Sakana AI Series B: A $135M bet against giant AI models

Tokyo-based Sakana AI has just closed a massive 20 billion yen ($135 million) Series B funding round, but it’s not to join the brute-force race of building ever-larger foundation models. Instead, the company is making a contrarian bet: that the future of AI will be defined by efficiency and evolution, not just raw computational power.

The funding brings in heavy-hitters like Mitsubishi UFJ Financial Group (MUFG), Khosla Ventures, and Macquarie Capital, alongside a notable investment from In-Q-Tel (IQT), the venture capital arm that serves US intelligence agencies. This isn’t just another cash infusion; it’s a validation of Sakana’s philosophy that genuine innovation comes from constraints.

While Silicon Valley giants are locked in a seemingly unsustainable arms race, burning through capital and energy to train models on the assumption of “near limitless resources,” as Sakana puts it, the Tokyo startup is taking a different path. “We believe that intelligent life has arisen not from an abundance of resources but rather from the lack of it,” the company stated in its announcement. “Nature ultimately selects systems that are able to do more with less.”

This principle is at the core of Sakana’s technology. Rather than training a new large language model from scratch, the company has pioneered methods to combine and improve existing models. Its research focuses on techniques like “Evolutionary Model Merge,” which fuses capabilities from different open-source models, and using tree search algorithms to orchestrate collaboration between closed models. In essence, Sakana is building AI that can self-improve and evolve, creating powerful new systems from the building blocks that already exist.

Evolution, Not Revolution

This approach is particularly resonant for Japan, a nation grappling with a declining workforce and limited domestic resources. Sakana AI is positioning itself as the architect of Japan’s “Sovereign AI”, an ecosystem tailored to the country’s specific cultural, industrial, and strategic needs. The company argues that for Japan, the right path isn’t to compete in the large-scale model competition but to innovate on the post-training and optimization layer, making frontier models work efficiently for Japanese enterprise.

And it’s already working. Over the past year, Sakana has built a significant enterprise business, partnering with some of Japan’s largest companies, including a strategic partnership with MUFG to develop custom AI for finance. Hironori Kamezawa, CEO of MUFG, noted that the investment aims to extend the benefits of AI “to Japan’s diverse industries.”

With the new funds, Sakana plans to double down on this strategy. The company is strengthening its applied AI team to deepen its work in finance while aggressively expanding into the defense, intelligence, and manufacturing sectors. The investment from IQT is a clear signal of the strategic importance of Sakana’s technology for national security applications, both in Japan and potentially for its allies.

Sakana is making a bet that the most powerful AI won’t necessarily be the biggest, but the one that is the most adaptable, efficient, and intelligently designed. As the industry questions the economic and environmental sustainability of the current AI trajectory, Sakana AI’s resource-constrained approach offers a compelling vision for a different, and perhaps smarter, future.