Trailing Behind: The 2020–2021 Japanese AI Landscape

Part I of a Two-Part Series

2020–2021 Landscape of Japanese AI Companies

The 2010s established and normalized “Big Data” as a household term across the world due to the acceleration of datafication and the growth of the FAANG (Facebook, Apple, Amazon, Netflix, Google) companies and related startups. In 2021 and beyond, Big Data and the more recent resurgence of AI show promise for unprecedented economic growth through automation and efficiency, but the growth will hardly be distributed evenly. Technology industries are well known for their winner-take-all nature and walled gardens, and the question for the Japanese data and AI industry is “how will the domestic industry catch up with the West and China while simultaneously overcoming its myriad number of challenging social issues including a super aging society, a persistent homogenous mindset, workforce shortages, and the novel coronavirus?” In order to address this question, this report first takes a historic look at Japan’s AI development and then digs into the data of the first exhaustive market landscape for AI in Japan, which was built over the past four years.

This work would not have been possible without the significant contributions of Sean Furuta and Kentaro Asaba.

Disclaimer 1: All data was collected through public sources. I believe the presented landscape is reasonably exhaustive, but as new companies rise and existing ones fall or are acquired, it will always be a work in progress. If your company was missed or mischaracterized or if you have any comments or feedback, please reach out via the Contact page on my website https://simeonbochev.com/contact by selecting “AI Landscape Commentary & Usage” as the “Type of Inquiry”.

Disclaimer 2: As I receive a lot of similar request requests: feel free to use the charts and data from this market landscape in books, conferences, presentations, etc. I only ask that when using these materials, you follow these three requirements:

1) Do not alter/edit the material in any way without my explicit consent

2) Provide clear attribution to Simeon Bochev, Sean Furuta, and Kentaro Asaba, AND

3) Notify me via the Contact page on my website https://simeonbochev.com/contact by selecting “AI Landscape Commentary & Usage” as the “Type of Inquiry”

This report is a summary from a proprietary database of over 400 active AI companies in Japan that for each company includes their service & product offerings, funding level, team size, related PR, known partnerships & clients, as well as much more information. If you are interested in access to this database, you can reach out to me through my website following the instructions above.

Disclaimer 3: As illustrated by this Medium story, there is a non-trivial subset of companies that claim to be “AI companies,” without having integrated true AI into their core business or at all. My data gathering aimed to filter out such companies; however, since I relied solely on public self-descriptions of companies and their services it is possible to have missed some such companies.

Without further ado, let’s begin!

Starting from Behind

Despite being a world leader in high-tech consumer goods and hard sciences and the world’s third largest economy, Japan has lagged other leading world nations in the development of data science and AI. One telling statistic comes from the output of AI related research papers from 2019 with 57% coming from the US while Japan only contributed 2%, placing Japan eighth in global rankings. One might wonder how Japan fell so far behind. Perhaps it was because Japan did not invest in the technology early enough? The answer might be the exact opposite: Japan was too early and became discouraged.

In 1982, the Japanese Ministry of International Trade and Industry (MITI)[now named the Ministry of Economy, Trade and Industry (METI)] established the Fifth Generation Computer Systems (FGCS) to research and develop highly parallelized computers and AI. MITI invested $448 million towards the project and committed to 10 years of support. However, in 1992, the project was declared a failure and not renewed, discouraging further investment in the field. Although the United States had similar early AI initiatives, the United States diversified and invested in other emerging technologies, such as high-speed processors like Intel’s X86, which rode the Moore’s Law revolution of the late 20th century that helped establish Western dominance in computing through the present day.

One of the keys to the current global AI boom was in 2004 when Google published the seminal MapReduce paper, which was critical in enabling massive parallelized computers to process large datasets efficiently and cheaply. Google’s MapReduce gave rise to integral Big Data technologies such as Hadoop and Spark, and powered Google’s ranking system and other services for nearly a decade until they transitioned to cloud computing. While the Japanese government was on the correct path, its failure to produce meaningful results shifted focus towards other hardware technologies, like robotics and automobiles. This advancement in hardware has given Japan an advantage in sensors and data collection, lowering one of the barriers to AI adoption.

Today, Japanese companies are faced with large datasets that are either completely unusable due to a lack of digitization of data or not fully utilized because of shortages in research and human capital. According to Canadian startup Element AI, Japan only has approximately 3.6% of the top AI talent worldwide, while the United States has more than 50% followed by China and the United Kingdom. Unfortunately, the domestic outlook for growth in future talent within the AI field does not look bright with METI estimating a rapidly growing shortage of 120,000 AI experts by 2030. Further, according to a 2015 survey conducted by Nikkei Asia, this shortage has forced a majority of companies to recruit talent from other STEM-related fields and provide internal training to build data and AI skills. What remains to be seen is if Japan, with its shortage of top AI talent, can successfully upskill and train its shrinking domestic workforce or if Japan will have to rely on attracting experts from abroad.

Despite these shortcomings, data science and AI startups are slowly popping up across Japan with the number of new startups more than tripling in the past decade from 84 to 291 (as of the end of 2020), based on research of publicly collected data sources. Refer to Figure 1 below for a summary graph of newly created AI startups in Japan between 2000–2020. The drop off in 2019 to 2020 can be attributed to my conservatism in keeping founding dates blank when they were unclear, coupled with the way data was gathered; namely, by collecting information from various news outlets and company websites directly, it is reasonable to assume many very young startups that lack an online or PR presence were missed. Notwithstanding the tripling of AI startups in Japan over the past decade, Japan still dramatically lags behind other developed nations in the creation of new AI companies and technologies.

Figure 1. Yearly Count of Newly Created AI Startups in Japan Between 2000–2020

Research: More Than 2%

The Japanese Government has shown a willingness to invest in the AI industry through its Japan Revitalization Strategy. This 2015 proposal has several initiatives that are likely to accelerate the digital transformation of Japan, such as increased local access network (LAN) connectivity (to eliminate the number of data “dead spots” in Japan in order to enable more IoT devices); government councils for robotics, cyber security, and AI development; and a government data platform for publishing government data for use by the public. Although these are welcome steps towards creating a flourishing data ecosystem, such infrastructure has already been a priority in the United States and China, the world leaders in AI.

The investments in the AI field go beyond government plans as several public and private institutions are recognizing the need to establish research facilities and startup programs. Universities across Japan have established several support mechanisms for their faculty and for bringing their ideas into action. Below are some representative examples of recently established AI research initiatives and university-company partnerships:

• Hokkaido University created a startup venture program that has led to the creation of several burgeoning startups such as AWL, TIL, and Chowa Giken. These three companies are currently collaborating on a COVID-19 solution to prevent customer harassment, detect congestion, and help customers distance themselves within stores.

• In 2016, the University of Tokyo established the Next Generation Artificial Intelligence Research Center (AI Center) which brings together professors from a wide array of backgrounds to collaborate on AI. SoftBank has committed ¥20 billion (~$200 million) over 10 years towards facilities, equipment, and labor costs at the AI Center. The AI Center has also recently announced a partnership with Toyota Central R&D Labs called “Intelligent Mobility Society Design, Social Cooperation Program.” Well-established companies like NVIDIA and Dwango are also making donations and scholarships at the center.

• Tohoku University is partnering with Hitachi on Metro-Area Distributed Storage System which will make data widely available and protected during disaster scenarios.

• In 2016, CyberAgent established an AI Lab which has partnered with researchers from the University of Tokyo, Meiji University and Yale University to improve ad technologies. Also, the AI Lab has led to tangible results with the establishment of Hanjuku-Kaso (HK), an R&D startup focused on applying economic theory, specifically counterfactual/causal ML, for social decision-making and targeting deployment within traditional industries and the public policy domain. Hanjuku-Kaso is led by Yusuke Narita, a Yale economics professor who has partnered with the AI Lab since 2016.

Other public institutions are also developing AI labs across Japan, such as RIKEN (Institute of Physical and Chemical Research). In 2016, RIKEN established its Center for Advanced Intelligence Project (AIP) under the guidance of the Japanese Ministry of Education, Culture, Sports, Science and Technology (MEXT). AIP aims to be a collaboration spot between the government and 20 various AI-related companies. Also, in May 2015, the National Institute of Advanced Industrial Science and Technology (AIST) established its Artificial Intelligence Research Center (AIRC). The Center’s goal is to accelerate the research surrounding AI applications such as medical services, robotics, and self-driving cars.

These public-private partnerships are essential not only for increased financial investment in the AI field, but also investments in human capital. As more money flows through university programs, young engineers will find viable career paths, alleviating part of the employee shortages and creating a more competitive supply side to the labor market. Although these partnerships represent steady progress, they are not nearly enough to fully close the rapidly widening talent gap, which will require a combination of education reform, further increases in funding, and welcoming foreign talent to solve.

Japan is clearly making strides towards creating a core research infrastructure with lofty public goals, which are the critical first steps to advancing in the globally competitive field of AI. However, these goals should be scrutinized for their feasibility as the talent, innovation ecosystem, and funding required to reach these ambitions are simply not there today. Based on where Japan is at the start of 2021, it is hard to believe that Japan will be close in the race to certain technologies that require AI, such as self-driving cars, compared to the United States and China. As a result, Japan must strengthen its research initiatives by not only funding various ventures, but also investing in talent expansion. Developing partnerships with foreign researchers and incentivizing them to contribute to the AI field in Japan will be a key short-term lever in breaking beyond the 2% research barrier.

Japan’s AI Landscape: Expanding Horizons

The AI ecosystem is constantly evolving globally. This is true especially in Japan, which as a result of its own nascent ecosystem, presents a large gap to be bridged. Many larger, traditional Japanese companies as well as established technology companies are making heavy investments in the budding local AI ecosystem. Figure 2 illustrates the scope of the 2020–2021 Japanese AI landscape. Companies have been grouped into the following categories (and sub-categories) based on their primary capabilities:

Figure 2. Japanese AI Landscape 2020–2021

Enterprise Applications:

• Business Administration
• Cloud & Data Analyst Platforms
• Human Capital
• Legal
• Marketing
• Security & Crime Prevention

Data Analysis:

• General Applications
• Geographic
• Image Recognition
• Internet of Things (IoT)
• Natural Language Processing (NLP), Chatbots & Communications

Industry Applications:

• Consulting & Digital Transformation
• Construction
• E-Commerce (EC) & Marketplaces
• Education & Training
• Energy
• Entertainment, Media & News
• Fashion
• Finance & Investment
• Food
• Manufacturing
• Medical & Healthcare
• Personal & Corporate Matching
• Real Estate
• Robotics
• Telecommunications
• Other

It is reasonable to expect that there is cross-pollination across groupings. As noted before, this is a continual work in progress and feedback is welcome.

Big Fish in a Small Pond

The completed research has found 438 different companies active in the AI field in Japan, ranging from 291 that are early to mid-stage startups (e.g., Hanjuku-Kaso, Sakeai, etc.), 127 that are late-stage startups (e.g., ABEJA, Cogent Labs, etc.), and 20 that are large and established companies (e.g., CyberAgent, NTT Group, Softbank Group, etc.). From the companies for which a public valuation exists, the total Japanese AI market capitalization came out to roughly ¥500 billion (~$5 billion). One caveat: all of the 20 large companies and a significant number of the late-stage startups are not pure AI companies, meaning their contribution to this AI market capitalization figure is a subjective vs. objective estimate. Additionally, the number of companies with a market cap over ¥1 billion (~$10 million) is approximately 50. Below are outlined some of noteworthy companies of various sizes from the list of 438:

Hanjuku-Kaso: Series A Startup with VC Backing

Hanjuku-Kaso is a data science R&D startup focused on applying economic theory and causal inference to various fields in both the private and public domains. Data science differs from AI by utilizing data to draw specific insights through statistical visualization, prediction, and experimentation. HK emerged from CyberAgent’s AI Lab and remains a close partner. Within digital advertising, HK partners with multiple private companies including CyberAgent to explore dynamic pricing for ads, hyperpersonalization, and recommendation systems, among others. The advent of Big Data has enabled small, agile data science startups like HK to emerge and partner with existing large companies. Due to the nature of data science, a small team of capable data scientist paired with sizable datasets can provide incredible value to large companies.

CyberAgent: Publicly Traded Digital Marketing Firm

CyberAgent is a digital advertising, media, internet, and game company. In the field of AI, they have set up their AI Lab in order to develop AI solutions for AdTech and other tangential purposes. They also have a project to provide chatbot-related services. Through their partnership with academic spaces and their venture capital arm, CyberAgent Capital, CyberAgent has become a major force in the AI space in Japan. Most of CyberAgent’s investments have been towards the ad space. For example, CyberAgent Capital invested in Lockon ($2.1 million Series A), a digital marketing AI firm, currently working on its AD EBiS platform. CyberAgent also has numerous subsidiaries working on AI technologies such as AJA, a small startup developing state-of-the-art advertising products to help improve the profitability of blue-chip media. However, CyberAgent’s investments are not limited to digital marketing as seen by their investment in Kabuku ($9.1 million), a manufacturing AI firm focusing on on-demand manufacturing services known as MMS Connect.

NTT Group: Publicly Traded Telecommunications Conglomerate

The Nippon Telegraph and Telephone Corporation (NTT) is a Japanese telecommunications company. The NTT Group consists of five main group companies: NTT East, NTT West, NTT Communications, NTT DoCoMo and NTT Data, many of which have project relating to the data and AI industry. NTT also has several subsidiaries that work on AI products and services, utilizing research from NTT Research. NTT has also invested in more than 20 AI-related startups covering a wide range of fields. For example, Shelve SCAN-AI is a service that can automatically recognize products from store shelving photos using AI. Until now, store employees had read product data with scanners and manually converted it into dedicated software when optimizing shelve layouts. Shelve SCAN-AI automates these tasks with deep learning. Another NTT product which reduces labor hours is Okudake Reception. Okudake Reception enables bi-directional communication and supports voice search by utilizing AI. You can search for the person in charge and directly call them. It can be used simply by installing a tablet terminal and peripheral devices and no wiring is required.

SoftBank: Publicly Traded Telecommunications Conglomerate

SoftBank Group Corp. is a Japanese multinational conglomerate holding company headquartered in Tokyo. SoftBank owns stakes in many technology, energy, and finance companies. It also runs its Vision Funds (I & II), the world’s largest technology-focused venture capital funds, with over $100 billion in capital in Fund I alone. SoftBank has many subsidiaries working on various AI solutions such as agoop ($3.5 million market valuation), which is a location information company that provides businesses with detailed and comprehensive information about users. SoftBank has partnered with IBM to introduce IBM Watson into the data and AI industry in Japan. One joint collaboration project includes a partnership with Mizuho Bank in which a combination of IBM’s Watson and SoftBank’s Pepper humanoid robot provide customer phone support. A similar partnership has been started with JR East to deploy guidance robots in stations. This partnership also extends beyond banking and mobility to include education, healthcare, and insurance. SoftBank’s partnership with IBM is a strong model for how to blend externally sourced Western AI technologies with domestic Japanese technologies to push the bounds of customer interaction.

Preferred Network (PFN): Series D AI Startup

Preferred Networks was founded in March 2014 as a spinoff from Preferred Infrastructure. PFN researches and develops innovative AI solutions in three domains: transportation, manufacturing, and bio/healthcare. They have partnered with Toyota to develop autonomous driving and connected cars. They have also partnered with Fanuc and Hitachi to apply ML and deep learning technologies to robotics and machine tools. PFN has also collaborated with the National Cancer Center of Japan to develop a system that analyzes CT and MRI medical images to allow early diagnosis of cancer using blood samples. PFN’s partnership with Aillis ($28.5 million Series A) led to the development of AI-enabled medical devices. Currently, they are focusing on their own AI hardware chip design called MN Core. These chips will be used for in-house computation for deep learning. Their hardware push follows a global trend of AI firms, such as Huawei and Google, in developing their own silicon.

ABEJA: Series D AI Startup

ABEJA enables retail and manufacturing optimization through its ABEJA platform which uses footage from store cameras to estimate the age and gender of customers (among other factors) to create customer profiles for more personalized customer service offerings. By combining the profile with other data sources like point-of-sale and weather, ABEJA insights can predict whether the customer will return. ABEJA’s success has also drawn the eyes of Western companies like Google which invested in ABEJA in 2018. ABEJA has also been able to track the impact of the coronavirus on store foot traffic and found traffic decreased by 44.9% in March 2020 when compared to March 2019.

PKSHA Technology: Publicly Trade AI Software Firm

PKSHA is an AI startup focusing on NLP and image recognition that has developed an algorithm licensing business for these technologies. PKSHA sells algorithm modules and software for smartphones in the automatic response and image recognition area. BEDORE is one example of PKSHA’s software, which is an automated text and voice response system that can reduce the need for voice telephone conversations with customers by approximately 30%. BEDORE has seen the highest sales among all the algorithms PKSHA sells. Also, like many other successful Japanese AI startups, PKSHA has also partnered with larger companies in the space like Sparks Group to create the PAKSHA Spark Algorithm Fund which invests in AI and robot-related startups.

Looking Forward

What has hopefully been made clear from Part I of this report is the wide array of companies in the AI field that are pushing the capabilities of the AI industry in Japan, but that despite small bright spots, the AI ecosystem in Japan is still in its infant stages and dwarfed by the West (primarily the United States) and China. In order to compete on a global scale, Japan’s AI industry must expand rapidly to increase innovation and collaboration. Such expansion starts with dedicating the necessary resources towards vital areas in the industry and diversifying talent sources, among other needed changes. It is my hope that this research and the associated landscape report can serve as the foundation for a more structured, strategic and calculated path forward for the Japanese AI industry.

Stay tuned for Part II, which will delve deeper into the data behind the Japanese AI landscape including the current models of collaboration, the trends that seem to be taking hold in Japan, and the opportunities that can propel Japan to successfully compete on the global AI stage.

Thank you for reading! I look forward to your comments and to continuing the dialogue that this research will hopefully start. Here’s to a fantastic 2021 for all!