Track Hyper | NVIDIA and SoftBank join forces to achieve a major breakthrough in telecommunications technology

Author: Zhou Yuan / Wall Street News

Even Masayoshi Son of SoftBank makes mistakes, and not just once.

On November 13, Jensen Huang, founder and CEO of NVIDIA, publicly "slapped" SoftBank founder, chairman, and CEO Masayoshi Son at the AI Summit Japan 2024: "Mr. Masa, can you imagine what would happen if you became the largest shareholder of NVIDIA?"

Son's reaction drew laughter from the audience: he hugged Huang and pretended to cry. Huang interjected, "We can cry together." Son admitted that he had made three investment proposals to NVIDIA in the past, but none of them materialized.

In June of this year, Son expressed regret for selling NVIDIA shares years ago.

In 2019, Son sold all of SoftBank's NVIDIA shares. This decision left him regretting it in the future, as he lost at least $150 billion.

As of the close of U.S. stocks on November 13, NVIDIA's market capitalization was $3.59 trillion, surpassing Apple ($3.4 trillion) to become the world's most valuable technology company.

Now, Son says he is ready to choose to press the "restart button" on Japan's technological development with Huang: both parties announced that NVIDIA will provide AI infrastructure for SoftBank's largest AI factory in Japan.

"We are going to create incredible value together. NVIDIA and SoftBank will collaborate to do something significant in Japan," Son said. "In the face of this new AI revolution, we cannot miss the moment."

On November 13, NVIDIA announced a significant achievement in collaboration with SoftBank: SoftBank successfully trialed the world's first combination of artificial intelligence and 5G telecommunications networks using NVIDIA's AI Aerial accelerated computing platform, marking a breakthrough in the telecommunications computing field and opening up potential AI revenue streams worth billions of dollars for telecom operators.

It sounds convoluted and confusing, right?

This is a new type of telecommunications network capable of running AI and 5G workloads simultaneously, known as Artificial Intelligence Radio Access Network (AI-RAN). This new infrastructure is supported by the telecommunications industry's ecosystem, as it allows operators to transform base stations from cost centers into AI revenue-generating assets.

This is very attractive.

This AI revenue platform allows telecom operators to achieve approximately $5 in AI inference revenue for every $1 invested in AI-RAN infrastructure (based on SoftBank's outdoor trials in Kanagawa Prefecture, Japan, and NVIDIA's estimates).

SoftBank's trials indicate that the AI-RAN solution can achieve operator-level 5G performance while utilizing excess network capacity to run AI inference workloads simultaneously.

After a comprehensive assessment of operating expenses and capital expenditure costs, SoftBank reached an estimated conclusion: each AI-RAN server added to its infrastructure can yield returns of up to 219%.

Huang said, "The AI Aerial accelerated computing platform can provide robust infrastructure support for future technology applications such as autonomous driving and remote robotic control, while also more effectively reducing power consumption." This is all about money.

For example, by dynamically allocating 5G and AI workloads based on demand within the same GPU, the AI Aerial platform can improve capacity utilization by 2-3 times, enhance energy efficiency and space utilization, and subsequently reduce the operational costs of isolated infrastructure.

NVIDIA's official press release claims that SoftBank will receive the world's first official DGX B200 system. This system will serve as the foundation for SoftBank's next-generation NVIDIA DGX SuperPOD solution supercomputer.

This supercomputer will also utilize the Quantum-2 InfiniBand network and, upon completion, will become Japan's most powerful AI supercomputer in history. In the future, it will support SoftBank's own generative AI development and AI-related businesses, and will be open to Japanese universities, research institutions, and enterprises.

SoftBank's next-generation supercomputer will also adopt NVIDIA's solution, targeting "extremely compute-intensive" workloads. The initial plan is to design it based on NVIDIA's Grace Blackwell platform, using the GB200 NVL72 rack system.

Masayoshi Son stated that SoftBank's goal is to create an AI market that meets local secure AI computing needs using NVIDIA AI Enterprise software. To achieve this goal, Masayoshi Son told Jensen Huang that SoftBank is prepared to purchase a large number of AI devices to build a more complete infrastructure.

In May of this year, SoftBank announced a $9 billion annual investment plan for artificial intelligence.

However, Masayoshi Son clearly believes that AI is a new technology distinct from the traditional IT industry, capable of giving rise to entirely new business models.

Therefore, Masayoshi Son believes that AI technology can not only truly rival human intelligence but that AI will be 10,000 times smarter than the human brain. To reach this level, Masayoshi Son said, "Companies will have more than 200 million GPU chips, with total capital expenditures reaching as high as $9 trillion. Considering the potential enormous profits that super artificial intelligence could bring, this could be a worthwhile deal."