Preferred Networks, a Tokyo-based Japanese AI company, is planning a stock market listing within three to five years to fund mass production of its MN-Core processors, claiming they deliver inference up to ten times faster than traditional GPUs. The company is valued at approximately two billion dollars with around four hundred fifty employees, and counts Toyota, SBI Group, Fanuc, and NTT among its strategic investors. The MN-Core L Series chips use 3D-stacked architecture to maximize memory bandwidth, the main bottleneck for generative AI inference. Preferred Networks partners with TSMC for twelve-nanometer and seven-nanometer processes, and with Samsung for an ambitious push toward two-nanometer chip production.
Source: Read the original article
