The world of AI infrastructure is poised for transformation as SoftBank (SFTBY) recently announced a collaboration with semiconductor powerhouse AMD. Despite witnessing a modest decline in share prices this week, the partnership promises groundbreaking advancements in GPU utilization for increasingly complex artificial intelligence workloads.
At the core of this joint venture is the exploration of AMD’s cutting-edge Instinct GPUs aimed at enhancing SoftBank’s proprietary resource management platform, Orchestrator. Originally crafted to meet the demands of both AI computing and telecommunications, the system is now being engineered to efficiently manage multiple AI applications on a single GPU.
Unlocking the Potential of AMD GPU Partitioning
This collaboration primarily focuses on the validation of AMD’s innovative GPU partitioning technology. This technology allows a single GPU to function like several logical devices, enabling a dynamic allocation of computing resources tailored to the size of various AI models and the intensity of concurrent workload demands.
SoftBank’s Vice President, Ryuji Wakikawa, underscored that this partnership aims to enhance resource efficiency across AI applications. Meanwhile, AMD’s Kumaran Siva pointed out the flexibility this technology offers for diverse AI services. This capability enables enterprises to “fractionally lease” GPU resources, significantly reducing costs and improving scalability.
Revamping Orchestrator for Enhanced Performance
At the heart of SoftBank’s journey into AI-RAN (Artificial Intelligence–Radio Access Network) solutions, the Orchestrator plays a vital role in dynamically reallocating computing resources between AI functions and virtualized telecom services. Until recently, this platform primarily operated on NVIDIA’s GH200 Grace Hopper Superchip-based infrastructure. The integration of AMD GPUs symbolizes a strategic expansion, paving the way for more versatile multi-use deployments.
By enabling the simultaneous management of AI and telecom operations on the same server, SoftBank is set to fulfill growing demand for AI-driven services. The ability to support various workloads on a singular GPU is a crucial advancement, aiming to drastically enhance performance and minimize hardware redundancy.
Realizing Multi-Tenant AI Infrastructure
Central to this initiative is AMD’s innovative CPX mode, permitting one Instinct MI300X GPU to be divided into as many as eight independent logical devices. This multi-tenant setup is particularly advantageous for handling smaller AI inference tasks and developmental workloads without the need for dedicated hardware.
Platforms such as ClearML can efficiently run multiple AI tasks simultaneously on a single physical GPU. Red Hat’s OpenStack Services on OpenShift further support multi-tenant environments, proving compatibility with both NVIDIA and AMD GPUs. The collaboration reflects an accelerating trend towards adaptable and scalable AI infrastructure, catering to the evolving needs of enterprises and cloud providers alike.
Showcasing Innovations at MWC Barcelona 2026
Looking ahead, SoftBank and AMD have scheduled a live demonstration for the Mobile World Congress (MWC) in Barcelona in 2026, where they will unveil the GPU partitioning technology alongside the latest enhancements to Orchestrator. This showcase is set to reveal the potential for scalable AI deployments across various sectors, illustrating how multi-purpose GPUs can effectively elevate resource efficiency while preserving performance quality.
While SoftBank’s stock took a minor hit following the announcement, analysts view this move as a strategic long-term investment. It highlights the company’s commitment to advancing AI infrastructure, positioning SoftBank favorably as the demand for AI compute resources continues to surge.
