Preferred Networks seeks IPO to mass-produce AI chips that could outpace Nvidia’s GPUs

A Japanese AI company most people outside Tokyo have never heard of is plotting an IPO with a bold pitch: its custom-built chips can run generative AI workloads up to ten times faster than conventional GPUs. Preferred Networks, founded in 2014, wants to go public specifically to fund mass production of its MN-Core processor line, a move that could inject fresh competition into a semiconductor market where Nvidia has enjoyed something close to a monopoly on AI training and inference hardware. The company currently carries a valuation of around $2B, employs roughly 450 people, and counts Toyota, SBI Group, Fanuc, and NTT among its strategic investors. Its leadership has indicated that an IPO window sits approximately 3 to 5 years out from March 2025, which puts the listing somewhere in the 2028-2030 range, contingent on when its chips actually roll off production lines at scale. The chip and its architecture Preferred Networks’ MN-Core L Series processors take a fundamentally different approach to AI computation than the GPUs that dominate data centers today. The chips use 3D-stacked architecture to dramatically boost memory bandwidth, which is the real bottleneck for generative AI i...