(This article contains 1,142 words and takes approximately 3 minutes to read)

On November 13, 2024, at the Japan AI Summit held in Tokyo, NVIDIA officially announced a partnership with SoftBank to accelerate the rise of AI technology in Japan. As a leading semiconductor company globally, NVIDIA is collaborating with SoftBank to build Japan's most powerful AI supercomputer, paving the way for Japan to become a global AI powerhouse.

The core of this collaboration lies in NVIDIA's newly released Blackwell platform. SoftBank plans to construct an AI supercomputer based on the x86 architecture using this platform, becoming one of the first companies globally to acquire the NVIDIA DGX B200 system. This system not only includes x86 processors but is also equipped with eight B200 GPU modules, offering powerful parallel computing capabilities and the ability to handle massive data training tasks, making it particularly suitable for the development of generative AI models.

SoftBank's supercomputer will become Japan's most potent AI supercomputing device, aiding Japan in gaining a significant advantage in the global AI technology competition. SoftBank also plans to build an even more advanced supercomputer in the future, based on NVIDIA's Grace Blackwell platform. This platform will utilize the GB200 NVL72 multi-node liquid-cooled system, combined with energy-efficient Arm-based Grace CPUs and Blackwell GPUs, enhancing computational efficiency and energy consumption performance. This supercomputer is expected to be dedicated to extremely dense computing workloads, such as training large language models and other complex AI applications.

Beyond the supercomputer project, NVIDIA and SoftBank are also advancing the construction of the world's first 5G AI-RAN (Artificial Intelligence Radio Access Network). AI-RAN is a novel telecommunications network architecture that integrates AI and 5G technology, aiming to transform traditional 5G base stations from mere data transmission platforms into efficient AI inference centers. Traditional 5G networks often have two-thirds of their capacity idle during off-peak times due to the need to handle peak loads, whereas AI-RAN can fully utilize this idle capacity by converting it into AI inference services, generating additional revenue streams.

SoftBank has demonstrated in initial tests that this innovative network can support emerging application scenarios such as autonomous vehicles and remote robots, effectively reducing network operating costs while enhancing the commercial viability of telecommunications networks. NVIDIA and SoftBank estimate that for every dollar spent on capital expenditures for AI-RAN infrastructure, telecom operators can expect to generate $5 in AI inference revenue. Such high returns have attracted the attention of telecommunications companies worldwide and laid the foundation for deeper integration of AI and 5G technology in the future.

While SoftBank has become the first enterprise to deploy the DGX B200 system, it is not the only player in the global AI supercomputer race. Foreign media reports indicate that Microsoft is also conducting AI device tests based on the Blackwell platform and is poised to become the first company to deploy these devices. NVIDIA's DGX B200 system has attracted a wave of procurement from major global enterprises due to its exceptional performance. Companies such as Tesla and xAI, owned by Elon Musk, have already purchased tens of thousands of NVIDIA processors for training their AI models. Whether SoftBank can secure sufficient GPU resources to compete with these companies remains uncertain, but it is clearly determined to make a significant impact in the AI field.

At the AI Summit, NVIDIA CEO Jen-Hsun Huang and SoftBank CEO Masayoshi Son engaged in an in-depth dialogue about the future of AI. Son revealed that he had missed three opportunities to become a major shareholder in NVIDIA for various reasons, which became an unfillable regret for him. However, through this collaboration, SoftBank's strategic layout in AI computing and telecommunications may compensate for this regret , helping it rise again in the AI era.

*Disclaimer: The above content is compiled from online sources and is intended for communication and learning purposes only. If there are any content or copyright issues, please contact us via comments for removal.