Japan's Sakana AI plans IPO as early as fall next year; weighs dual listing in the U.S. and Japan

【Financial News, Tokyo—Correspondent Seo Hye-jin】Japanese artificial intelligence (AI) startup Sakana AI plans to pursue an initial public offering (IPO) as early as fall next year. It is also considering a simultaneous listing on the U.S. and Japanese stock markets. The funds raised will go toward developing proprietary AI models and securing talent. The company aims to expand from combining existing foreign models to developing its own, reducing reliance on overseas AI in key fields such as defense and finance.
According to the Nihon Keizai Shimbun (The Nikkei) and the Yomiuri Shimbun on the 10th, Sakana AI recently appointed Goldman Sachs as an adviser and began preparing for a listing. In Japan, it is also reportedly in talks with Daiwa Securities Group and SMBC Nikko Securities, among others.
It is considering the Tokyo Stock Exchange and the U.S. Nasdaq, among others, as potential venues, with a target date as early as fall next year. If the plan proceeds as intended, it would be the first listing by a Japanese generative AI specialist.
Some observers believe its market capitalization based on the initial trading price could exceed Mercari's approximately ¥680 billion (about 5.7868 trillion won) when it went public in 2018. It could rank among the largest IPOs ever by a Japanese startup, but the listing venue and timing, offering size and valuation have yet to be finalized.
Sakana AI plans to use the funds raised through its listing to develop proprietary AI models, hire talent and expand sales. A key focus is “Sakana Sawara” (provisional name), a Japanese-made generative AI model announced last month. The company plans to release a smaller model next spring, then move on to developing a large-scale model.
Sakana AI has so far focused on technology that combines multiple existing AI models to lower costs while improving performance. “Sakana Fugu,” which became generally available in June, uses the same approach. Although the company operates in Japan, it uses models developed overseas, leaving its performance potentially subject to the limitations of the underlying models.
The proprietary generative AI model “Sakana Sawara” (provisional name) is being developed to address this limitation. Sakana AI is developing it in-house from the pretraining stage, in which vast amounts of data are fed into the model in advance. It plans to release a smaller model next spring and then scale up to a large-scale model. The strategy is to reduce reliance on foreign technology by going beyond operating AI in Japan and securing core models of its own.
The company has defense and cyber defense in mind as key areas of application. The more sensitive the information handled, the greater the need to reduce reliance on foreign technology and operate autonomously. Sakana AI already signed a contract in July with Japan's Ministry of Defense for a demonstration project involving AI-powered information analysis.
It is also expanding its customer base in the financial and corporate markets. It has secured MUFG Bank, Sumitomo Mitsui Banking Corporation and Sumitomo Corporation as clients, among others, and is pursuing demand for business operations innovation as well as from government and public institutions.
The company forecasts that annualized revenue next year will exceed ¥30 billion (about 255.3 billion won), roughly 10 times this year's projected figure. Annualized revenue is a metric that converts the revenue level at a given point in time to an annual basis.
The key challenge is to contain the costs of developing its own models while achieving competitive performance. As the company expands from combining existing models to carrying out pretraining itself, development capabilities and funding have become increasingly important.
Sakana AI plans to use AI in the development process as well to improve efficiency and reduce costs. Proving the performance and commercial viability of its proprietary models is expected to be a key challenge ahead of the listing.


[email protected] Seo Hye-jin Reporter