Sovereign AI Heavily Reliant on Nvidia Chips, Posing Challenge for National AI Competition

Despite the fervor for 'sovereign AI' aimed at data sovereignty, an analysis has revealed a high dependency on a specific company for hardware. Market research firm Counterpoint Research, in its 'Sovereign AI Large Language Model (LLM) Research Report,' stated that 92.4% of independent AI systems worldwide utilize Nvidia's AI semiconductors.

By company, AMD accounted for 4.1% and Cerebras for 1.7% after Nvidia. The majority of domestic LLM developers are training their models using Nvidia chips due to the influence of Nvidia's software platform, the CUDA ecosystem.

To reduce reliance on Nvidia, some countries and companies, including South Korea, are attempting to diversify their semiconductor supply chains by exploring alternative solutions. AI startup More is developing LLMs based on AMD semiconductors, and Upstage has adopted AMD's graphics processing units (GPUs). Semiconductor company Rebellions is pursuing an expansion of its market share in inference chips through a partnership with SK Telecom.