In November 2025, Sakana AI raised 20 billion yen (US$135 million) in its Series B, establishing a post‑money valuation of approximately 400 billion yen (US$2.635 billion).

Investors in this round included Mitsubishi UFJ Financial Group (MUFG), Khosla Ventures, Factorial, Macquarie Capital, Fundomo, Mouro Capital, New Enterprise Associates, Geodesic Capital, Lux Capital, Ora Global, MPower Partners, Shikoku Electric Power (STNet), and In‑Q‑Tel (IQT).

This brings Sakana AI’s total funding to approximately 52 billion yen (US$347 million).

Sakana AI builds nature-inspired AI systems that treat artificial intelligence as biological populations rather than monolithic models. The company's Evolutionary Model Merge technology automatically breeds new foundation models by combining layers and weights from existing models without retraining.

Users point the system at a pool of base models on platforms like Hugging Face and specify a task-specific fitness function. The platform generates hundreds of child models, benchmarks their performance, selects the top performers as parents, and repeats the process over multiple generations. This evolutionary approach produced EvoLLM-JP, a 7-billion parameter model that outperformed previous 70-billion parameter Japanese language models.

The company's AB-MCTS system orchestrates multiple frontier AI models at inference time through Monte Carlo Tree Search. When given a complex problem, the system decides whether to explore fresh solutions, refine promising answers, or switch between different AI providers like ChatGPT, Gemini, or DeepSeek based on which performs best for specific sub-problems.

ShinkaEvolve serves as a generalized evolutionary code-search engine that discovers new algorithms and agent architectures in hundreds rather than thousands of trials. The platform couples novelty-weighted sampling with AI-ensemble mutation to find optimal solutions across domains from mathematical optimization to reasoning agent design.

Sakana operates as a B2B AI research lab that commercializes nature-inspired artificial intelligence through enterprise licensing and partnerships. The company's go-to-market strategy centers on multi-year comprehensive partnerships with large Japanese corporations, as demonstrated by its foundational MUFG banking automation deal.

The business model leverages evolutionary computing to create specialized AI models at a fraction of traditional training costs. Rather than spending months and millions of dollars training models from scratch, Sakana's evolutionary approach can produce high-performing domain-specific models in days or weeks by intelligently combining existing foundation models.

Revenue comes primarily through software licensing and implementation services for custom AI solutions. The MUFG partnership includes both technology licensing and ongoing support for document automation workflows, suggesting a hybrid model that combines upfront licensing fees with recurring service revenue.

Sakana's cost structure benefits from its evolutionary approach, which requires significantly less computational resources than traditional model training. The company can deliver enterprise-grade AI capabilities without the massive GPU clusters typically needed for foundation model development, creating favorable unit economics compared to competitors building models from scratch.

Large Japanese technology conglomerates have launched their own foundation model initiatives backed by substantial internal resources. Rakuten developed its AI 2.0 model family integrated across its e-commerce and fintech ecosystem, while CyberAgent built CALM3-22B specifically for advertising technology pipelines.

SoftBank's Sarashina project benefits from ¥421 billion in government subsidies and dedicated NVIDIA Blackwell GPU clusters, with plans for a 390-billion parameter model. Preferred Networks targets enterprise customers with its PLaMo family, offering on-premises deployment for banks that cannot move sensitive data to public clouds.

These incumbents compete through vertical integration and proprietary data access, but their models require massive upfront investment and lengthy training cycles that Sakana's evolutionary approach can potentially circumvent.

Sakana competes directly with research-focused organizations like Anthropic and OpenAI in developing novel AI architectures and capabilities. Anthropic has gained traction with its Claude models driving significant API business growth, while OpenAI dominates consumer markets through ChatGPT integration.

These competitors focus on scaling transformer architectures and training increasingly large models, contrasting with Sakana's nature-inspired approach that emphasizes efficiency and specialization over raw parameter count.

The competitive dynamic centers on whether evolutionary model merging can match or exceed the performance of traditional large-scale training, with Sakana betting that biological inspiration provides a more sustainable path to AI advancement.

Japanese startups like Karakuri and Elyza are exploring alternative training approaches to reduce the computational costs of foundation model development. Karakuri's MoE model trained on AWS Trainium claims 50% GPU cost reduction, while Elyza focuses on efficient Japanese language processing.

These competitors share Sakana's focus on cost efficiency but pursue different technical approaches. The race centers on which methodology can deliver enterprise-grade performance at the lowest computational cost, with evolutionary merging competing against novel training architectures and specialized hardware optimization.

Sakana's evolutionary pipeline has expanded beyond language models to produce EvoVLM-JP for vision-language tasks and EvoSDXL-JP for image generation. This demonstrates the platform's ability to create specialized models across multiple AI modalities without requiring separate training infrastructure for each domain.

The AI Scientist and ShinkaEvolve tools transform Sakana's research capabilities into commercial software that can propose experiments, execute them, and generate research papers or code. Packaging these as lab-as-a-service offerings opens pharmaceutical, materials science, and corporate R&D markets far larger than pure model licensing.

AB-MCTS creates opportunities for managed inference services that optimize query allocation across multiple AI providers. This positions Sakana as an orchestration layer for cost-aware AI workloads, capturing usage-based revenue even when customers bring their own models.

The MUFG partnership establishes Sakana's credibility in financial services, creating opportunities with regional banks, insurers, and securities firms that require similar document automation and risk analysis capabilities. Many of these institutions participated in Sakana's Series A funding round, providing built-in sales channels.

Strategic investors including NEC, Fujitsu, ITOCHU, and KDDI offer embedded distribution into manufacturing quality control, supply chain planning, and telecommunications network optimization. These industrial applications benefit from Japanese-language processing and edge deployment capabilities that global competitors may struggle to match.

Sakana's evolutionary models can run on consumer-grade hardware, opening Japan's 3.5 million small business market plus robotics and IoT vendors requiring sub-7-billion parameter models for on-device deployment.

The evolutionary merging pipeline that created Japanese-optimized models can combine Korean, Thai, or Bahasa models with English experts to address underserved Asian markets. This approach avoids the trillion-token pretraining costs typically required for new language coverage.

Pan-Asian expansion leverages similar regulatory environments and enterprise software adoption patterns across the region, while Sakana's nature-inspired approach may resonate culturally in markets that value biological metaphors for technological systems.

Research translation: Sakana's valuation depends on successful commercialization of evolutionary AI research, but translation of academic breakthroughs into profitable enterprise products remains unproven. Revenue depends on evidence that nature-inspired approaches can consistently outperform traditional AI development methods at commercial scale.

Commoditization pressure: Open-source libraries and tools are replicating model merging capabilities, potentially commoditizing Sakana's core evolutionary techniques. As competitors like mergekit and Optuna Hub provide similar functionality, Sakana must continue advancing its methods to maintain technical differentiation and pricing power.

Scaling limitations: The evolutionary approach may face constraints when applied to the largest and most complex AI models that enterprises increasingly demand. If biological inspiration cannot match the performance of massive transformer architectures, Sakana's efficiency advantages may become irrelevant as computational costs continue declining across the industry.