Key Takeaways:
Nvidia's decision to use SK Hynix memory in its first standalone data-center CPU marks a deepening of the two companies' ties as AI infrastructure demand outstrips supply chain capacity.
Nvidia Corp. will use memory supplied by SK Hynix Inc. in its new Vera data-center processor, Chief Executive Jensen Huang said Sunday, cementing a multiyear partnership that locks in supply for the chipmaker's AI factory buildout through 2027. The Vera processor, unveiled at Computex Taipei this month, is Nvidia's first standalone central processing unit designed for data centers, directly challenging Intel Corp.'s Xeon and Advanced Micro Devices Inc.'s Epyc lines as well as Amazon.com Inc.'s Graviton series.
"The collaboration between Nvidia and SK Hynix is projected to experience significant expansion from the latter half of 2026 continuing through 2027," Huang said during a weekend visit to Seoul, where he met with SK Group Chairman Chey Tae-won and SK Hynix Chief Executive Kwak Noh-jung. The two companies plan to present their formal partnership strategy to media on Monday, according to a person familiar with the matter. The official announcement, described as a "multiyear technology partnership," covers co-development of next-generation memory for AI Factories and expanded supply commitments.
The Vera processor will incorporate SK Hynix DRAM, though specific memory configurations — including whether it uses the latest HBM4 (high-bandwidth memory, the fastest type of DRAM for AI workloads) — have not yet been disclosed. Nvidia's current H100 and B200 graphics processors already use SK Hynix HBM3 memory. The partnership extends beyond CPUs into AI supercomputing systems and robotics implementations, Huang said. SK Hynix shares on the Korea Stock Exchange surged following the disclosure, extending gains of more than 220 percent in 2026, according to exchange data.
Supply constraints will persist for years
Huang offered a blunt assessment of the semiconductor supply chain, saying shortages affecting everything from silicon wafers to advanced packaging materials and silicon photonics components will remain problematic. "It is going to persist for several years," he said, reinforcing that demand from cloud service providers and enterprise organizations currently exceeds the industry's production capabilities.
The bottleneck extends beyond memory. Nvidia relies on Taiwan Semiconductor Manufacturing Co. for its advanced chips, using CoWoS (chip-on-wafer-on-substrate) packaging — a specialized technique that stacks chips vertically to improve performance — which has been a persistent constraint. TSMC has been expanding CoWoS capacity but has said it will take through 2027 to fully meet demand.
For SK Hynix, the partnership provides a locked-in demand channel as the memory maker competes with Samsung Electronics Co. and Micron Technology Inc. to supply the AI boom. Huang's Seoul itinerary includes meetings with Samsung, Hyundai Motor Group, and LG Group, signaling Nvidia's broader push into Korean supply chains.
What it means for investors
Nvidia shares closed Friday at $205.10, giving the company a market capitalization of $4.97 trillion and a forward price-to-earnings ratio of 22.9. The stock has gained more than 170 percent over the trailing 12 months. SK Hynix trades on the Korea Stock Exchange under ticker 000660.
The Vera CPU positions Nvidia to capture a slice of the $30 billion-plus data-center CPU market dominated by Intel and AMD, while the memory partnership with SK Hynix reduces single-source risk for Nvidia's core GPU business. For SK Hynix, the deal locks in pricing power at a time when memory supply remains constrained — a dynamic that supports margins for both companies through at least 2027.
This article is for informational purposes only and does not constitute investment advice.
