SoftBank Corp unveiled its Telco AI Cloud platform at MWC26. The aforementioned platform is an integrated architecture designed to transform telecom networks into AI-native infrastructures.
By unifying distributed AI data centres with the vendor’s Aitras AI-RAN orchestrator and Infrinia AI Cloud OS, SoftBank’s goal is to evolve from a traditional mobile network operator into a scalable AI infrastructure provider.
Ryuji Wakikawa, VP and head of the company’s Research Institute of Advanced Technology, explained in a briefing Softbank’s Telco AI Cloud infrastructure integrates large data centres across Japan with distributed GPU resources deployed nationwide, including Aitrus at the far edge and “regional brain” clusters with smaller GPUs per prefecture.
Wakikawa explained this architecture supports both AI model training and inference. It addresses latency-sensitive applications by placing GPUs closer to users while maintaining large-scale centralised training capabilities in gigawatt data centres.
SoftBank teamed with Ericsson to highlight one of the use cases of the Telco AI Cloud platform. Using AI-on-RAN, the two companies demonstrated how robots with limited onboard GPU capacity can offload heavier AI models to mobile edge GPUs while operating in complex environments like shopping centres.
The two companies validated a low-latency, high-reliability AI-RAN architecture which combines dynamic AI processing offloading with network slicing, establishing a foundation capable of supporting scalable physical AI applications.
For enterprises, Mitsubishi Heavy Industries and SoftBank conducted a field trial of an edge AI application using Aitras in an on-premises environment.
SoftBank’s’ Aitras for Biz platform is designed for commercial enterprise use, especially in factories. It leverages stable 5G connectivity and nearby GPUs to deploy AI applications.
“This is a dedicated product for enterprise segment. We’re going to run multiple applications on here to showcase how powerful AI RAN is for enterprise use.”
Open source
Softbank also announced its Aitras orchestrator is now available through open source. By open-sourcing the dynamic scoring framework (DSF) of its orchestrator, SoftBank enables RAN vendors and MNOs to implement AI-RAN orchestration more easily, accelerating commercial deployments and expanding the ecosystem through OSS collaboration.
DSF is open sourced under the Linux Foundation’s CNCF programme to promote interoperability and multi-vendor support.
Once GPUs are allocated for radio use, vendors like Ericsson and Nokia can deploy and manage radio units via the service management orchestrator (SMO) for improved radio resource management.
Using the Aitras orchestrator, SoftBank has partnered with Nokia to enable execution of external AI workloads on AI-RAN to create new revenue opportunities across communications infrastructure.
Ericsson and SoftBank also announced at MWC26 they achieved interworking between the Aitras orchestrator and an open RAN compliant SMO.
“After I trust the orchestrator to give the resource to the vendor, the vendor can start deploying the radio on that resource through the SMO. This kind of interaction has already been confirmed with these two vendors and we’re very excited to have these kinds of innovations”, Wakikawa added.
Source: Mobile World Live
Image Credit: Stock Image