Nvidia Announces Q2 Earnings
Unveils Roadmap for Next-Gen AI Chips
Production of Blackwell and Blackwell Ultra Accelerates
Rubin Confirmed for Launch Next Year
Blackwell to Feature HBM3E, Rubin to Adopt HBM4
Samsung and SK Hynix Poised for Increased HBM Orders
Supply Chain Restructuring Expected with HBM4
New Developments for HBM in China-Bound H20 Chips
Nvidia, the largest customer for Korean semiconductor companies, drew industry attention on August 28 by announcing second-quarter (May to July) results that exceeded expectations, along with an optimistic outlook for next-generation products and hopes for increased exports to China. If Nvidia’s plans proceed as intended, positive prospects are emerging for Samsung Electronics and SK Hynix, both of which supply high-bandwidth memory (HBM) to Nvidia.
Samsung Electronics Chairman Jay Y Lee is greeting Jensen Huang, CEO of Nvidia, at the Korea-US Business Roundtable held on the 25th (local time) at the Willard Hotel in Washington DC, USA. Photo by Yonhap News
Nvidia’s management, including CEO Jensen Huang, expressed confidence during the company’s second-quarter earnings announcement and subsequent conference call, stating that its next-generation artificial intelligence (AI) accelerator products will proceed to mass production and launch as scheduled, maintaining strong momentum in the AI semiconductor market.
Colette Kress, Chief Financial Officer, confirmed that the ‘Rubin’ series currently under development “will be released next year as planned,” adding, “This will allow us to maintain our annual product launch cycle.” CEO Huang also stated, “Rubin will incorporate a host of new ideas. We will have plenty of time over the next year to explain the many innovations Rubin will bring. While I cannot share details right now, we will unveil them at the GTC event in October.”
Rubin is Nvidia’s upcoming next-generation AI accelerator series, set to follow Blackwell and Blackwell Ultra. It is expected to be the first to fully adopt sixth-generation HBM, known as HBM4. As demand for Rubin increases, the need for HBM will also rise significantly, leading to expectations that Samsung Electronics and SK Hynix will receive more orders.
Ahn Ki-hyun, Executive Director of the Korea Semiconductor Industry Association, commented, “Compared to recent trends, there were no notable changes in Nvidia’s latest earnings report. AI demand remains strong, and the outlook for our companies supplying HBM to Nvidia continues to be bright.”
However, foreign media and industry sources have suggested that Nvidia may restructure its supply chain starting with HBM4. While SK Hynix dominated supply up to the previous generation HBM3E, it is widely believed that Nvidia will make some adjustments for HBM4, increasing the volume supplied by Samsung Electronics and Micron.
Leading US securities firm Morgan Stanley noted, “SK Hynix’s share of Nvidia’s HBM supply, which reached about 90% this year, could drop to around 50% next year, with Samsung Electronics and Micron likely to fill the gap.” The firm provided specific figures, projecting Samsung Electronics at 25-30% and Micron at 20-25%.
All three companies-SK Hynix, Samsung Electronics, and Micron-have produced HBM4 samples this year and provided them to Nvidia, awaiting feedback on future supply contracts. Competition among the three to secure a larger share of Nvidia’s supply chain is expected to intensify going forward.
In the near term, Nvidia is also expected to significantly increase its procurement of previous-generation HBM products such as HBM3E from SK Hynix. CEO Huang emphasized that demand for Blackwell and Blackwell Ultra, which precede Rubin, remains extremely high.
He described Blackwell as “the AI platform the world has been waiting for, achieving tremendous advancements unlike anything before,” and said of Ultra, “Production is ramping up at maximum speed, and demand is enormous.” Both Blackwell and Blackwell Ultra are reportedly equipped with SK Hynix’s 12-layer HBM3E.
Plans for the Chinese market are also encouraging for Samsung Electronics and SK Hynix. Nvidia is seeking ways to export its H20 chip to China, and there is a high likelihood that the HBM used in the H20 will be supplied by Samsung Electronics or SK Hynix. Although several variables remain, it appears unlikely that Nvidia will easily abandon its efforts to export to China.
CEO Huang stressed, “If we respond to the Chinese market with competitive products, there is an opportunity to generate approximately 50 billion dollars in revenue this year.” He added, “After 50 billion dollars this year, we expect annual growth of 50% in subsequent years.”
Nvidia’s H20 chip, which had its exports to China restricted by the US Trump administration in April, received approval to resume sales last month. In exchange, Nvidia agreed to pay 15% of its Chinese export revenue as taxes to the US government. While the US government’s export approval process is still delayed and the Chinese government has instructed domestic companies to halt H20 purchases, most industry observers and foreign media believe these issues will soon be resolved.
A new phase may be emerging. Until now, Nvidia has sold H20 chips in China equipped with HBM3 supplied by Samsung Electronics, but it is reportedly considering upgrading to HBM3E for the Chinese market to improve performance. As a result, a new supplier for HBM3E may be designated; Samsung Electronics may continue to supply, or SK Hynix could be newly selected.
Meanwhile, Nvidia announced that it posted second-quarter revenue of 46.74 billion dollars (approximately 65.1555 trillion won) and earnings per share of 1.05 dollars (about 1,463 won). Both figures slightly exceeded Wall Street’s expectations of 46.06 billion dollars in revenue and 1.01 dollars in earnings per share, while revenue rose 56% compared to the same period last year.
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
