SK Hynix Inc.’s shares slid 4.7% after its record quarterly profit and modest capex plans failed to impress investors betting on insatiable AI demand for advanced memory chips.
Nvidia Corp.’s main supplier of high-bandwidth memory reported a more than 20-fold jump in operating profit to 8.08 trillion won ($5.6 billion) in the December quarter on a 75% rise in revenue, in line with analysts’ estimates. The South Korean company also boosted its annual dividend by 25% to 1,500 won a share, and said it expects HBM sales to more than double this year.
The numbers underscore how SK Hynix remains at the forefront in the HBM that Nvidia accelerators need to train AI—at the expense of longtime rival Samsung Electronics Co. They may also help assuage concerns that a global AI-fueled datacenter spending spree by Big Tech companies such as Microsoft Corp. and Meta Platforms Inc. has peaked.
That boom got a new boost this week, however. SoftBank Group Corp., OpenAI, Oracle Corp. and Abu Dhabi-backed MGX said they plan to set up a $100 billion venture to bankroll and build datacenters, targeting $500 billion worth of investments over four years. News of the Stargate project, promoted by President Donald Trump, fueled a rally in AI players from Nvidia to Arm Holdings Plc.
Still, the Korean chipmaker’s stock fell by its most in a week, shaving off some of the 30% gain SK Hynix had racked up so far this year. That recent stock rally means it traded before Thursday at a price-to-book ratio of roughly 2.8, well above its three-year average of 1.6.
“SK Hynix will be a big beneficiary” of Stargate, said Sanjeev Rana, an analyst at CLSA Securities Korea. “They have a significant lead, whether in terms of product quality of production yields…It will take some time for competitors to catch up.”
Well ahead of Samsung and Micron Technology Inc. in designing and supplying HBM, SK Hynix said it now plans to ship cutting-edge 16-layer HBM4 chips in the second half of next year. But the company, which said its capital spending would increase slightly this year, also needs to stay ahead of Samsung.
HBM made up 40% of its overall DRAM chip revenue in the quarter. Demand for such high-end memory will continue to increase as investment in AI servers grow and inference—making predictions from data—grows in importance, the company predicted. While the Korean firm expects inventory adjustments in the consumer market, sales of PCs and smartphones equipped with AI features will expand, fueling a second-half pickup, it said.
SK Group Chairman Chey Tae-won told reporters at CES in Las Vegas that SK Hynix has quickened its pace of development to keep up with Nvidia’s demands. During that trip, he met Nvidia co-founder Jensen Huang to discuss ways to deepen their relationship.
The Korean company is now racing to supply customers with its next HBM offering this year. The company will increase capital spending slightly in 2025, it said last year said it’s investing about $15 billion in South Korea to meet surging demand for high-end chips, on top of a plan to spend $3.9 billion on an advanced packaging plant and research center for AI products in Indiana.
Samsung has said it plans to mass-produce HBM4 chips in the second half of 2025, seeking to catch up to its smaller rival with the more advanced products. On Wednesday, Samsung unveiled its latest Galaxy S25 marquee, promising improved AI capabilities.
