South Korean neural processing units (NPUs) are emerging as a cost-effective alternative to Nvidia’s graphics processing units (GPUs), presenting a challenge to the company’s dominance in AI inference. Nvidia recently secured a contract to supply 260,000 GPUs to South Korea for sovereign AI training. In response, local firms like Rebellions, backed by Samsung, are now mass-producing NPUs specifically designed for AI inference tasks. The South Korean government plans to invest $634.5 million by 2030 to support the development of local AI semiconductor technologies, aiming to lessen the nation’s dependence on expensive foreign GPUs. IT service companies are beginning to adopt these domestic NPUs to reduce costs and enhance their competitiveness in AI transformation projects. This dual approach—utilizing Nvidia GPUs for training while fostering local NPUs for inference—indicates a potential shift in the semiconductor landscape. As Nvidia (NVDA) encounters increasing competition in the inference market, it will be important to monitor whether Korean NPUs can expand beyond niche applications and impact Nvidia’s global market share.
This Shorty is an educational market impulse and does not constitute trading or investment advice. Always do your own research.