co-authored by Jerry Chi and Ilya Kulyatin

Japan is playing a unique, crucial role in the global AI ecosystem. In this post, we go beyond hand-wavy claims like “jApAn is sTiLL iNnoVatiVe” and concretely summarize why Tokyo and Japan are hugely relevant for AI globally (and show that we’re not just Japan AI shills 😜 ).

Macro backdrop: Japan’s unique positioning

Hold up. Let’s first cover some macro fundamentals.

The US and China are the top 2 superpowers, both in terms of GDP and AI capabilities. Japan, with the 3rd-largest GDP, is closely intertwined with them, being the 2nd-largest trading partner of both superpowers outside of North America. Increasing geopolitical tension between the US and China makes Japan’s role increasingly important in APAC, the region accounting for about 40% of global GDP and 60% of global GDP growth.

Many AI-related companies see Japan as the 2nd most important market after the U.S., especially since China is too tricky to work with in terms of geopolitics and legal issues. Japan is already the 2nd biggest country for cloud services. Just AWS by itself accounts for tens of billions (USD) in investment in Japan, underscoring Japan’s market size of AI and cloud-related services.

Japan is estimated to be 2nd highest country in terms of the potential boost to productivity from AI adoption, with its AI-related market potentially reaching $736 billion by 2030. A good amount will be unlocked through solutions, integrating AI into SMEs, enterprise, and government. Furthermore, Japan plays important roles in key growing industries relevant for AI, including media/entertainment and other creative industries, robotics, healthcare, new materials, biotech, and manufacturing (needed for e.g. making NVIDIA chips). In a16z’s recent article Anime is Eating the World, the global market for Japanese-style anime (including games) is estimated at >$50 billion. Also, Japan has great infrastructure conducive to building and using AI; for example, Japan ranks in the top 3 OECD countries for fiber optic network infrastructure.

Different country, different strengths

Although the US has much of the top AI talent, that by itself is often not enough to have direct real-world impact in certain industries, since domain knowledge and access to certain data are often crucial. Collaboration between the US and Japan’s strong players in certain industries can help. For example, Bearing AI, a Silicon Valley startup backed by Andrew Ng’s AI Fund, is partnering closely with Japan’s Mitsui OSK Lines, one of the world’s largest shipping companies with deep maritime expertise, for joint development and commercialization of AI solutions for the maritime industry, including a Smart Routing Engine that helps ships optimize routes. Such development probably couldn’t have been done if limited to the US, which lacks such a shipping company.

Japan also differs from countries like the US in that the employee turnover rate is much lower, often due to culture or loyalty. While in some cases this might actually hurt enterprise AI adoption in the short term due to fear of “stealing” human jobs, in the long term it may pay off for many companies that figure out how to train their workforce to use AI and augment their work and creativity effectively, treating AI as copilots (e.g. refer to ad agency Hakuhodo’s Human-Centered AI Institute); this is in contrast to some company executives in the US etc. who are eager to lay off workers, replace them with AI, and boost quarterly earnings, which might work well financially in the short term but may not always be the best approach for the company or for the well-being of society in the long term.

A magnet for talent

While Japan may trail the US and China in the sheer volume of AI talent, its AI workforce is growing rapidly. More broadly, over the last 10 years, the number of foreign IT professionals in Japan has tripled to 85,000. Global tech giants such as Indeed have actively worked to bring engineering talent from top Asian universities to Tokyo. For Tokyo-based Sakana AI, an AI research startup that achieved unicorn status in less than 1 year, two-thirds of applicants are from overseas.

Anecdotally, many Asian AI researchers/engineers would prefer living in Japan to western countries, for reasons such as lifestyle/cuisine or cultural preferences and being only a short flight away from family members. It also doesn’t hurt that Tokyo and Osaka are the top two most liveable cities in Asia according to The Economist. Furthermore, Japan is in a similar time zone with countries with strong AI talent such as China, India, Singapore, and Korea, enabling close collaboration for remote teams. Granted, Japanese tech salaries have historically trailed those in the US, but more recently, salaries of ML engineers (especially in Tokyo) have been getting closer to those in the UK and the EU, while keeping the cost of living below, and quality of life above most of those countries. This now makes Japan a valid destination after graduation from e.g. UK universities with lots of AI talent. Furthermore, some foreign companies in Japan offer considerably higher salaries, especially the US tech giants like Amazon or Google, as well as certain startups flush with cash.

A thriving ecosystem

Tokyo, the most populous metropolitan area in the world, hosts a vast array of AI-related events such as meetups, conferences, pitch contests, and hackathons. For example, Tokyo AI (TAI), the biggest in-person AI community in Japan, despite being only 1 year old, has more than 1,700 Tokyo-based members, of which 65% are international. Even non-AI events increasingly involve AI components. For example, the Tokyo Game Show 2024 (about 300K attendees) featured an AI Technology Pavilion.

There are several thriving areas of Tokyo. One such example is Shibuya, which has become a hotbed for AI innovation. Known for its vibrant startup scene, Shibuya hosts numerous AI-focused organizations and incubators. For example, UP is a joint accelerator initiative between the local government and private sector which has attracted AI founders from around the world. UP’s portfolio startups bring products related to anime, music, logistics, agetech, and retail technology to the Japanese market. Programs such as UP act as a launchpad and provide valuable support to entrepreneurs, lowering the barrier to entry for international talent to contribute to the Japanese startup ecosystem. Also, given the success of the startup visa program in areas like Shibuya, the program has recently expanded nationwide.

Venture capital, a key part of the AI startup ecosystem, has also been ramping up. VC investment in Japan increased by an order of magnitude over the last decade. Prominent Silicon Valley VCs like Plug and Play, DNX, Geodesic, and DCM have been active in Japan for several years, and Andreessen Horowitz (a16z) is opening an office in Tokyo. In addition to considering Japanese startups for investment, a16z sees Japan as a promising market for its portfolio companies. There are many VC funds with a complete or partial focus on AI startups in Japan, including the Generative AI Fund by ANOBAKA and DEEPCORE (which runs the deep-tech focused KERNEL community with a number of AI startups). There are also UTokyo IPC and UTEC VC funds, with strong ties to the University of Tokyo, which are focused on deep tech and academic spin-outs. Also, over the past 2 years, three global accelerators launched in Tokyo: Antler, Techstars, and Alchemist. Of course, investment is not limited to venture capital. Recently, Blackstone, the largest private equity firm in the world, did a $1.7B deal in the Japanese manga industry, which is also increasingly affected by AI.

Many AI-native startups have been founded in recent years, and many other startups are actively developing products and services leveraging AI. A partial list of Japanese GenAI startups can be found here and here.

A supportive, pro-AI government

We’ve seen how the AI Act in Europe is hindering the AI industry and AI product releases. In contrast, in Japan, politicians and government officials are in general very supportive of AI. One possible reason is that people realize that with a shrinking, aging population, AI-driven productivity gains may be crucial for saving the economy. Regulation is reasonable for AI companies, and clarity is provided for key issues such as using copyrighted data for training AI models. Indeed, Yann LeCun, Chief AI Scientist at Meta, described Japan as a “machine learning paradise.” Lawmakers actively solicit the opinions of AI companies big and small, domestic and foreign, to figure out how to make the AI industry thrive while addressing concerns like safety.

Large support programs with big budgets are underway. In November 2024, Prime Minister Shigeru Ishiba unveiled a plan to spend over ¥10 trillion (approx. $65 billion) to support the semiconductor and AI sectors. Also, GENIAC (Generative AI Accelerator Challenge) is a project launched by the Ministry of Economy, Trade and Industry (METI) in early 2024 to enhance Japan’s capabilities in developing generative AI. Already 30 startups have been selected for support, which includes free access to considerable GPU resources.

AI research contributions: past, present, and future

Japan has a history of big contributions to AI research and also research more broadly, with the Tokyo-Yokohama region ranked as the #1 Science & Technology cluster in the Global Innovation Index and ~$140B annual R&D spend. Some notable contributions include:

• “A Theory of Adaptive Pattern Classifiers”, a seminal work in 1967 which helped pave the way to modern neural network optimization techniques
• The neocognitron, the precursor to modern convolutional neural networks, a core component of most computer vision and image generation systems.
• Chainer, the precursor to PyTorch. Developed by Preferred Networks, a Japanese AI unicorn.
• Sentencepiece, a tokenization algorithm used for many LLMs including some of Meta’s Llama models. The paper has ~4,000 citations.
• The “let’s think step by step” paper (Large Language Models are Zero-Shot Reasoners) surprised the world with the reasoning power unlocked by prompt engineering and inspired many other papers and experiments on prompt engineering and LLM capabilities. The paper has ~4,000 citations.
• The first paper to reconstruct viewed images from brain scans based on a diffusion model (~300 citations)
• Recently, AI Scientist generated considerable buzz for its interesting approach to using AI to conduct AI research autonomously. Japan has been researching the role of AI in scientific discovery for many years.

Granted, there are areas of AI research where the US and China are hugely outperforming in terms of citation count etc. However, there are some important areas where Japan is uniquely positioned to make large contributions to, such as AI R&D for comics/animation/games, for elderly care, or for robotics (Japan accounts for 38% of global robot production). Perhaps that is why Microsoft established an AI research lab in Tokyo, not to mention significant teams in Tokyo working at global AI powerhouses like Google Deepmind, Sony, Softbank, and IBM.

An exciting future for AI and Japan

While challenges remain—such as addressing gaps in domestic AI education and fostering homegrown talent—Japan’s trajectory in the AI landscape is undeniably promising. By combining government support and a thriving ecosystem of startups and corporations, Japan is poised to play a pivotal role in shaping the future of AI, especially considering its unique blend of culture and technology and its international positioning. As new investments, startups, and initiatives continue to emerge, Japan’s influence on the global AI stage will likely grow stronger.

We’re super excited to see how this plays out, and we’re committed to continuing our work to help this ecosystem grow.

About the authors

Jerry Chi is leading research at Mewtant, which runs the anime-style image generation platform PixAI and is developing an AI-powered manga/webtoon creation tool. Previously he was Head of Japan at Stability AI and held various machine learning / data science roles at Google, Supercell, SmartNews, and Indeed. Jerry holds degrees from Stanford (engineering) and U Penn. Also, as an ex-entrepreneur, Jerry does a bit of angel investing / startup advising as a hobby. He is deeply passionate about creative AI. Website: jerrychi.com.

Ilya Kulyatin is a 2x founder of data-driven Fintechs in Singapore and NYC, with global professional and academic experience in the US, EU, UK, Singapore, and Japan. He has a BSc in Economics, an MA in Finance, and an MSc in Machine Learning (UCL). For the past year, Ilya focused on his objective of making Japan a global AI hub, starting with building an AI community with 1700+ AI professionals based in Tokyo.