Nvidia chips. Samsung has been slower than rivals such as SK Hynix Inc. in producing HBM that meets Nvidia's standards, and Huang acknowledged those challenges during a press briefing at the CES conference in Las Vegas Tuesday. "They have to engineer a new design," Huang, Nvidia's co-founder and chief executive officer, said at the event. "But they can do it. They are working very fast. They're very committed to do it." Samsung is the largest overall maker of memory chips, but it's fallen behind rivals in the lucrative AI market. That weighed on its latest quarterly results, released Wednesday morning, which fell well short of analysts' estimates. Sales of AI computing systems powered by Nvidia have exploded over the past two years. Nvidia's chips serve as the heart of the machines, which create AI models by bombarding the software with data. Memory chips are key to the operation because they supply the processors with information. Trying to increase the speed and capacity of memory -- and tightly integrate the components with processors -- has added new levels of complexity to production of the chips. Samsung has been working through those obstacles. "They're going to succeed -- no question," Huang said. "I have confidence that Samsung will succeed with HBM."

Nvidia Says Samsung Faces Challenges With AI Memory Chip Design

Nvidia Corp. head Jensen Huang said that Samsung Electronics Co. has faced difficulties producing a new type of memory chip for AI systems, but he expressed confidence that the partner company would overcome the challenges. The technology in question -- the latest type of high-bandwidth memory, or HBM -- is a vital part of new artificial intelligence systems that feature Nvidia chips. Samsung has been slower than rivals such as SK Hynix Inc. in producing HBM that meets Nvidia's standards, and Huang acknowledged those challenges during a press briefing at the CES conference in Las Vegas Tuesday.