South Korea Secures 21,000 of 35,000 GPUs to Be Delivered Directly by Nvidia
Key Facts
South Korea said it has secured about 21,000 graphics processing units out of the roughly 35,000 it plans to obtain directly from Nvidia, with the remainder expected to arrive in early 2027. The minister of science and ICT said on Nov. 13 that the government’s initiative to secure 50,000 units was on track. The remarks followed a meeting the ministry held with the company earlier in the day to review outstanding issues. The ministry linked the meeting to concerns raised about the impact of CEO Jensen Huang’s resignation on dealings with South Korea. The statement marks tangible progress on part of the supply plan, while the quantity delivered to the government remains below the total it will obtain directly from the company.
The figures in the statement span a government initiative targeting 50,000 units and direct purchases from Nvidia totaling about 35,000; the minister said about 21,000 of the latter had been secured as of Nov. 13. That means units from the direct-purchase plan are still pending, with early next year set as the expected arrival time. The statement did not specify a separate date for each batch or provide details of the phased delivery schedule. It also did not explain how the direct-purchase amount is counted within the broader government initiative. The minister said the initiative was "on track," while the arrival date for the remainder remained tied to the stated expectation of early 2027.
The supply is tied to South Korea’s effort to combine the development of AI computing infrastructure with domestic semiconductor development. The minister said the cooperation would include using Nvidia’s NVLink technology to integrate domestically developed Korean chips with the company’s platforms. This puts cooperation between the two parties on two parallel tracks: providing computing units and working to connect local semiconductors to Nvidia’s platforms. At the APEC summit, the company had previously announced plans to work with South Korea to expand infrastructure by more than a quarter of a million units through sovereign clouds and AI factories. That earlier plan shows the broader scope of the announced project compared with the direct purchase the minister discussed in his latest statement.
The announced industrial plans include using the units in projects by Korean companies, alongside government needs. Nvidia said Hyundai Motor Group plans to work on autonomous mobility, smart factories and robotics technologies using 50,000 Blackwell units. It also announced that Naver Cloud plans to deploy more than 60,000 units for enterprise AI and physical AI work. In a separate announcement dated Oct. 31, 2025, the company said Samsung Electronics was building an AI factory with more than 50,000 units to accelerate its AI, semiconductor and digital transformation plans. These figures show that the broader infrastructure announcement includes multiple plans by the government and companies, each with its stated uses.
Nvidia announced on Oct. 31, 2025, that SK Group planned to build an AI factory with more than 50,000 units, alongside an industrial cloud aimed at robotics and physical AI applications. The company said the factory would use Nvidia units and the cloud would be based on server-dedicated Blackwell units. In the same announcement, Hyundai Motor linked its plan to an investment of about $3 billion to develop the physical AI sector in South Korea. The initiatives included an AI technology center, a physical AI applications center and regional data centers. These projects provide context for the broader industrial partnership under which the government supply is taking place, without the minister’s statement specifying the status or implementation date of each project.
Nvidia’s expansion plan announced in October 2025 included deploying 13,000 units in an initial phase through Naver Cloud, NHN Cloud and Kakao, with deployment set to grow over the following years. The announcement linked this phase to support for initiatives including the National AI Computing Center. In a parallel effort, the company said Naver Cloud, LG AI Research, SK Telecom, NC AI and Upstage were working with it to develop Korean foundational language models to support local applications. Huang described South Korea’s ambition as turning its industrial capabilities into a source of intelligence exports, saying, "The country can now produce intelligence as a new export that drives global transformation." The Nov. 13 statement says the government expects the remainder of its direct purchases to arrive in early 2027, while the broader expansion plans remain spread across the following years, according to the company’s earlier announcement.