Sakana AI's IPO Talk Tests Japan's AI Champion Bet Against Mercari's ¥680 Billion Mark
A Nikkei report of a listing as soon as autumn 2027 comes as the Tokyo developer adds U.S. intelligence veterans, builds its own models and publishes a Japanese finance benchmark.

What to know
- Sakana AI is preparing an IPO in Japan or the U.S. as early as autumn 2027 at a value that could exceed Mercari's ¥680 billion ($4.3 billion), Nikkei reported on Nov. 9.
- The Tokyo developer said on X that it has no near-term IPO plans, The Japan Times reported, after raising ¥32 billion in a Series B round in November 2025.
- Sakana AI said on Nov. 11 that former NSA Director Michael Rogers and former CIA station chief Tom Moran had joined as senior advisers on defense and cybersecurity.
TOKYO
Sakana AI, the Tokyo developer that bills itself as Japan's national AI champion, is preparing for an initial public offering in Japan or the United States as early as autumn 2027, at a market value that could exceed the ¥680 billion ($4.3 billion) set by flea-market app Mercari, Nikkei reported on Nov. 9.
That would make it one of the largest listings by a Japanese startup and, according to Nikkei, the first by a Japanese AI developer. For investors in Asia, the report puts a price tag on a question Tokyo has spent heavily to answer: whether Japan can grow a domestic model maker big enough to stand next to American and Chinese rivals, and whether public markets will pay for one.
The company pushed back. In a post on X on Monday, Nov. 9, Sakana AI said it had no near-term plans for an IPO, The Japan Times reported. Its priority, the post said, remains building "a generational AI company that serves as Japan's national AI champion."
What Nikkei reported about the Sakana AI listing
Goldman Sachs has been hired as an adviser, according to the Nikkei report as cited by The Japan Times. The listing venue is open: the Tokyo Stock Exchange, Nasdaq or both in a dual listing.
The timing is awkward for any AI float. The Japan Times noted that the Nasdaq was about 9% below its all-time high as of Nov. 10, partly because of a retreat in AI shares. Capital Times has also covered the Tokyo debut of AI firm Allganize earlier this month.
Sakana AI's private backers have so far paid up. The company raised ¥32 billion, or $200 million at an exchange rate of 160 yen to the dollar, in a Series B round announced on Nov. 17, 2025. Investors listed in that announcement include Mitsubishi UFJ Financial Group, Khosla Ventures, Macquarie Capital, Mitsubishi Electric, Salesforce Ventures, Google, Citi, JAFCO, Shikoku Electric Power and In-Q-Tel. MUFG Group Chief Executive Hironori Kamezawa said in that release that the bank had recognized Sakana AI's technology from the start and hoped AI's benefits would reach beyond banking to other Japanese industries.
From Chinese base models to its own
Investors weighing a Sakana AI listing will look closely at what the company actually owns. Its first large language model, Sakana Namazu, was built on Kimi 2.6, an open-weight model from China's Moonshot AI, according to The Japan Times. Moonshot's own listing plans are covered in our report on its Hong Kong IPO.
That dependence sits uneasily with Japan's push for home-grown models. The Japan Times reported that the government has made domestic language models a priority because of national security concerns about relying on Chinese and U.S. systems, and has reportedly put $6 billion toward Noetra, a consortium whose members include Sakana AI, SoftBank and Sony. On Oct. 26, according to the same report, Sakana AI said it was training its own model from scratch, a planned Sakana Sawara series with a large and a small version.
The company's research output this week points the same way. In a benchmark paper dated Nov. 17 on its website, Sakana AI introduced JFinBench, a test of 321 questions drawn from the 2026 editions of six Japanese finance certification exams, including the certified public accountant and tax accountant exams. Of the 12 recent models it tested, Moonshot's Kimi K3.1 Pro, which the company said is reported to rival OpenAI's GPT-6 Astra on mainstream benchmarks, trailed that model by about 10 points. The main cause of failure, Sakana AI said, was missing knowledge of Japanese rules and institutions rather than weak arithmetic.
For a company selling to Japanese banks, that finding doubles as a sales argument: general-purpose models still miss local tax and accounting detail.
Defense and intelligence move to the front
The other strand is national security. On Nov. 11, Sakana AI said Admiral Michael Rogers, U.S. Navy (Ret.), who led both the National Security Agency and U.S. Cyber Command from 2014 to 2018, and Tom Moran, who spent close to 30 years at the CIA and served six times as a chief of station, had joined as senior advisers. The company said they will advise on strengthening its defense business and its cybersecurity work, and that it is hiring engineers, project managers and technical editors for a team that supports Japan's defense and intelligence.
The appointments follow the arrival in September of AI researcher Jürgen Schmidhuber, according to the company.
That mix of finance, defense and sovereign-model work is what a listing would ask public investors to value, against the ¥680 billion yardstick Nikkei used. The next concrete marker is the Sawara series. Once those home-trained models are released, investors will be able to judge how much of Sakana AI's technology is its own, well before the autumn 2027 window Nikkei described.
Sources
- Nikkei Asia: Sakana AI gears up for one of Japan's biggest startup IPOs (Nov. 9, 2026)
- The Japan Times: Sakana AI may go public in 2027 (Nov. 10, 2026)
- Sakana AI: U.S. Admiral Michael Rogers and Former CIA Chief of Station Tom Moran Join Sakana AI as Senior Advisors (Nov. 11, 2026)
- Sakana AI: Announcing Our Series B (Nov. 17, 2025)
- Sakana AI: JFinBench: a Japanese financial benchmark built from the latest certification exams (Nov. 17, 2026)
Our standards: Capital Times reports under the principles set out in our Editorial Standards and Ethics Policy. Spotted an error? Report it to our editors.







