South Korea is seeking to prioritize the supply of Vera Rubin, with HBM4 capacity potentially becoming a bargaining chip in negotiations with NVIDIA Corporation.

South Korean Minister of Science, Technology and Information Communications, Bae Kyung-hoon, announced on Monday that due to delivery delays for NVIDIA's next-generation Vera Rubin graphics processing unit (GPU), South Korea will officially request priority supply in order to obtain the chip within this year. B300 on time, Vera Rubin delay triggers "grabbing goods" Bae Kyung-hoon stated that the South Korean government had issued a notification regarding the GPU project earlier that day, and expects the supply of the NVIDIA B300 chip to arrive on time. Bae Kyung-hoon told reporters at a press conference, "It looks like the supply of the B300 can arrive on time, but there may be a slight delay with Vera Rubin, so we will request priority supply for it." He revealed that South Korea is striving to obtain supply of the Vera Rubin GPU within this year. Just a few days ago, NVIDIA finalized Samsung, SK Hynix, and Micron as suppliers for the Vera Rubin AI computing platform's HBM4, all three have passed qualification audits and are in full production to ensure supply. This statement comes at a crucial moment as South Korea accelerates the development of its AI infrastructure. On Monday, the Ministry of Science, Technology and Information Communications of South Korea announced the selection of three cloud service companies - Naver Cloud, Samsung SDS, and Elice Group - to participate in the "Advanced GPU Procurement and Operations Support" project, with a planned total investment of 2.08 trillion Korean won (approximately 1.35 billion US dollars) to purchase 9704 advanced GPUs for the construction of the national "AI highway". This is a core project for South Korea in building an independent AI infrastructure model and supporting national AI strategies. Notably, in South Korea's strategic plan, Vera Rubin is explicitly marked as the core target for the next generation GPU procurement. According to the government's specific plan, out of a total of 9704 GPUs, there are 2016 "Verarubin" (Vera Rubin) and 7688 B300. In terms of allocation, Naver Cloud will receive 1008 Vera Rubin and 3112 B300, Samsung SDS will receive 1008 Vera Rubin and 2016 B300, and Elice will receive 2560 B300. The government stated that although the total quantity has been reduced from the initial plan of 15000 (based on B200), the actual performance is equivalent to approximately 19000 B200, which is about 30% higher than the original target. 52,000 "arms race" chips: Softbank, Samsung, Naver all enter South Korea's government's demand for AI computing power has been long-standing. At the end of last year, Bae Kyung-hoon revealed that South Korea plans to purchase 52,000 high-performance GPUs by 2028, and expand this number to 260,000 by 2030 through public-private partnerships. By the end of 2025, the South Korean government announced plans to fully advance AI computing infrastructure construction in 2026, with semiconductors and AI listed as core strategic industries. At the same time, giants like Softbank, Samsung, and Naver are accelerating their layout around South Korea's AI ecosystem. NVIDIA and LG are collaborating to build an artificial intelligence factory covering Siasun Robot & Automation, autonomous driving, and GPU cloud services. NVIDIA also plans to establish a research and development center in South Korea and initiate recruitment. SK Hynix has reached a long-term technology cooperation with NVIDIA to jointly develop Vera Rubin AI supercomputing dedicated memory, entering new markets such as AI infrastructure, personal AI, and physics AI that NVIDIA is pioneering. Samsung's HBM4 and the logic of tying it to South Korea's "priority" The reason why South Korea is urgently prioritizing Vera Rubin on its "priority supply" list is rooted in its representation of NVIDIA's next-generation AI computing core transition. As a new architecture following Blackwell, the Vera Rubin platform was officially unveiled at CES in January, 2026, as NVIDIA's third-generation tens of thousands of card-level rack AI supercomputing platform. Previously, Huang Renxun stated that Vera Rubin is currently in full production and is scheduled to begin delivery in the third quarter of this year. The Ministry of Science, Technology and Information Communications of South Korea expects that Vera Rubin related services will gradually come online starting from the first half of 2027. As South Korea is the world's largest producer of High Bandwidth Memory (HBM), with Samsung and SK Hynix dominating the global HBM market, and Vera Rubin platform highly depending on HBM4, in the context of tight supply and demand for HBM, NVIDIA's supply chain security is deeply linked to South Korea's HBM production capacity. After NVIDIA finalized Samsung, SK Hynix, and Micron as suppliers, the South Korean government immediately sought priority supply in high-level talks, indicating a potential "capacity for priority" equal game between the two parties. For South Korea, the supply right of HBM4 is a bargaining chip to ensure firm priority supply of Vera Rubin - and locking in Vera Rubin supply early will directly affect whether South Korea can keep up in the global AI computing arms race. NVIDIA CEO Huang Renxun's frequent visits and a series of investment commitments in South Korea are adding warmth to these negotiations. However, for South Korea, time and computing power wait for no one - the slight delay in Vera Rubin has already raised the strategic priority for this semiconductor-led AI follower.