Korean chip stocks rallied after Nvidia Chief Executive Jensen Huang said the company expects to sell twice as many chips next year as it does this year.
Huang Vows to Double Chip Sales, Citing Confidence in AI Demand
According to CNBC and other outlets on the 18th, Huang told reporters on the 17th ahead of an artificial intelligence summit in Scotland hosted by Britain's King Charles III that Nvidia expects to sell twice as many chips next year as this year.
Huang pointed to expanding AI investment around the world. "It's because AI contributes so much to many industries and many economies," he said, adding that in nearly every country Nvidia has entered, people want to invest in AI.
The "double" figure Huang cited refers to unit volume, not revenue. Nvidia sells not only data center GPUs but also central processing units, switch and optical networking chips, notebook chips and Jetson modules for robots and vehicles, so doubling chip shipments does not translate into doubling revenue. The company does not disclose its total chip shipments. Last October, Huang said Nvidia had shipped 6 million Blackwell GPUs over four quarters.
Nvidia has also issued aggressive earnings guidance. In its second-quarter results announced on the 26th of last month, the company projected fiscal 2028 revenue of about $673 billion, up 70% from a year earlier.
Samsung and SK hynix Rally as Next-Generation HBM Shift Accelerates
Huang's comments also gave a boost to Korean chip stocks. On the main board on the 18th, Samsung Electronics (005930.KS) rose more than 2% and SK hynix (000660.KS) gained around 5% during the session. Samsung Electronics closed at 261,000 won, up 8,500 won or 3.37% from the previous session, while SK hynix ended at 1.857 million won, up 112,000 won or 6.42%.
The market is focusing on the possibility that Nvidia's expanding sales will drive demand for next-generation high-bandwidth memory from Samsung Electronics and SK hynix. Vera Rubin, the high-performance GPU Nvidia recently began mass-producing, uses HBM4, the sixth generation of the memory. Rubin Ultra, the successor due next year, is expected to use seventh-generation HBM4E, while Feynman, slated for 2028, is expected to use HBM5.
While HBM3E has been the mainstay of memory makers' HBM revenue until now, the center of gravity is expected to shift to HBM4 and HBM4E from the second half of this year through next year. Because HBM prices rise 30% to 50% with each generation, a larger share of next-generation products could benefit profitability at Samsung Electronics and SK hynix.
If GPUs for AI data centers drive Nvidia's sales growth in particular, the pace of HBM supply expansion could quicken as well. In last month's earnings announcement, Nvidia said it had secured orders from major technology companies and that Vera Rubin would ramp up production faster than any product in the company's history.
More Orders, More Problems: Capacity Is Key to Winning HBM4
The challenge is supply. Samsung Electronics and SK hynix are expanding next-generation memory capacity, but analysts say supply pressure could persist through next year if demand from Nvidia and other AI companies grows faster than expected.
Rising HBM output can also affect supply of conventional DRAM. Concentrating limited capacity on higher-margin HBM leaves less room for standard DRAM production, which could tighten supply and demand across the memory market.
Market attention appears to be shifting from whether AI demand actually exists to whether producers can secure the capacity to meet it. "As market attention has shifted from AI-related demand itself to whether capacity can be secured going forward, expectations for the data center CPU business have come back into focus," Kim Rok-ho, an analyst at Hana Securities, told Yonhap News on the 18th.
An official in the semiconductor industry likewise said that as the market moves into an era of customized HBM, securing HBM4 is what will determine market leadership, adding that how quickly capacity can be expanded is the key question.
