Sakana AI Claims 100x Boost in AI Training Efficiency

Tokyo-based startup Sakana AI has unveiled a system that could dramatically accelerate AI development and deployment, potentially speeding up the process by 10 to 100 times, Nikkei Asia reports. The company's AI CUDA Engineer system automates the creation of code used to control Nvidia's GPUs, which are essential for AI training and inference.

"We believe the path to develop much stronger AI systems is to automate the development of AI using AI," Sakana stated in a press release.

The breakthrough centers on automating the generation of CUDA instructions, a software platform developed by Nvidia to optimize GPU performance for AI applications. Sakana's approach leverages its proprietary technology, inspired by biological evolution, to generate CUDA instructions from more general software code. These instructions are then "bred," selecting the most efficient offspring through a process of iterative optimization.

The pursuit of more efficient AI development is driven by the high cost of acquiring Nvidia chips and the massive energy consumption associated with advanced AI model training. Companies like OpenAI and other large American tech firms are currently engaged in a "money game," as some AI scientists describe it, investing heavily to develop superior AI models.

However, this approach is being challenged by companies like DeepSeek, which claims to have achieved comparable capabilities to larger AI models at a fraction of the cost through a unique development methodology. Sakana also suggests that the current high-spending approach to AI may not be sustainable in the long term.

"Currently, our AI systems consume immense resources, and if the technology continues to scale without thought for efficiency and energy consumption, the result will not be sustainable," Sakana said. "This project is an important step toward making AI a million times faster."

Sakana, co-founded in 2023 by CEO David Ha, a former leader of Google Brain's research team in Tokyo, and Llion Jones, another former Google engineer, aims to address the challenges of AI development by focusing on efficiency and sustainability.