In a recent and very interesting interview, I should say, on “Bloomberg: The Asia Trade,” David Ha, CEO and Co-Founder of Sakana AI, shared his insights on the evolving landscape of artificial intelligence (AI), particularly focusing on efficiency and the democratization of AI technologies. Ha emphasized a shift in the narrative from building larger, more energy-consuming models to developing more efficient AI that can run on less powerful hardware.
Ha discussed Sakana AI’s approach with projects like Tiny Swallow, which utilizes a new model distillation method called TADE. This method enables the creation of models that are significantly smaller — about one-hundredth the size of traditional large models — yet maintain similar performance levels. Tiny Swallow, trained on Japanese data, exemplifies this innovation by functioning entirely on local devices like smartphones or within web browsers without needing an API, highlighting a trend for 2025 towards more efficient foundation models.
Addressing the emergence of Chinese AI startup DeepSeek, which aims to make AI skills cheaper and more efficient, Ha sees this as a sign of the times. He argues that the focus on scaling up models with massive investments is not the only path forward. Instead, the future of AI will see a proliferation of smaller, optimized models that cater to a broader range of applications and users, much like how mainframe computers evolved into personal computers and smartphones.
Ha envisions a future where AI models are not monolithic but diverse, with various models tailored for different needs, derived from open-source technologies. This diversification means that individuals and organizations won’t be locked into one vendor or model, challenging the current investment narrative that there will be a single dominant player in AI.
On the topic of Nvidia (NVDA), Ha disclosed their partnership, where Nvidia serves not only as a major investor in Sakana AI’s Series A funding but also as a collaborator in R&D. This collaboration focuses on using AI to enhance AI development, scientific discovery, and data center efficiency, particularly in Japan. Ha sees this partnership as crucial for fostering an AI ecosystem in Japan, aiming to ensure that core AI technologies are developed locally alongside global leaders like the U.S. and China.
Regarding fundraising, Ha mentioned that while there are always options for securing additional funds, Sakana AI is currently in a strong position post-Series A, with a primary focus on advancing their technology.
When questioned about the potential impact of more efficient AI models on Nvidia’s market position, given recent volatility in tech stocks, Ha suggested that rather than diminishing Nvidia’s prospects, the demand for GPUs would likely increase. This demand would shift from model training to model inference, broadening the user base and potentially benefiting Nvidia due to increased global compute needs.
David Ha’s vision for AI is one of accessibility and efficiency, where technology becomes a tool for widespread use rather than an exclusive domain of those with significant capital to invest in large-scale models. His insights suggest a future where AI is as integral and diverse in application as the devices we use daily.
WallStreetPit does not provide investment advice. All rights reserved.
Leave a Reply