Toru Nishikawa (right), co-founder and chief executive officer of Preferred Networks, and Daisuke Okanohara, co-founder and chief executive researcher, with a Fanuc Corp. robotic arm at their company’s headquarters in Tokyo, Japan, March 16, 2018.
Bloomberg | Bloomberg | Getty Images
Preferred Networks is a little-known name outside of Japan, but the Toyota-backed artificial intelligence unicorn has big plans to use deep learning to fix “real-world” issues — with the end goal of going global.
“For the past 10 years, we have been working on using AI to solve real-world problems,” chief executive researcher, Daisuke Okanohara, told ‘s “Managing Asia.” The company has ventured into industries as varied as trucking, health care and robots.
Preferred Networks has drawn the attention of big names such as automotive giant Toyota, which in 2017 invested 10.5 billion yen ($95.4 million) in the company on top of an initial 1 billion yen in 2015.
Preferred Networks is one of the few unicorns in Japan. Though Japan’s startup ecosystem has grown, Japanese companies valued at over $1 billion make up just 0.5% of the world’s total, according to an International Monetary Fund working paper published in December.
The deep-learning startup is also in a joint venture with trading house Mitsui & Co. in the field of autonomous driving. The two have invested in a company, T2, in order to address challenges associated with the trucking industry, including “extremely long” working hours that stretch into the night and a declining pool of drivers, said CEO Toru Nishikawa.
But competition in the car industry was also a factor. Acknowledging the company’s current limitations, Okanohara added, “If we were to tackle the problem of regular cars, it would be extremely difficult. There are many companies working on autonomous driving, and it is a very competitive field.”
Preferred Networks’ AI ambitions have since gone much further afield. But compared with other AI solutions in the digital space, in which it is “relatively easy to achieve results with users,” dealing with real-world problems requires more time, Okanohara said.
When asked how long it takes for the company’s technologies to start making money, Nishikawa said it could take as many as five years in a field like AI-based materials discovery.
“First, we start joint research with companies and then assess whether or not it can be commercialized. Once we determine that it can be commercialized, we work to create services and products that can be provided to customers,” he said, adding that “in total, this process takes about three to five years before commercialization and practical launch.”
As for the latest AI developments, Okanoharo said that Preferred Networks had recognized DeepSeek’s “very good technology” and “paid close attention to it” before it was catapulted to fame, adding that “various players” will soon enter the chip industry.
Preferred Networks is hoping to be among them. Recognizing that Nvidia’s generative AI strategy has fared well thus far but has a graphics processing unit design that is not yet “optimal,” Nishikawa said Preferred Networks is in the midst of “creating more advanced processors” to gain a competitive advantage.
IPO plans
Whether the company will be able to achieve the scale of its ambitions remains to be seen. For now, it’s eyeing foreign investors, and is receiving “a lot inquiries in the fields of entertainment, semiconductors and computers,” according to Nishikawa, who noted that “there are limits to what we can do in Japan alone, and the global market is much bigger than the Japanese market.”
Getting an initial public offering is one of the company’s goals, he added. When asked for a time frame, Okanohara said, “I think that the need for huge funds will come at the time when we introduce our hardware products, such as semiconductors, to the world. So I think we will aim to go public at that timing,” adding that could take about three to five years.
