Preferred Networks

Preferred Networks leads Japan's deep-learning revolution, training the world's fastest AI supercomputer, MN-3, at 23.2 petaflops for robotics and drug discovery. Its open-source framework, Chainer, outperforms #10 Fujitsu's legacy AI tools by 40% in training speed for autonomous driving models. In 2025, it secured ¥15.6 billion in funding, 30% more than the average Japanese AI startup, and its drug-discovery partnership with Takeda reduced candidate screening time from 12 months to 7 months.