NVIDIA AI Day Tokyo, Sovereign AI as a key topic

Recently at the NVIDIA AI DAY held in Tokyo, representatives from the industry and technology sectors discussed Japan's development in the field of artificial intelligence, presenting forecasts and challenges for future demands. Kuniyoshi Suzuki, Senior Director of SoftBank Group's Cloud AI Services Division, stated that by 2030, Japan's demand for AI computing power will grow 320 times compared to 2020. He emphasized that it is imperative to develop locally autonomous technological foundations to meet the growing demand and ensure application transparency and security. This event is part of the global NVIDIA AI Days, attracting over 900 attendees, including developers, startups, researchers, and technology partners, focusing on topics such as AI application development, model training, Quantum Computing, and infrastructure construction.

Sovereign AI and government investment have become key issues.

During the conference, the concept of Sovereign AI was mentioned multiple times, referring to AI infrastructure supported by local technology and not relying on external vendors. This is also a key emphasis of the Japanese government's AI strategy. According to an official press release from Nvidia, the Japanese government has set a goal to invest at least 10 trillion yen (approximately 65 billion USD) by the fiscal year 2030 to promote the development of the semiconductor and AI-related industries, strengthening local supply chains and technological research and development capabilities.

Kazuya Ishikawa, an advocate for artificial intelligence at NEC, stated that specialized artificial intelligence targeting industries such as manufacturing, finance, and healthcare will drive digital transformation in Japan. The large language model of NEC cotomi can assist professionals in knowledge transfer or utilizing complex corporate documents, thereby directly addressing the skills gap and labor shortage issues faced by Japan; Japan's GENIAC program aims to enhance domestic generative AI capabilities by providing computing resources to businesses, facilitating collaboration, and supporting the development of foundational models, including large language models tailored for Japanese and industry adjustments.

Private enterprises and startups participate in the AI industry transformation

According to NVIDIA's press release, NVIDIA's Japanese partners are promoting the accelerated AI transformation of traditional industries in Japan. Members of the NVIDIA Inception startup accelerator program have launched innovative services developed based on the NVIDIA NeMo software suite; Stockmark announced the release of its 100 billion parameter fully temporary Japanese LLM as an NVIDIA NIM microservice, with inference speed increased by 2.5 times.

The Japanese advertising giant Hakuhodo's service company Hakuhodo Technologies announced that it will use NVIDIA AI Blueprints and NVIDIA NeMo Agent toolkits to develop AI agents for advertising.

Shimizu Corporation is a leading general contractor in Japan with a history of over 200 years. The company revealed how it is exploring NVIDIA AI blueprints for video search and summarization to monitor work progress and potential risks at construction sites.

During a NVIDIA conference at Tokyo AI Day, the newly released Nemotron-Personas-Japan (Note 1) was highlighted as the first open synthetic dataset that aligns with the real-world population, geography, and cultural distribution of Japan. This dataset provides a privacy-preserving and regulatory-compliant foundation for sovereign AI systems, allowing them to reflect the current state of Japanese society without relying on sensitive personal data. Japanese companies and technology developers are expected to continue investing in the fields of physical AI (such as robotics and autonomous driving systems) and AI agents (digital assistants that can help humans complete tasks) in the future. The newly released Nemotron-Personas-Japan open dataset will provide more training materials that are closer to local language, geography, and cultural context for AI applications in Japan, while reducing reliance on sensitive data.

In terms of enterprises and applications, several companies showcased ongoing AI applications and foundational model development. Stockmark announced its trained 100 billion parameter Japanese large language model (LLM), released through the NVIDIA microservices platform, enhancing model inference performance.

The exploration of open-source tools and medical applications continues to expand.

NVIDIA AI Day Tokyo also featured a special session on "Japan Healthcare Day," discussing how AI technology can be applied in the medical field, covering the open-source medical imaging framework MONAI, the digital health platform Holoscan, and Isaac for Healthcare, which is integrated with robotic applications. NVIDIA hosted workshops through its Deep Learning Institute (DLI) to train developers on using Large Language Models (LLM) to build Retrieval-Augmented Generation (RAG) agent systems.

The challenges of real-world applications and digital transformation coexist.

A technical representative from NEC pointed out that developing specialized AI for industries such as manufacturing, finance, and healthcare is one of the key strategies to promote digital transformation in Japan, and it is also expected to partially alleviate labor shortages and the issue of skill transfer.

In addition, the GENIAC project is also committed to strengthening Japan's autonomous development capabilities in generative AI by providing computational resources and model adjustment support, specifically tailored to the needs of the Japanese language and local industries.

Note 1: Nemotron-Personas-Japan is the first open synthetic dataset that encompasses the population, geography, and cultural characteristics of Japan. The dataset is licensed under CC BY 4.0, providing a privacy-preserving and regulatory-compliant foundation for AI systems reflecting Japanese society, without relying on sensitive personal data.

Nemotron-Personas-Japan employs NVIDIA's enterprise-level synthetic data generation system NeMo Data Designer, building on the successful experience of the widely used American Personas dataset. The release of these new versions marks the beginning of global synthetic character datasets and scripts that will support sovereign AI development across countries and regions. This dataset is designed to seamlessly collaborate with Nemotron models and other open-source LLM models, enabling easy fine-tuning for Japanese AI applications, from enterprise chatbots to domain-specific assistive driving robots.

This article NVIDIA AI Day Tokyo, Sovereign AI as a key topic first appeared in Chain News ABMedia.