
The term "domestically developed AI model" reveals a slightly different picture when looking at the lower layers. Swallow from Tokyo University of Science, Sarashina from SB Intuitions, and tsuzumi 2 from NTT Data are all built on NVIDIA’s Nemotron datasets and libraries. The national AI infrastructure also relies on NVIDIA’s configuration: 13,750 Vera CPUs and 27,500 Rubin GPUs. This raises the question: at which layer are we defining something as “domestically developed”?