SoftBank Corp., under the leadership of President and CEO Junichi Miyakawa, has unveiled a significant development from its Infrinia Team. This team, specializing in the creation of next-generation AI infrastructure architectures and systems, has designed the Infrinia AI Cloud OS. This is a robust software stack custom-built for AI data centers.
The rollout of Infrinia AI Cloud OS provides AI data center operators with the ability to present kubernetes-as-a-service (KaaS) in multi-tenant environments. It also facilitates inference-as-a-service (Inf-aaS), which offers large language model inference via APIs. These features can be incorporated into operators’ individual GPU cloud services. Compared to custom-made or internally developed options, the software stack is anticipated to decrease total cost of ownership (TCO) and simplify operational intricacies. This facilitates a quicker deployment of GPU cloud services by providers while supporting the entire AI lifecycle, ranging from model training to inference, both flexibly and efficiently.
SoftBank’s initial strategy is to incorporate Infrinia AI Cloud OS into its own GPU cloud offerings. In the future, the Infrinia Team plans to extend the deployment to overseas data centers and cloud platforms with the aim of achieving global acceptance.
The request for GPU-accelerated AI computing is escalating swiftly across various sectors including generative AI, autonomous robotics, simulation, drug discovery, and materials science. As a result, user prerequisites for AI computing have become diverse and complex. These encompass access to fully managed infrastructure with abstracted GPU bare-metal servers, cost-effective inference services which eradicate the need for direct GPU management, and sophisticated operational models where AI training is centralized and inference is deployed at the edge.
Addressing these diverse needs places significant demands on GPU cloud service providers. Building and operating such environments call for in-depth technical know-how and involve complicated operational procedures. To surmount these hurdles, the Infrinia Team invented Infrinia AI Cloud OS. This solution maximizes GPU performance while easing the deployment and management of advanced GPU cloud services.
The software stack’s KaaS capabilities automate the entire infrastructure stratum—from BIOS and RAID configurations to operating systems, GPU drivers, networking, Kubernetes controllers, and storage—on advanced GPU platforms such as NVIDIA GB200 NVL72. Additionally, it supports software-defined, on-demand reconfiguration of physical connectivity and memory, which allows clusters to be created, modified, or removed based on specific AI workload requirements. Automatic node allocation based on GPU proximity and NVIDIA NVLink domains further diminishes latency and enhances GPU-to-GPU bandwidth for large-scale, distributed workloads.
The Inf-aaS component is crafted to empower users to deploy inference workloads with ease, providing scalable and efficient access to AI model inference via managed services.
CEO Junichi Miyakawa emphasized SoftBank’s commitment to pioneering the AI era. The company aims to provide essential capabilities required for the large-scale deployment of AI in society through a new GPU cloud service and software business. The Infrinia AI Cloud OS software forms the core of this initiative, designed to seamlessly connect AI data centers, enterprises, service providers, and developers.
What is Infrinia AI Cloud OS?
Infrinia AI Cloud OS is a robust software stack custom-built for AI data centers developed by SoftBank’s Infrinia Team.
What features does Infrinia AI Cloud OS offer?
It provides kubernetes-as-a-service (KaaS) and inference-as-a-service (Inf-aaS) capabilities. It can also be integrated into operators’ individual GPU cloud services.
What are the benefits of Infrinia AI Cloud OS?
Infrinia AI Cloud OS decreases total cost of ownership (TCO) and simplifies operational complexities. This facilitates quicker deployment of GPU cloud services by providers while supporting the entire AI lifecycle flexibly and efficiently.
