NVIDIA Helps Build AI Factories Faster Than Ever with NVIDIA DGX SuperPOD

In a cavernous room at an undisclosed location in Japan, a digital revolution is unfolding. Racks of servers stand like giants, their sleek frames linked by thousands of cables humming with potential.

Until last year, this sprawling AI factory didn’t exist. Now it’s poised to anchor SoftBank Corporation’s vision for AI-powered innovation, a vision rooted in creating a society that coexists with AI, using advanced infrastructure to drive progress. 

For SoftBank, every day shaved off deployment isn’t just a technical win. Instead, it’s various competitive advantages. In the rapid-fire world of AI supercomputing, speed is everything.

SoftBank, one of Japan’s leading tech giants, partnered with NVIDIA to bring this system to life. Together, they’ve built two of the most advanced computing clusters in the world—clusters poised to process vast amounts of data at record-breaking speed and accelerate the development of large language models, or LLMs, being developed by SoftBank subsidiary SB Intuitions.

However, creating an AI factory isn’t just about plugging in hardware. It’s a carefully choreographed effort where every cable, connection, and component must align perfectly.

That’s where NVIDIA Infrastructure Specialists (NVIS) comes in. NVIS is an expert team that accelerates AI deployments with precision and efficiency. Using a proven playbook, NVIS enables customers to transform bare metal into production-ready AI infrastructure in record time.

Speed through collaboration

When SoftBank asked for an accelerated timeline, NVIDIA was ready to help turn an ambitious goal into reality. Despite SoftBank’s limited experience with large-scale AI platforms, its agility—combined with NVIDIA’s expertise and reference architecture—ensured success.

SoftBank’s NVIDIA DGX SuperPOD deployment was precisely orchestrated. Every step was mapped to milestones and timelines, with the project split into two clusters to help manage the pace.

When SoftBank requested an earlier finish 10 days before the initial planned date of operation, NVIS recalibrated to meet the new deadline. This kind of agile response called for careful resource management, precise logistics and daily coordination.

The numbers tell the story. Thousands of cables and hundreds of network switches connected 510 NVIDIA DGX B200 systems, the powerful computing nodes at the heart of a DGX SuperPOD. Both clusters delivered exceptional FP64 precision performance: one reaching 89.78 gigaflops and the other 91.94 gigaflops, a measurement of their ability to process vast amounts of data per second.

“The precision and speed of deploying 510 NVIDIA DGX systems across these DGX SuperPOD clusters highlight what’s possible when expertise and collaboration align,” said Hironobu Tamba, vice president of data platform strategy and division head of the technology unit at SoftBank. “This infrastructure isn’t just fast—it’s setting a new standard for AI development in Japan.”

With the largest AI computing infrastructure in Japan, SoftBank is now well-positioned as a leader in the country’s AI ecosystem.

Overcoming challenges with precision and agility

Of course, no project of this scale unfolds without its share of hurdles. In SoftBank Corp.’s case, limited power availability required NVIS to conduct some testing during off-hours. When connectivity issues arose due to networking components, the team found a creative solution, repurposing parts from the second cluster to keep the first on schedule.

NVIDIA teams meticulously tracked and resolved every challenge in real time, ensuring the deployment stayed on schedule.

“This wasn’t just a job,” one team member said. “It was more like a Formula 1 pit crew, working in sync, each of us knowing that every day we saved meant more value for SoftBank.”

Pioneering Japan’s AI future

SoftBank’s deployment is more than a technical achievement. It’s a step toward building next-generation social infrastructure, critical for accelerating in-house projects like LLM development and enabling a broader ecosystem of generative AI developers across Japan.

When it’s operational, this platform will also serve external companies, contributing to the development of Japan’s domestic AI capabilities.

“SoftBank Corp.’s AI factory is a leap forward for Japan, accelerating innovation and empowering the country’s AI ecosystem to thrive,” Tamba said. 

NVIS playbook: Speed, accuracy and collaboration

SoftBank’s deployment showcases the NVIS playbook, a streamlined approach to managing large-scale AI projects quickly and precisely. With NVIS, NVIDIA can install thousands of GPUs daily, transforming massive AI centers into efficient AI factories.

At the heart of SoftBank’s DGX SuperPOD is the NVIDIA Quantum-2 InfiniBand networking platform, designed to handle immense data loads without bottlenecks. This networking technology, combined with the digital twin capabilities of NVIDIA Air, a virtual replica of the data center that allows for pre-deployment testing and validation, ensured that the project ran smoothly from planning to execution.

CEO math: Why speed matters

For companies that adopt NVIDIA’s technology, speed isn’t just technical, it’s strategic. The faster its new AI factory goes live, the quicker it can gain insights, streamline operations, and launch new AI-driven products.

Internal NVIDIA analysis suggests that if the installation time was accelerated from a typical 6+ months to just three weeks, customers can avoid up to $150M in costs associated with prolonged deployment downtime for large AI systems. This estimate is based on the operational costs of a large-scale deployment, which NVIDIA calculates at approximately $1M per day for a 1K-server data center with 8K GPUs.

Faster deployment also enables customers to generate revenue sooner by running workloads such as LLMs without delays. Each additional day of operation unlocked by NVIS means avoiding $1M in expenses and capturing revenue opportunities otherwise lost due to downtime.

Build faster with DGX SuperPOD

Accelerate your DGX SuperPOD deployment with NVIDIA Infrastructure Specialists (NVIS) and cut your time to first training to a fraction of the norm. For more information, see AI Infrastructure Services | NVIDIA NVIS.