Sakana AI
Tokyo's research-lab bet that the next big ideas in AI won't come from scaling.
sakana.ai ↗Sakana is what happens when a world-class research pair decides the best lab to build isn't in San Francisco. Ha and Jones are betting that scale isn't the only way forward — that evolutionary model merging, small efficient models, and multi-agent systems will open research doors that brute-force pre-training is closing. Whether that thesis plays out at frontier-model scale is uncertain; what's not uncertain is that Sakana is the only AI lab in Asia with the talent, capital, and distribution to find out.
What's going for them.
- 01Founded by Llion Jones — co-author of 'Attention Is All You Need,' the 2017 paper that created the modern LLM era — and David Ha (ex-Google Brain, Stability). Research pedigree no other non-US lab can match.
- 02Japan's most valuable AI unicorn at $2.65B, with a cap table that includes every major Japanese financial and telco institution: MUFG, Nomura, SMBC, NTT, KDDI.
- 03Evolutionary and nature-inspired AI methods — model merging, small-model efficiency, emergent multi-agent systems — a research agenda that's genuinely orthogonal to the 'train bigger' consensus.
- 04Enterprise positioning into Japanese banking, government, and industrial corporates where US labs cannot practically operate due to language, data residency, and relationship requirements.
- 05Defense-adjacent work with the Japanese Ministry of Defense on AI agents for logistics and information systems — a category no other AI lab in Asia has comparable access to.
What they built
Sakana ships foundation models optimized for the Japanese language, small-model efficiency, and evolutionary/nature-inspired architectures. The product line spans Evo (multi-agent research assistants), Tinkering (the internal research surface), and Japanese-optimized LLMs deployed with partners like NTT, KDDI, and major Japanese banks. The company’s research output — model-merging papers, evolutionary fine-tuning, AI Scientist self-generating research papers — has outsized influence in academic and applied ML circles relative to its headcount.
How they got here
David Ha and Llion Jones left Google and Stability in 2023 to start Sakana in Tokyo. The specific choice of Tokyo — not SF, not London, not Singapore — was the founding insight: Japan had world-class fundamental research, deep corporate capital, and effectively no frontier AI lab. Sakana’s $30M seed in January 2024 was followed by a $214M Series A at $1.5B valuation, then a ¥20B ($135M) Series B in November 2025 at $2.65B post-money.
The customer base tells the real story. Sakana’s cap table and commercial pipeline are dominated by MUFG, Nomura, SMBC, NTT, KDDI, Mitsubishi, and several Japanese government R&D relationships — the exact cohort that US labs have almost zero organic distribution into. This is a lab that’s being built as much by the Japanese corporate establishment as by its founders.
What’s ahead
Three tensions define where Sakana goes. First, research thesis: the evolutionary and multi-agent research agenda is genuinely promising but hasn’t yet produced a flagship consumer or enterprise product comparable to Claude or GPT. Second, commercial depth: the Japanese corporate pipeline is excellent, but Sakana needs to prove it can scale beyond lighthouse accounts into recurring revenue. Third, talent gravity: Tokyo is not yet a net importer of frontier AI researchers; Sakana will be the single biggest data point on whether a non-US lab can win global research talent.
Why it matters
Sakana is the most interesting AI lab outside the US–Europe–China triangle. For founders, it’s proof that geographic concentration isn’t destiny — a team can ship world-class research from Tokyo if the talent and capital align. For investors interested in Asia-Pacific AI exposure, it’s the cleanest single position, and one of the few AI companies in the region that’s not a foreign-lab regional franchise.
Founder interview coming soon.
We'll be sitting down with the founders and operators of the companies we profile — on fundraising, product decisions, and what they're building next. If you're part of the Sakana AI team and want to share a perspective, get in touch.
Thinking about fundraising or M&A?
Amafi Advisory works with AI companies on strategic, fundraising, M&A, and technical advisory. Even if you're just exploring.