NVIDIA and Softbank build Japan’s fastest AI supercomputer using Blackwell. Now is the time for a non-“manufacturing” business revolution

NVIDIA will hold a big event titled “AI Summit Japan 2024” in Tokyo on the 12th and 13th. On the second day, Jensen Huang, CEO and founder of the company, gave a keynote speech and announced that SoftBank would introduce the latest Blackwell platform to build a supercomputer for AI. In addition, in the second half of the keynote, Masayoshi Son was invited to the stage for a talk show.

The content of the keynote is long, so I’ll list the presentations below first, but the main focus will be on the big partnership with Softbank.

Softbank chooses Blackwell to build Japan’s most powerful AI supercomputer

SoftBank acquired the “DGX B200” supercomputer platform system equipped with Blackwell and built the “DGX SuperPOD” supercomputer, which is expected to be the highest performance in Japan. In addition to being used in our own generational AI development projects and AI-related projects, we will provide it to universities, research institutes, and companies throughout Japan for AI development.

In addition to this DGX SuperPOD, we plan to build NVIDIA accelerated supercomputers to run very intensive workloads, with initial plans based on the Grace Blackwell platform together with Arm-node-based Grace CPUs, on liquid cooled, rack scale system.

The main announcement today is that Softbank is building Japan’s largest AI supercomputer DGX SuperPOD

NVIDIA CEO Jensen Huang with the Blackwell platform

Realizing AI-RAN that can run AI and 5G workloads simultaneously

AI-RAN runs AI and 5G workloads simultaneously on the same hardware. When 5G workloads are low, excess capacity is used to run AI decision workloads. Traditional communication networks are designed for peak loads, but on average only one third of their capacity is used. Further monetization can be achieved by giving the remaining two-thirds of capacity to users for AI decision-making services.

SoftBank has used NVIDIA AI Enterprise to build AI decision-making applications such as remote support for self-driving cars, robot control, and multi-modal RAG (research-enhanced generation), as well as a comprehensive NVIDIA Aerial CUDA Accelerated RAN Library found a 5G wireless stack. In an outdoor test conducted in Kanagawa Prefecture, we demonstrated the ability to simultaneously run an AI decision workload using excess capacity while achieving carrier-grade 5G performance.

The companies estimate that for every $1 invested in AI-RAN infrastructure, carriers will receive $5 in AI decision revenue. SoftBank believes it can achieve a profit margin of up to 219% per AI-RAN server, including operating costs.

Various edge AI applications will be realized through collaboration with Softbank.

The negative side of “manufacturing” in Japan

At this point, Mr. Huang called Mr. Masayoshi Son to the stage to discuss AI technology. Here, Mac’s motives, strategies and intentions for being involved in building an AI supercomputer are made clear.

Masayoshi Son (right) takes the stage

Looking back, Mr. Huang said, ”Mr. Son was also a major shareholder of NVIDIA,” and Mr. Mac showed a gesture of shouting at Mr. Huang, which delighted the audience.

“I have had many discussions with large Japanese companies, but they have always focused on the physical side of ‘manufacturing.’ Software is virtual, so they underestimate its value, I think it has been passed down from generation to generation.

However, the age when software has value has arrived, which means AI. Japan has been said to have seen “30 lost years”, but with the rise of AI, we believe there is an opportunity to reset the entire business stack and make up for the past 30 years lose

Fortunately for Japanese startups, unlike other countries, there is no specific move by the government to stop the AI ​​revolution. There are 350 AI startups in Japan, and for young entrepreneurs and innovators to take off, they need support, and infrastructure needs to be in place to make this happen. “SoftBank will support these entrepreneurs and innovators by building Japan’s largest AI supercomputer and making it available for almost free trial use.”

View from the conversation. Anyway, in the past there was a negotiation for Softbank to acquire NVIDIA while they were getting private meals.

He also said, “Just as Bill Gates advocated a PC on every desktop, and Steve Jobs put a smartphone on everyone,” I want to bring an AI agent to everyone.

Just as children as young as one year old can now operate smartphones naturally and without any struggle, in the future AI agents will grow up with children as young as one year old. He believes that if he becomes a doctor who helps people when they are sick, he will be a “personal Aristotle,” a teacher who knows everything. We can realize such an amazing future in Japan.”