NVIDIA CEO Jensen Huang visited Korea in late October 2025 to deliver a keynote speech in which he emphasized his vision of the AI era and the importance of cooperation within the Asia-Pacific region. Huang also announced a massive supply agreement, pledging to supply 260,000 GPUs to South Korea. The GPUs will be distributed among the government and major private companies like Naver, Samsung Electronics, SK Group, Hyundai Motor Group. Jensen Huang stated that Korea would receive priority access to GPU allocations. This is critical because companies worldwide usually have to wait months or years for these chips.



What is a GPU?



A GPU (Graphics Processing Unit) is the core part of a graphics card. It processes large-scale graphic and numerical computations at high speed. Unlike CPUs that process small numbers of complex computations, GPUs can process a massive number of simple computations simultaneously. The process of AI deep learning is similar to a large number of simultaneous simple computations; therefore, GPUs have become the core component of the AI industry.



Why is NVIDIA so strong?



Among the companies that produce GPUs, NVIDIA achieved a 92% share of the Add In Board GPU market in the first quarter of 2025. Moreover, in 2006, NVIDIA created a software programming tool called “CUDA” that enables AI development using NVIDIA GPUs. Before CUDA, there was no platform that enabled the use of GPUs for general purpose. This made CUDA the first widely-adopted platform that enabled GPU use in fields outside of graphics. CUDA provided optimized performance and abundant tools for developers and thus many have used the program for developing and studying AI. If developers were to use GPUs from other companies, they would need to study the new platform, rewrite the code, and adjust for optimization. As developers are reluctant to make these enormous efforts, NVIDIA remains an almost complete monopoly in the GPU market.



What Other Companies Are Doing To challenge



NVIDIA’s monopoly, graphics card competitor AMD has developed ROCm (Radeon Open Compute), AMD’s opensource alternative to CUDA. Although ROCm has historically lagged behind, recent updates (ROCm 6.x and 7) have significantly improved stability.



AMD has also partnered with OpenAI to target both hardware supply and software compatibility. The hardware partnership centers around OpenAI’s commitment to deploy AMD’s nexgeneration GPUs. The development of OpenAI’s programming language called Triton acts as a middle layer. It allows developers to write code that runs efficiently on both NVIDIA and AMD chips. Through the partnership, OpenAI and AMD are working together to make Triton run on AMD hardware. The deal also includes a special option for OpenAI to buy stock in AMD. This helps both companies work toward the same goal: relying less on NVIDIA.



Other big tech companies like Microsoft, Google, and Amazon have also built their own specialized chips. Microsoft has deployed “Maia” to accelerate AI tasks for OpenAI. Google powers its Gemini models using its Tensor Processing Units (TPUs). Amazon targets cloud customers with two distinct chips: “Trainium” for training AI models and “Inferentia” for running them.



Impact on Korea



Before Jensen Huang’s announcement of the NVIDIA supply agreement, South Korea had an estimated inventor y of only about 60,000 enterprise-grade GPUs. The NVIDIA deal effectively quadruples the nation’s AI compute capacity.



In Huang’s keynote, he said “Korea has the data, the language, and the culture. You must own your own intelligence. If you rely entirely on others’ models, you lose the ability to encode your own nation’s common sense.” He defined sovereign AI as a matter of national security and cultural identity. The GPUs allocated to Naver and the Korean government are expected to help achieve these goals.



Reporters



Jinpyo Hong

undohere@seoultech.ac.kr



Seongbeom Hong

hongsb@seoultech.ac.kr