Announcing Our Series B

Sakana AI, the Tokyo-based AI research outfit, has closed a Series B round worth 32 billion yen, or roughly 200 million US dollars, and it's using the moment to make a pointed argument about where the rest of the industry is headed. The company says the current wave of AI investment is pouring record capital into compute, often funding businesses with no clear route to profitability, and burning through energy at a pace that assumes resources are basically infinite. Sakana's founders are asking, bluntly, whether that's the model Japan should be copying.

Their answer is no. The company frames itself as betting on constraint rather than scale. Instead of training another giant foundation model from the ground up, Sakana has built its research around combining existing open-source models through evolutionary methods, using tree search to merge closed models, and pushing models to self-improve. Last year the company also used LLM agents to automate parts of AI science itself, aiming to let AI discover more efficient algorithms on its own, and it's been building energy-efficient language models meant to run on edge devices rather than in massive data centers.

On the business side, Sakana has spent the past year building what it calls a healthy and growing enterprise AI operation, working with major Japanese companies including MUFG and Daiwa Securities Group on custom finance applications. That work exposed something the company keeps returning to: general-purpose models are good at generic tasks but miss the tacit, uncodified knowledge that professional work actually runs on. Closing that gap, Sakana argues, takes engineers embedded deeply in a specific domain, not another bigger model. The company is now extending that approach beyond finance into defense, intelligence and manufacturing.

The funding round drew backing from a long list of new and existing investors, including MUFG, Google, Citi, Salesforce Ventures, Khosla Ventures, and In-Q-Tel, among others. Sakana says the money will go toward accelerating both its research and its push to embed AI into Japanese business and public-sector applications, with a specific focus on post-training work, tailoring existing frontier models to Japan's own culture and needs rather than competing head-on in the race to build ever-larger base models.

Sakana was established about two years ago, and it's framing this raise as the start of what it calls its most exciting phase yet, moving research out of the lab and into deployment across industry and government. Given Japan's shrinking workforce and aging population, the company's read is that the payoff from getting this right could be substantial, even if the path there looks nothing like the compute-maximalist approach dominating headlines elsewhere.