SoftBank unveiled a software stack for its AI data centers, which it claims can “maximize GPU performance.”

The Infrinia AI Cloud Operating System was developed by a team of infrastructure architecture engineers who sit within SoftBank’s SB Telecom America subsidiary.

The software offering focuses on automating GPU drivers, networking, and storage components powering GPU platforms to reduce the operational burdens of managing both physical infrastructure and Kubernetes software layers.

Junichi Miyakawa, president and CEO of SoftBank, said the software-defined Infrinia stack will act as the core of its GPU cloud service, with it set to “seamlessly connect AI data centers, enterprises, service providers, and developers.”

SoftBank is positioning the internally developed Infrinia OS as a foundation for inference-as-a-service offerings. The Japanese giant suggests the stack will allow users to deploy services by selecting their desired large language model (LLM) without needing to directly manage Kubernetes or the underlying infrastructure.

The inclusion of OpenAI-compatible APIs provides users with drop-in integration with existing AI applications. It also facilitates the automation of operational maintenance, including system monitoring and failover. Infrinia also features tenant isolation capabilities to secure customer data when inference workloads are run on shared infrastructure.

The SoftBank stack makes use of networking components from long-time partner Nvidia to further boost its performance capabilities, with NVLink interconnect systems helping to reduce latency and maximize GPU-to-GPU bandwidth for highly distributed jobs

“The advancement of AI infrastructure requires not only physical components such as GPU servers and storage, but also software that integrates these resources and enables them to be delivered flexibly and at scale,” Miyakawa added. “Through Infrinia, SoftBank will play a central role in building the cloud foundation for the AI era and delivering sustainable value to society.”

While initial deployments of the Infrinia AI Cloud OS will focus on SoftBank’s own GPU cloud services, its Infrinia team will look to bring it to “overseas data centers and cloud environments with a view to global adoption.”

SoftBank's plans for a GPU cloud service come as the Japanese firm looks to be doubling down on the digital infrastructure market.

The Masa Son-led investment giant revealed to the world last December that it pipped 26North to snap up data center and network infrastructure firm DigitalBridge in a deal worth $4 billion.

That deal, which is expected to close in the second half of 2026, would see SoftBank acquire a firm managing more than $100 billion in assets, including 5.4 gigawatts (GWs) of data center capacity from providers such as Switch, Vantage Data Centers, and Yondr Group.

Included in that capacity are a near-gigawatt facility in Wisconsin linked to the SoftBank-backed Stargate project

Masa Son himself serves as Stargate’s chairman, while SoftBank is reportedly the driving force behind trying to secure investment for the OpenAI-led venture. The investment firm pumped $30 billion into an OpenAI funding round last April to cover some of the AI company’s Stargate costs. The ChatGPT maker did, however, come out this week saying it would commit to paying its own way in terms of energy for the project.

To fuel its Stargate investment, SoftBank sold its stake in Nvidia for $5.83 billion and a portion of its stake in T-Mobile US for $9.2 billion.