AMD CEO Lisa Su visited South Korea for the first time on March 18, meeting executives from Samsung Electronics and Naver to forge comprehensive semiconductor alliances spanning memory procurement, foundry contracts, AI chip supply, and joint technology development.
Under the agreements, Samsung Electronics will serve as the primary supplier of high-bandwidth memory (HBM) for AMD's latest AI chips, which Naver will then deploy to power Korea's highest-performing AI models and services. Industry observers said Su is moving to secure Korea's two leading AI and semiconductor companies as allies, building an independent ecosystem away from Nvidia.
Samsung Secures First-Supplier Status for HBM4
At Samsung's Pyeongtaek campus in Gyeonggi Province, Su signed a memorandum of understanding with Vice Chairman Young-hyun Jeon of the Device Solutions division and President Jin-man Han, head of the foundry business, to expand cooperation in next-generation AI memory and computing technologies.
Through the MOU, Su designated Samsung Electronics as the primary HBM4 supplier for AMD's new Instinct MI455 chip, choosing the Korean company over SK Hynix and Micron.
The Instinct MI455 is an AI chip AMD developed with inference performance ten times greater than its predecessor to compete against Nvidia's Vera Rubin. As AI chip competition has shifted to securing HBM—the core component for computation—AMD has joined Nvidia in selecting Samsung, which has achieved industry-leading performance, as its top partner.
Samsung developed HBM4 using the industry's most advanced 1c DRAM and 4nm base die processes, beginning mass production shipments last month. By securing AMD as its second major tech customer for HBM4 after Nvidia, Samsung has established a foundation to close the market share gap with SK Hynix.
Foundry Orders Expected
The two companies agreed to strengthen cooperation across all semiconductor areas, including foundry contracts and joint research on next-generation chips. Foundry orders are considered highly likely to materialize.
AMD has primarily relied on TSMC for AI chip manufacturing, but surging semiconductor demand has pushed TSMC's advanced process capacity to its limits, creating the need for new partners. Samsung, the world's second-largest foundry, has been absorbing demand from major tech companies seeking TSMC alternatives, recently taking on production of the new language processing unit Grok 3.
President Han, who had attended Nvidia's annual developer conference GTC 2026 in San Jose, California, on March 16, reportedly rushed back to Korea the next day to meet Su.
Su arrived at the Pyeongtaek facility around 2 p.m., toured the production infrastructure with Samsung guides, and said she was "very pleased to combine Samsung Electronics' advanced memory technology leadership with AMD platforms including Instinct."
President Han also attended a dinner between Samsung Electronics Chairman Jay Y. Lee and Su at Seungjiwon in Yongsan-gu, Seoul, where discussions on foundry orders continued.
The companies also agreed to strengthen cooperation in next-generation memory. AMD's architecture has already been applied to Samsung's Exynos smartphone application processor series, and the two are jointly developing processing-in-memory (PIM) technology. They plan to concentrate investment on advancing this research.
Naver Partnership Targets Chip Diversification
Earlier that morning, Su met Naver CEO Soo-yeon Choi at the company's headquarters in Seongnam, Gyeonggi Province, signing an MOU for "AI ecosystem expansion and next-generation infrastructure cooperation."
The companies plan to strengthen technical cooperation to build high-performance GPU computing environments optimized for Naver's large language model HyperCLOVA X, jointly advancing infrastructure technology for stable AI model operations. The goal is to enable Naver's AI models to run on AMD chips rather than Nvidia's through optimization and other technical collaboration.
"Cooperation with AMD will be a meaningful opportunity to secure technological diversity and enhance AI infrastructure competitiveness for Naver," Choi said.
Su responded: "Naver, with its world-class AI capabilities and cloud platform, is the ideal partner to innovatively implement AMD's next-generation AI GPU technology."
Industry observers said Su is partnering with Naver—a cloud operator and major server market player—to expand market share in the AI semiconductor market dominated by Nvidia.
"From AMD's perspective, securing Korea's leading AI company as a chip customer provides a reference case for global markets," an industry source said. "It's also notable that Su has scheduled a one-on-one meeting with Upstage, another AI powerhouse."
Aligned Interests Against Nvidia Dependence
Samsung and Naver share AMD's interest in reducing dependence on Nvidia. Samsung aims to capture the application-specific integrated circuit (ASIC) market, where competing tech giants are developing alternatives to Nvidia GPUs. According to Counterpoint Research, HBM demand for ASICs is projected to surge 35-fold between 2024 and 2028.
Naver also seeks to diversify chip procurement for its models. The company has already partnered with Intel on joint research to optimize the Gaudi AI chip for its models and previously attempted to develop a proprietary chip called Mach 1 with Samsung Electronics.
That evening, Su met with Chairman Lee for approximately three hours to discuss additional cooperation measures. On March 19, she is scheduled to meet Senior Presidential Secretary Ha Jung-woo to discuss cooperation with domestic semiconductor companies and government initiatives including the AI highway project. A meeting with Samsung Electronics President Tae-moon Roh of the Device Experience division is also planned.