Sakana AI

サカナAI / サカナAI株式会社

A Tokyo-based applied-AI company developing model-optimization methods and custom enterprise implementations for Japanese sovereign and regulated workflows.

Private company
Profile as of September 14, 2026
Not ranked: no published revenue or market capitalisation
How the ranking works →

Sakana AI is one of Japan's best-capitalized private AI companies, combining frontier-model research with strong access to domestic financial and industrial institutions. Its valuation reflects confidence in its research talent and strategic positioning, though its revenue, margins, and commercial repeatability remain undisclosed.

Key figures

Revenue: Not disclosed
Revenue growth: Not disclosed
Operating margin: Not disclosed
Net income: Not disclosed
R&D spend: Not disclosed
Capital expenditure: Not disclosed
Series B funding 2025-11-17: 32 billion yen (US$200 million)
Cumulative capital raised Post-Series B: 66 billion yen (US$412 million)
Post-money valuation Post-Series B: 432 billion yen (US$2.7 billion)

Overview

Sakana AI is a private Tokyo-based frontier-AI research and applied-AI company founded in July 2023 by David Ha, Ren Ito, and Llion Jones. Its commercial model centers on using or combining existing frontier and open-source models, then adapting systems for Japanese enterprise workflows and sovereign-AI requirements. This makes its technical position different from a hyperscaler or a foundation-model developer that bears the full cost of training a broad base model from scratch.

Named technologies include Evolutionary Model Merge, AB-MCTS, The AI Scientist, Darwin Gödel Machine, ShinkaEvolve, Continuous Thought Machine, Namazu Japanese-language models, Sakana Chat, Sakana Marlin, and Sakana Fugu. The company targets Japanese language, culture, domain knowledge, and regulated enterprise contexts through post-training and application engineering.

The AI angle

Sakana AI supplies AI research, model-optimization methods, enterprise implementation services, and user-facing software rather than chips, servers, data-center capacity, or cloud services. AI demand reaches revenue through custom enterprise deployments and associated research and engineering work. The connection is direct, but the company does not disclose revenue, bookings, number of paying customers, pricing, segment revenue, or the share of revenue tied to any product.

The company depends heavily on external compute and models. NVIDIA is a disclosed research and infrastructure collaborator, with Sakana AI receiving early access to NVIDIA-supported data centers in Japan for experimental work. Its model-merging and multi-model orchestration approaches also depend on access to existing open-source and closed frontier models. Specific cloud providers, GPU quantities, model licences, inference costs, and compute commitments are not disclosed.

Sakana AI matters most in the model-developer and enterprise-application layers because it combines frontier-AI research with Japanese-domain post-training and direct implementation for regulated or specialized workflows.

Technology and moat

Sakana AI's main technical differentiation is its focus on computationally efficient methods that combine, orchestrate, adapt, or improve models rather than competing primarily through the largest pre-training runs. Evolutionary Model Merge combines capabilities from existing open-source models. AB-MCTS uses adaptive branching Monte Carlo tree search for collective reasoning across multiple models. The Darwin Gödel Machine and ShinkaEvolve pursue self-improving code and program evolution, while The AI Scientist automates portions of scientific research workflows.

This approach lowers the requirement for proprietary hyperscale training clusters. Against domestic Japanese competitors, Sakana AI has a founder team with direct frontier-model research credentials, including Llion Jones's co-authorship of the original Transformer paper and David Ha's prior Google Brain leadership in Japan. However, its defensibility rests on research talent and accumulated implementation knowledge rather than exclusive patents, and its research methods are substantially public through papers and code.

Five-pillar assessment

Scale and market position: A private early-stage company with significant venture funding but undisclosed revenue, customer count, and commercial scale.
Technology and R&D: Differentiated by computationally efficient model-merging, multi-model orchestration, and self-improvement methods rather than proprietary hyperscale pre-training.
Supply-chain centrality: Strong strategic investor base and partnerships with major Japanese financial institutions, alongside infrastructure collaboration with NVIDIA.
Financial momentum: Rapid valuation growth to approximately 432 billion yen (US$2.7 billion) in its Series B, though operating financial momentum is undisclosed.
Governance and quality: Independent private company with limited transparency regarding board composition, ownership percentages, and financial operating metrics.

Supply chain and relationships

Sakana AI depends on external compute and models to function. NVIDIA is a disclosed research and infrastructure collaborator, providing early access to NVIDIA-supported data centers in Japan. The company's model-merging methods also depend on access to existing open-source and closed frontier models, exposing it to changes in third-party licences, APIs, and pricing.

On the customer side, named enterprise recipients include Mitsubishi UFJ Financial Group, Daiwa Securities Group, Mitsubishi Electric, Citi, and Datadog. The company has strategic partnerships with MUFG and Daiwa Securities Group for custom AI in finance, and plans to extend its applied work into defense, intelligence, and manufacturing. Customer revenue shares and concentration are not disclosed, but its disclosed applied-AI activity emphasizes finance and Japan's public sector.

Customers

Mitsubishi UFJ Financial GroupCustom AI for financial-services applications
Daiwa Securities GroupStrategic partnership for custom AI in finance

Suppliers

NVIDIAResearch, infrastructure, and data-center access

Partners

DatadogCombining AI research with observability and security capabilities
Mitsubishi ElectricPotential collaboration involving its Serendie digital platform
CitigroupFinancial-services innovation collaboration

Competitors

OpenAIInferredCompetes for Japanese enterprise generative-AI workloads
AnthropicInferredCompetes for enterprise foundation-model deployments in Japan
Google DeepMindInferredCompetes through foundation models and cloud ecosystem
Preferred NetworksInferredCompetes for Japanese industrial and enterprise AI projects
NTTInferredCompetes through domestic AI research and enterprise relationships

Geopolitics and risk

Sakana AI positions itself around sovereign AI and Japan-specific deployment, but regional fragmentation may constrain model access, cross-border data handling, foreign partnerships, and the availability of advanced compute. Its research and applications require externally supplied accelerated compute, exposing it to changing United States export controls that could affect the availability, cost, or deployment options of the compute it uses.

Expansion into defense, intelligence, finance, and misinformation mitigation raises requirements for security, safety, procurement compliance, data governance, explainability, and human accountability. The company was selected in June 2025 for a Ministry of Internal Affairs and Communications project concerning technologies to counter online false and misleading information, highlighting its exposure to domestic public-sector priorities.

Risk matrix
Risk	Severity	Why it matters
External GPU and data-center dependence	High	Research and applications require externally supplied accelerated compute.
United States export-control exposure	High	Controls could affect the availability, cost, or deployment options of the compute it uses.
Frontier-model platform dependence	High	Methods depend on access to open-source and closed models outside its control.
Commercial concentration in Japanese regulated sectors	High	Exposure to long procurement cycles, security obligations, and concentrated domestic demand.
Revenue and customer-concentration opacity	High	Lack of disclosure limits outside assessment of its commercial durability.
Competition from global model providers	High	Larger competitors can improve model quality or reduce prices faster.
Sensitive-use regulation	Medium	Expansion into defense and intelligence raises requirements for security and compliance.
Japan-China-United States technology fragmentation	Medium	Regional fragmentation may constrain model access and foreign partnerships.

Governance and ownership

Sakana AI is an independent private Japanese company. It is not state-owned, state-influenced through a government ownership stake, part of a chaebol, part of a keiretsu, or controlled by an industrial parent. Its investor base includes Japanese financial institutions and corporates alongside foreign venture firms and strategic investors such as NVIDIA, Google, Citi, Salesforce Ventures, Datadog, Khosla Ventures, NEA, and Lux Capital.

The company does not publish a full board roster, board-independence assessment, voting-rights structure, capitalization table, founder ownership, or audited financial statements. Its regulatory posture is shaped by its work in financial services, government-related misinformation countermeasures, defense, and intelligence.

What to watch

Whether Sakana AI publishes revenue, annual recurring revenue, customer count, gross margin, or a first audited financial statement after its Series B.
Whether the MUFG Bank and Daiwa Securities Group relationships progress into disclosed production deployments, contract values, or multi-year commercial revenue.
Whether Sakana AI names a proprietary Japanese model, reports benchmark performance, and documents the compute, data, or post-training method behind it.
Whether NVIDIA-supported Japanese data-center access becomes a defined long-term compute arrangement with disclosed capacity, GPU type, or cloud partner.
Whether expansion into defense, intelligence, and manufacturing produces named contracts while meeting Japan's security and public-procurement requirements.
Whether the company's evolutionary and multi-model research becomes productized software with disclosed adoption beyond bespoke consulting and implementation engagements.

Recent developments

February 24, 2026Citi announces a strategic investment in Sakana AI and a financial-services collaboration.
November 17, 2025Sakana AI announces a Series B of approximately 32 billion yen and US$200 million.
August 7, 2025Sakana AI holds its first Applied Engineer open house.
July 1, 2025Sakana AI introduces AB-MCTS, an inference-time scaling method that uses adaptive branching Monte Carlo tree search.
June 1, 2025Sakana AI identifies its selection for a Ministry of Internal Affairs and Communications project on countermeasures against online false and misleading information.
May 19, 2025Sakana AI announces a multiyear partnership with MUFG Bank, extending its enterprise focus in Japanese financial services.

Coverage on AsiaAI.FYI

Microsoft Plans to Triple Data Centers by 2032, Locking in a Decade of HBM and Foundry Demand September 11, 2026
Is OpenAI’s Astra halt the ultimate wake-up call for AI security? August 19, 2026
Can Anthropic’s Easily Erased Watermark Truly Protect Corporate IP? August 16, 2026
Will Samsung’s 18x Profit Surge Finally Convince Investors the Memory Glut Is Over? July 7, 2026
China Minerals Threatens EU; AI Warfare Dominates Japan, WeChat. June 23, 2026

Guides that cover Sakana AI

About this profile

Compiled with AI-assisted research from company filings, market data, and published reporting as of September 14, 2026, then reviewed by AsiaAI.FYI. Figures marked Estimate are not company-reported. Check primary filings before relying on any number.

Confidence: C. Company announcements substantiate identity, funding, partnerships, and selected technology claims, but audited financial statements and cap-table data are not public.

Main sources: Sakana AI corporate-information disclosures; Sakana AI funding announcements; Product and research announcements; Partnership announcements; Company blog updates.

All 24 sources

Position in the AI stack

Develops Japanese-oriented models, model-merging methods, and custom AI applications for enterprise and public-sector use cases.

Legal name: Sakana AI K.K.
Headquarters: Tokyo, Japan
Founded: 2023-07
Founders: David Ha
Ren Ito
Llion Jones
Leadership: David Ha, Co-Founder and Chief Executive Officer
Ren Ito, Co-Founder and Chairman
Llion Jones, Co-Founder and Chief Technology Officer
Ownership: Private company
Business: AI research, model optimization, and enterprise implementation
Sector: AI Software & Models
Website: sakana.ai