The video introduces the Japanese tech startup, Preferred Networks, which is noted for its pioneering work in artificial intelligence and deep learning to enhance efficiency across various sectors. Co-founders Tohru Nishikawa and Okano Harasan share their vision of using cutting-edge technology to address real-world challenges and their ongoing projects in fields such as autonomous driving and logistics. They also discuss the significant investment from Toyota and their ambition for a potential IPO to further expand their reach.
Preferred Networks is recognized for the development of Japan's most powerful supercomputer, emphasizing energy-efficient chip architecture. The company aims to differentiate its AI processors from global competitors like Nvidia by creating high-performance, power-efficient chips. The founders elaborate on their strategy to strengthen Japan's position in the global AI race and their groundbreaking work with Plamo, their open-source large language model, specifically designed for Japanese and other languages.
Main takeaways from the video:
Please remember to turn on the CC button to view the subtitles.
Key Vocabularies and Common Phrases:
1. unicorn [ˈjuːnɪkɔːrn] - (noun) - A startup company valued at over $1 billion. - Synonyms: (startup, scale-up, enterprise)
Today, preferred networks is one of the largest unicorns here in Japan.
2. autonomous [ɔːˈtɒnəməs] - (adjective) - Having self-government or functioning independently without human intervention. - Synonyms: (self-governing, independent, self-ruling)
With strategic collaborations across industries such as robotics, health care and autonomous driving...
3. processor [ˈprəʊsɛsər] - (noun) - An electronic circuit that performs operations on some external data source. - Synonyms: (CPU, microprocessor, central processing unit)
Developed in collaboration with Kobe University, the mncore AI processor series.
4. transistor [trænˈzɪstər] - (noun) - A semiconductor device used to amplify or switch electronic signals and electrical power. - Synonyms: (semiconductor, device, switch)
We focus on how much computing power we can pack into a limited number of transistors.
5. efficiency [ɪˈfɪʃənsi] - (noun) - Achieving maximal productivity with minimal wasted effort or expense. - Synonyms: (effectiveness, productivity, competence)
Today, we're shining the spotlight on Preferred Networks, a tech startup that's rapidly using deep learning and artificial intelligence to improve efficiency and accessibility across various sectors
6. optimization [ˈɒptɪmaɪˌzeɪʃən] - (noun) - The process of making something as effective or functional as possible. - Synonyms: (enhancement, improvement, refinement)
In addition, through tuning and optimization, using software has enabled us to create a high performance chip.
7. semiconductor [ˌsemɪkənˈdʌktər] - (noun) - A substance, usually solid, that can conduct electricity under certain conditions better than an insulator but not as well as a conductor. - Synonyms: (circuit, microchip, silicon wafer)
The biggest issue for all semiconductors in the next generation is that they require more power.
8. collaboration [kəˌlæbəˈreɪʃən] - (noun) - The action of working with someone to produce or create something. - Synonyms: (partnership, cooperation, alliance)
We have strategic collaborations across industries such as robotics, healthcare and autonomous driving.
9. architecture [ˈɑːrkɪˌtɛktʃər] - (noun) - The conceptual structure and logical organization of a computer or computer-based system. - Synonyms: (framework, structure, design)
What software and hardware architecture and design went into creating these chips?
10. proprietary [prəˈpraɪətɛri] - (adjective) - Relating to ownership, or relating to an owner or proprietor. - Synonyms: (exclusive, private, patented)
How do your proprietary MN Core AI processors compare to the chipsets made by Nvidia?
How Japan’s largest AI unicorn is shaping the future of deep learning
Hello and welcome to Managing Asia. In this first episode of a two part series, we're diving into the world of innovative Japanese companies that are truly making an impact. Today, we're shining the spotlight on Preferred Networks, a tech startup that's rapidly using deep learning and artificial intelligence to improve efficiency and accessibility across various sectors.
Preferred Networks is one of the largest unicorns in Japan who is leading the charge in artificial intelligence innovation in the country. Founded in 2014, the company specializes in AI, deep learning and machine learning solutions. With strategic collaborations across industries such as robotics, health care and autonomous driving, the company is developing smarter technologies to address real world challenges.
I recently sat down with the founders in Tokyo to discuss their journey and how they plan to boost Japan's global tech presence with their innovations. Both of you founded preferred networks in 2014. Today, preferred networks is one of the largest unicorns here in Japan. Nishikawa san, as CEO, what was your vision for the company? Were you trying to use AI and deep learning to solve pressing social problems and issues?
Here in Japan, we are trying to create a lot of new value in the world by focusing on computers and computer science, which we love. Our vision is to make the world a better place by making the latest technology available in the shortest possible time. And we are working on various real world issues.
It's been 10 years, Okano Harasan as the chief executive researcher, how challenging has it been for you to realize this vision that you've created for yourselves? For the past 10 years, we have been working on using AI to solve real world problems. We have been working on a wide range of issues, including cars, robots and medicine.
And in the real world, each field has its unique problems. For example, there are fields where there are data problems. There are fields where AI technology itself is still underdeveloped. There was also the problem that society had not yet demanded it. We have been solving these problems one by one over the years.
Well, this vision of yours certainly attracted the attention of Toyota. They became a key investor in your company investing 11.5 billion yen in two preferred networks. Nishikawa san, what does it feel like to have one of the biggest automakers in the world as a key investor?
We have received a large investment from Toyota, but I cannot give details about what is currently in progress. When we first received the investment, we had a strong desire to apply AI to the field for automated driving. With this in mind, we received investment from Toyota and accelerated this business.
And one area they were hoping to address with its autonomous driving technology is Japan's trucking industry. The technology would mean that elderly truck drivers remain active in the workforce while reducing the physical and mental strain that comes with long hours on the road.
As part of our current efforts in the field of automated driving, we have invested in a company called T2. This company is kind of a joint venture that originally came around through a collaboration between us and Mitsui and Koi. The field we are currently working on is the automated driving of trucks. The problem in the world of logistics is that you have to drive for many hours and sometimes you have to drive late at night. However, there is a problem in Japan that the number of truck drivers is decreasing.
Like any unicorn, the ultimate ambition is to list one day. Is that your goal and how soon? Investment is also necessary for software development. But compared to software, investment in hardware requires a much larger amount of money. In order to raise such large scale funding, we believe that raising funds from the market will become essential in the future. Therefore, we aim for an IPO as one of the goals of our business.
What kind of time frame? 3, 5 years? Yes, around that. I guess within that time frame. I'm here with Tohru Nishikawa, Co founder and CEO of Preferred Networks, one of Japan's largest tech unicorns. He is giving me an exclusive look inside the company's chip room where all the groundbreaking research and development takes place. One of their most impressive achievements is the creation of Japan's most powerful supercomputer. Developed in collaboration with Kobe University, the mncore AI processor series. Your MN Core series managed to top Green 500's list for the world's most energy efficient computer chips.
In terms of software and architecture, I'm looking at the chips now. What software and hardware architecture and design went into creating these chips? We develop both software and hardware for hardware. We focus on how much computing power we can pack into a limited number of transistors. And we use a very unique architecture. In addition, through tuning and optimization, using software has enabled us to create a high performance chip. We have been able to greatly increase the performance per power.
When you talk about AI technologies, you cannot not talk about the big elephant in the room. Nvidia. How do your proprietary MN Core AI processors compare to the chipsets made by Nvidia? By focusing on deep learning and AI, we are creating more advanced processors and are aiming to gain a competitive advantage by only installing what is necessary for the use of AI. We believe that this approach will help us maintain our competitive advantage.
Okano Harasan when it comes to developing those AI processes, do you have Nvidia in mind? When you want to beat them in terms of speed, in terms of power, one of the things we place importance on is performance and efficiency per unit of power. The biggest issue for all semiconductors in the next generation is that they require more power, but this also generates heat and limits the amount of computing power they can perform.
The processes and related devices we are currently developing are chips that handle data transfer and they require a lot of power. We are working hard to install new innovations into these important areas as quickly as possible to give us a competitive advantage.
Nvidia is the dominant player in the entire generative AI industry. As a tech startup here in Japan, do you think Preferred Networks has what it takes to really beat Nvidia in this field? Regarding our processors, for example, Nvidia dominates the field of AI, but of course we aim to break into this field.
And as Deepseek has shown, even if you don't have the most advanced hardware, you can still create cutting edge models by improving software. I think in the future various players will enter the world of chips and continue to create new products. Okana, Haruh San, I want to talk to you about China's open source AI model, Deepseek, trying to shake up Nvidia.
How do you view this threat? Are you trying to change your own AI strategy as a result? First of all, even before Deepseak became famous, we thought it was a company with very good technology and paid close attention to it. So to some extent we expected it to achieve the position it has now. I think some companies will continue to strengthen their models as they have been doing, while others will challenge themselves to take a new approach and break new ground. I think Deepseek was the first to take the lead in this.
When you look at the AI technologies out there, it seems that US and China are leading the race. Japan is lagging behind as founders of Preferred Networks. Do you want to change that?
Yes. Yes. How so? We have been working on semiconductors and we have also been focusing on several areas in the field of AI. And as a result, we hope that Japanese industry will become more competitive through us. Above all, we want to provide our products and services to the world in the future.
Preferred Networks is leading the way in AI innovation with Plamo, its open source large language model which the company built from scratch. Plamo excels at Japanese specific tasks while also supporting multiple languages, including English. I sat down with Daisuke Okanohara to learn more about Plamo's capabilities and the company's vision for its future.
The Plumo 2 now supports 30 languages as well. But where most current industrial area is edge site. So therefore we focused on say the robotics or automobile or industry site. Is that where the money is?
Yeah, yeah. So in the robotics or automobile, this is a promising field because now the maybe the generative AI technology can change the world. And so therefore maybe it is challenging to get money only from the LLM business, but we maybe find a way to get money from such a robotics or other edge size business.
In terms of leadership and management style, are both of you very different or very similar? I think there are a lot of similarities in our management styles. In terms of management style, it has changed a lot over the last 10 years depending on the state of the company. When the number of employees was quite small, the organizational structure was quite flat. But as the number of employees gradually increased, we created a more structured organization in order to make it easier to gather opinions and easily share them.
In terms of leading and managing the employees, how do you differ from Nishikawa san? I'm more of a person who trusts and delegates to others and if there are talented people, I will select and promote them. In the first place, our company is involved in various projects, so I will choose and promote people who are suitable for each project. I think I often get into the details and manage things myself.
When it comes to important things, whether it's semiconductors or AI, how do you attract the best and the brightest people to work for you? One of our company mottos is Learn or Die, which expresses our stance on learning. Being a specialist is good, but when it comes to hiring, we look at whether the person has a willingness to learn and is flexible.
I want both of you to look back at your 10 year journey. Have you made any mistakes? What are some of the hard lessons learned? There was a period where our company grew rapidly after receiving investment. As the company grew, I was supposed to change my management style.
But I found ourselves in a situation where everybody needed to be facing the same direction, but they were no longer facing the same direction. In order to fix that, I had to do things that involved some painful decisions. I think this was something that I need to reflect on very deeply from the perspective of technological research, given foreign countries like the US and China are already far ahead in the field of AI research.
I am still trying to figure out how Japan can become more competitive in the field. However, even in this situation, I believe that we can still produce world class technologies and research results. What is important for that is to identify what problems have yet to be solved in the world.
Alfani where do both of you see preferred networks in the next 10 years? What will the unicorn you've built look like? We want our processes and applications to be successful in the global market. We aim to have our products installed in various devices globally and for people to use our services through various devices.
In order for Japanese companies like us to be able to compete on the world stage, we need to have bases in other countries and we also need to understand the local culture as we develop our business. So I think it will be a very long road.
In addition to what Nishikawa said, as we introduce AI into various fields in the real world, the problems that can be solved at this moment are only a small part of the whole. There are still many problems that we humans will need to solve in society in the future. As part of these efforts, I hope that we can work on fields that are major problems worldwide and actually make a significant contribution in the future.
These are exciting times for prepared networks. Thank you so much for talking to me. Thank you. Thank you very much.
TECHNOLOGY, INNOVATION, ENTREPRENEURSHIP, AI PROCESSORS, JAPANESE TECH, LEADERSHIP, CNBC INTERNATIONAL