Korea bets a trillion on sovereign AI, and Nvidia collects

Domestic control of AI services can coexist with dependence on a foreign chip supplier. That is one argument in SemiAnalysis's report on South Korea: national model projects and locally operated computing infrastructure could broaden NVIDIA's customer base, while Korea still lacks a scaled domestic accelerator supplier.

The supplier's own Korean infrastructure announcement, dated October 30, 2025, describes plans involving more than a quarter-million GPUs across government and industrial projects. Those plans illustrate the distinction: a country can choose where its models run and who operates them while purchasing the underlying hardware abroad. The announced buildout is consistent with the report's demand argument, but it does not test the report's forecasts, most of which sit behind a paywall that was not read.