For the last few years, the AI industry has settled for one common formula – you need a massive large language model (LLM) to run the smartest and most capable AI systems. Google, OpenAI and Anthropic have poured billions of dollars in large language models to power all the smartness we hail as AI. Japan’s Sakana AI, however, has a different approach, and it’s inspired by fish. You must have read about Fugu recently.
While the summer of 2026 was marked by geopolitical restrictions on AI exports, Sakana AI, a Tokyo-based startup, quietly flipped the script. They launched Fugu, an elegant and hyper-efficient AI system named after the Japanese blowfish. Unlike Claude Mythos or GPT-5.6, Fugu didn’t try to outscale the American titans.
Instead, it out-smarted them, all by coordinating smaller, specialised AI models into a single, seamless network, like a school of fish!
Almost overnight, Fugu put Japan at the centre of the global frontier AI map. And the architect behind this paradigm shift? David Ha, a former finance executive, is rewriting the rules of technology by looking at the natural world.
While Sakana AI achieved popularity overnight, its approach to developing AI models got early backing from major investors like Khosla Ventures, Google, Nvidia, MUFG and SBI Group. As a result, Sakana AI is presently valued at approximately $2.6 billion to $2.65 billion – a milestone reached following its $135 million Series B round in late 2025.
From Goldman Sachs to Google Brain: The journey of David Ha
Before he was training world-beating neural networks, David Ha was managing financial risk. In 2008, he was serving as the Managing Director and Head of Interest Rates Trading for Goldman Sachs in Tokyo. For almost a decade, his daily life was dedicated to the global markets and how it affected the economy. His intellectual curiosity, however, pulled him toward deep learning.
In the midst of his career in finance, Ha left Wall Street, earned a PhD at the University of Tokyo, and joined Google Brain. By 2016, he was leading Google’s entire AI research division in Japan.
While at Google, Ha started looking outside the box. While his peers were obsessed with buying more data centres, Ha looked to biology. He pioneered breakthroughs in “neuroevolution”, using evolutionary algorithms to organically “grow” AI software.
For Ha, nature had already solved the efficiency problems that Silicon Valley was trying to overcome with capital and brute force.
Assembling the Avengers
Hence, in 2023, Ha co-founded Sakana AI alongside Llion Jones – one of the co-authors of the historic Google paper that invented the Transformer. They wanted to build a school of small AI models that moved together flawlessly and were collectively adaptive, just like fish; hence the name Sakana.
Fugu was what the vision started to look like. Unlike existing AI LLMs that rely on one massive, expensive brain to solve a problem, Fugu acts as a smart digital conductor. When a user asks it to complete a complex task, Fugu automatically assembles a team of smaller AI experts – one to plan, one to write code, and one to verify the accuracy.
Speaking on Japanese television shortly after the launch, Ha summarised the strategy in a way that would get Marvel fans excited: “If you assemble a team of Avengers, that team will be better than the individual heroes. The same is true for AI.”
The best thing about Fugu is that geopolitics won’t affect its service. Because Fugu can swap out its underlying models at a moment’s notice, it is incredibly resilient. If a foreign vendor pulls API access due to changing trade laws (like the US restrictions), Fugu simply routes around the disruption, plugging in an alternative model without the user ever noticing a glitch.
No wonder that Nvidia’s Jensen Huang stated, “The team at Sakana AI is democratising AI in Japan.”
Fighting for AI sovereignty
With Fugu, Ha did not want to achieve a commercial milestone. Instead, he describes his creation as a way to bypass the dependence of the global AI power on just a few companies from the US. Ha wanted to push for AI sovereignty.
In his post on X (formerly Twitter), he wrote, “Access to top models can disappear overnight. Relying on a single company’s APIs for critical infrastructure, finance, or governance is a material vulnerability. This risk is no longer a hypothetical possibility, but a reality,” he said, referring to the restriction on Claude Mythos and Fable 5.
Ha, who is a graduate in Engineering Science and earned his master’s in Mathematical Finance from the University of Toronto, followed by a PhD from the University of Tokyo, proved that a lightweight and localised AI system could match the performance of Western tech giants, showing the world how it can settle for sovereign AI.
Ha, however, still believes in a deeply philosophical view to further his AI work. “Human intelligence is fundamentally a collective intelligence. We solve complex problems by participating in a vast cultural network that builds upon ideas across generations. I believe the strongest AI systems will become a collective intelligence, too,” he wrote.