TOKYO — Preferred Networks, a Japanese artificial intelligence startup, is looking for new lines of business. Although the company enjoys backing from a number of big names in Japan, its latest move is motivated by its founder’s growing unease over how long it is taking to develop practical uses for its technology.
Preferred Networks was worth an estimated 351.5 billion yen ($3.33 billion) at the end of September last year, making the Tokyo-based company Japan’s most valuable “unicorn,” according to the Next Unicorn survey published by Nikkei. A unicorn is an unlisted company with a valuation of $1 billion or more.
CEO Toru Nishikawa set up a company in 2006 called Preferred Infrastructure while he was in graduate school at the University of Tokyo. The search engine developer spun off its deep-learning business in 2014, which became Preferred Networks.
Few of the startup’s businesses have turned a profit so far, but its advanced deep learning and other AI technologies have raised its corporate value.
Preferred Networks CEO Toru Nishikawa says the company needs to explore new industries to make money. Preferred Networks backers include carmaker Toyota Motor and robot manufacturer Fanuc. They are not pushing for quick profits, which gives the startup ample time to devote to research and development. Still, Nishikawa is putting pressure on himself. He says the company needs to look into other industries to make money.
Like big information technology companies in the U.S. and China, Preferred Networks is involved in numerous sectors, hoping to put its technology to work. In 2019, it expanded R&D into chemicals, forming a tie-up with oil and metals producer Eneos Holdings and signing a deal in cosmetics with Kao. In June, it entered into still another tie-up with game company Gree. It is also eying retailing.
Preferred Networks’ latest foray is in education. It has a programming education business and plans to begin offering classes in August using a curriculum developed together with a cram school operator. The programming classes are a departure for Preferred Networks, which up to now has applied its programming skills rather than its deep learning technology.
“We have a sense of crisis about possibly ending up a mystery company, known only for its estimated high corporate value,” said Nishikawa, explaining the reason for the company’s push into diverse areas. “We shifted gears in the past year or two and started to focus on the business side of things.”
In 2014, Preferred Networks started a project on self-driving technology with Toyota. The automaker has invested about 11.5 billion yen in the company. The project aims to develop technology that will allow a car to understand its environment using video. But, said Nishikawa, “I doubt if we can realize a perfect self-driving system in five years.”
It remains unclear whether any of Preferred Networks’ other tie-ups with large companies will bear fruit. “We need to slow our pace to work with large companies in tie-ups,” Nishikawa said, “which can be beneficial for medium- to long-term research and development purposes, but we’ve not been able to accelerate the pace of development sufficiently to deliver cutting-edge technologies.”
At the end of last year, Preferred Networks terminated its work on Chainer, a deep-learning framework it announced in 2015. Chainer can correct calculation errors in data learning. Its technology came out ahead of that developed by big-name rivals such as Facebook or Google, but it lagged in getting others to adopt it.
Preferred Networks also developed the MN-3 supercomputer, which in June topped the energy efficiency rankings in the Green500, a global rating of supercomputer performance. The MN-3 crunches the most numbers with the least amount of power, using AI to process huge amounts of data.
Preferred Networks’ MN-3 supercomputer was recently rated the world’s most energy efficient putting it at the top of the Green500 ranking for such devices. “If you want to develop great software, you also need high-performance hardware to run it.” Nishikawa said.
Preferred Networks, meanwhile, has also started to bolster the business side of its operation. Last year, the company hired Tomonobu Tominaga, who was previously head of marketing at supermarket operator Seiyu. The goal is to transform the company from one led by technologists into one that can make a profit using advanced technology.
Of course, high tech does not stay that way for long. “The AI boom will soon run its course. We must find new technologies that can help us leap forward from the current AI,” Nishikawa said.
Preferred Networks is now looking to apply AI in different areas to determine where it can make money. “Technologies using AI have not totally matured yet. You may try to carefully select areas in which to focus, but you never know if you will succeed. You can’t stop searching for viable new business areas,” Nishikawa said.
Asked about his exit strategy, Nishikawa said: “I don’t want to do an IPO unless we need to do it to raise funds. But if there is something we want to do so much that we feel it’s worthwhile doing, then we’d do [a share flotation]. What I want is to have our technologies used around the world, like [Microsoft] Windows.”