NVIDIA's CEO, Jensen Huang, confirmed a significant business development during his visit to South Korea, revealing that the company's newly launched data center CPU, Vera, will incorporate DRAM memory chips from SK Hynix.
During a dinner in Seoul with SK Group Chairman Chey Tae-won, SK Hynix CEO Kwak Noh-Jung, and SK Telecom executives, Huang spoke to reporters about the close collaboration. He stated, "This year has been very important with SK Hynix. We are preparing for a very large volume of business in the second half and next year." He added, "We introduced the Vera CPU, a revolutionary CPU, and it will also use SK Hynix DRAM."
NVIDIA's First Standalone Data Center CPU
The selection of SK Hynix for the Vera CPU is a deliberate strategic move. The Vera CPU represents NVIDIA's most significant foray into the data center CPU market, which has long been dominated by Intel's Xeon series and Advanced Micro Devices' Epyc chips, and also competes with operator-designed chips like Amazon's Graviton.
Officially unveiled at Computex 2026, Vera is NVIDIA's first independent data center microprocessor, designed specifically for AI agents. It features 88 custom Olympus cores and offers a memory bandwidth of 1.2TB/s, reportedly completing tasks 1.8 times faster than traditional x86 processors. Huang described it as offering "the single-threaded performance improvement we haven't seen in 25 years."
Industry analysis suggests Vera's design logic fundamentally differs from traditional CPUs. While traditional CPUs focus on scaling core counts for simultaneous user access, Vera is optimized to allow AI agents to complete tasks faster, preventing expensive GPU idle time.
Leading AI labs such as Anthropic, OpenAI, and SpaceXAI, along with hyperscale cloud providers like ByteDance and CoreWeave, have plans to adopt Vera. System manufacturers including Dell Technologies, HPE, Lenovo, and Supermicro are also integrating Vera into their AI infrastructure.
Huang made a notable observation about the market: "Every product we create, on the day it's created, the market is $0. We are not trying to take anything from anyone. Six months ago, this market didn't exist." He was referring to the new demand for computing power generated by AI agents.
HBM4 Supply Chain Takes Shape
Huang's visit also included a key meeting with Samsung Electronics Vice Chairman Jun Young-hyun, expected to cover HBM (High Bandwidth Memory), next-generation memory technology, and robotics. Jun, who leads Samsung's memory chip business, has pivoted the company's strategy toward AI chips, with HBM as a core focus.
Just prior to his trip, Huang announced that Samsung, SK Hynix, and Micron Technology have all passed NVIDIA's qualification to become suppliers of HBM4, a core component for NVIDIA's next-generation Vera Rubin computing platform. All three are now in mass production to meet supply demands.
In the broader DRAM market, first-quarter 2026 revenue surged 81% sequentially, according to TrendForce. Samsung led with a 38.5% market share, followed by SK Hynix at 28.8% and Micron. With HBM4 supplier qualifications finalized, competition among these memory giants in the high-end AI memory segment is entering a new phase.
"We are collaborating with many industries, from AI supercomputers to central processors, to new types of personal computers and robots," Huang explained in Seoul, outlining the strategic intent of his visit. "So we are here today to make plans, and perhaps there will be some announcements tomorrow."
Strategic Implications of the Partnership
Analysts note that the high-profile nature of Huang's itinerary underscores the critical importance NVIDIA places on the South Korean semiconductor supply chain in the global AI computing race. For NVIDIA, securing stable, high-quality HBM supply is foundational for its expanding GPU business and AI computing ambitions. For Samsung and SK Hynix, securing major NVIDIA orders translates to significant profit growth and solidified industry standing.
Furthermore, Huang has indicated that future telecom networks will be used for AI, and he is in discussions with telecom companies. This points to NVIDIA's deeper strategic planning for integrated AI edge networking and computing infrastructure, suggesting its industrial collaboration network in South Korea and globally will continue to expand.
Huang is scheduled to conclude his meetings and depart South Korea next week for the next stop on his Asian tour. Local media have drawn parallels to his visit last year, suggesting that the close "barbecue and fried chicken" rapport between NVIDIA and South Korea's semiconductor industry is evolving into a more strategic and systematic competitive-cooperative landscape.
Comments