NVIDIA and SoftBank Corp. Partners to Quicken Japan’s Transition to a Global AI Powerhouse

NVIDIA announced a number of partnerships with SoftBank Corp. that will help Japan advance its global technological leadership and accelerate its sovereign AI ambitions, while also opening up billions of dollars in AI income prospects for telecom companies around the world.

NVIDIA founder and CEO Jensen Huang said during his keynote address at the NVIDIA AI Summit Japan that SoftBank is utilizing the NVIDIA Blackwell platform to construct Japan’s most potent AI supercomputer and intends to employ the NVIDIA Grace Blackwell platform for its upcoming supercomputer.

Also Read: Nvidia Eyes Partnership with India for Chip Development, Targeting Expanding AI Market

Key Insights:

NVIDIA also disclosed that SoftBank has successfully piloted the world’s first combined AI and 5G telecom network using the NVIDIA AI Aerial accelerated computing platform. This is a computing breakthrough that opens up AI revenue streams for telecom operators that could be worth billions of dollars.

NVIDIA and SoftBank also revealed that SoftBank plans to develop an AI marketplace that can satisfy the need for safe, local AI computing by utilizing NVIDIA AI Enterprise software. By supporting AI training and edge AI inference, this new service puts SoftBank in a position to become Japan’s AI grid and open up new economic prospects for the development, deployment, and use of AI services by the nation’s consumers, businesses, and industries.

Also Read: Denmark’s First AI Supercomputer Gefion Launched by Jensen Huang of Nvidia & Danish King

Blackwell is First Received by SoftBank, with Plans for Grace Blackwell

The world’s first NVIDIA DGXTM B200 systems, which will form the foundation of SoftBank’s upcoming NVIDIA DGX SuperPODTM supercomputer, are scheduled to arrive.

For its own generative AI development and AI-related operations, as well as that of Japanese universities, research institutes, and companies, SoftBank intends to employ its Blackwell-powered DGX SuperPOD.

AI-RAN Achieves a New Achievement

SoftBank has developed a new type of telecommunications network that can handle AI and 5G workloads simultaneously, referred to by the industry as an artificial intelligence radio access network, or AI-RAN, in close collaboration with NVIDIA. This is a significant technological advancement.

Because it enables operators to turn their base stations from cost centers into AI revenue-producing assets, this new type of infrastructure enjoys widespread ecosystem support from the telecom sector.

Also Read: NVIDIA NVLM 1.0: Know All About Open-Source Multimodal LLM

AI-RAN Is Used for Real-World Inference

SoftBank developed real-world AI inference applications for the trial using NVIDIA AI Enterprise, such as robotics control, multimodal retrieval-automated generation at the edge, and autonomous vehicle remote assistance. SoftBank’s AI-RAN network was able to support all inference workloads with optimal performance.

NVIDIA’s AI computing platform is tailored for SoftBank’s fully software-defined 5G radio stack, which includes SoftBank-enhanced L1 software built on NVIDIA AerialTM CUDA®-accelerated RAN libraries. In the future, SoftBank intends to include NVIDIA Aerial RAN Computer-1 systems into its solution, which it claims can consume 40% less power than conventional 5G network infrastructure.

Red Hat and Fujitsu are two NVIDIA and SoftBank partners that helped with the SoftBank AI-RAN solution trial.

Also Read: Nvidia’s New Llama-3.1-Nemotron AI Model Takes on OpenAI & Anthropic with Superior Performance

How are supply and demand balanced?

SoftBank wants to create an ecosystem that links the supply and demand of AI technology by utilizing its in-house orchestrator and NVIDIA AI Enterprise serverless application programming interfaces. This is because an AI-RAN solution must be able to dynamically spin compute up or down in response to supply and demand without sacrificing carrier-grade performance in real-time. This allows SoftBank to provide localized, low-latency, secure inferencing services by sending external AI inferencing jobs to an AI-RAN server when computing resources are available.