In 2025, telcos need to continue to concentrate on efforts to maximize the value of existing infrastructure while balancing new revenue-generating opportunities through service innovation, dynamic pricing, and strategic partnerships with the likes of hyperscalers. As we have discussed in a long series of 5G blogs , Network monetization is also critical to Telco’s growth strategy, which is implemented by exploiting infrastructure and technology to create new revenue streams. This blog is based on an Nvidia/SoftBank partnership that seeks to unlock AI at the 5G RAN Edge.
https://developer.nvidia.com/blog/ai-ran-goes-live-and-unlocks-a-new-ai-opportunity-for-telcos/
Edge Monetization, the promised land?
Telcos have a unique opportunity to build on their edge infrastructure and workload management expertise as discussed at
https://www.vamsitalkstech.com/5g/accentures-maturity-model-of-edge-computing/
Telco’s extensive footprint across geographies gives them a unique opportunity to deliver enterprise-grade edge computing with a wide range of consumption-based pricing options. This infrastructure advantage allows businesses to process the data near its source, reducing latency radically and improving operational efficiency.
However, given the somewhat lukewarm monetization of edge computing in telecommunications, success in implementing AI-RAN may potentially help make or accelerate a critical second-mover advantage for telcos to be in the conversation for the AI economy.
Is AI-RAN the Telcos’ next chance at monetizing the Edge?
NVIDIA and SoftBank’s successful AI-RAN implementation, as highlighted in the blog, seems to offer a vital second opportunity following the missed edge computing wave of the past decade. While telcos failed to capitalize on edge computing due to business strategy limitations rather than technical constraints, the AI-RAN breakthrough presents a compelling case for transformation. This timing is particularly significant as AI infrastructure investment soars, with tech giants like Microsoft and Meta committing $333 billion to AI data centers, while 97% of telcos are investing in AI, though only a select few have successfully monetized these investments.
Telcos’ existing infrastructure presents a natural advantage for AI-RAN implementation, with distributed compute resources across central offices positioning them ideally for low-latency AI inferencing. The practical applications already demonstrated include autonomous vehicle support, factory monitoring with multi-modal AI integration, and robotics applications with edge AI inferencing. These use cases illustrate the versatility and immediate market potential of AI-RAN technology.
The technical POC (proof of concept) demonstrated in Fujisawa City, Japan, showcases the impressive integration of NVIDIA’s cutting-edge hardware, including the GH200 Grace Hopper Superchip, Bluefield-3 networking, and Spectrum-X fronthaul/backhaul connectivity. The system’s performance metrics are equally impressive, processing 20 5G cells per GH200 server with peak downlink speeds of 1.3 Gbps in ideal conditions and 816Mbps in real-world deployment. Perhaps most significantly, the solution increases infrastructure utilization from 33% to nearly 100%, marking a fundamental shift in network efficiency.
The economic implications of this transformation are substantial, with implementations showing 5x revenue generation per CapEx dollar and a 219% ROI over a five-year period in AI-heavy scenarios. The system also delivers significant operational benefits, including 40% power reduction compared to traditional RAN systems and 60% power savings versus x86-based vRAN implementations. These efficiency gains translate directly to improved operational economics and environmental sustainability.
Technical Architecture: NVIDIA and SoftBank’s AI-RAN Implementation
Technical Architecture:
The architecture diagram illustrates SoftBank’s AI-RAN implementation using NVIDIA technology.
The architecture diagram illustrates SoftBank’s AI-RAN implementation using NVIDIA technology. Here’s a detailed description:
- Hardware Foundation: The system is built on NVIDIA GH200 hardware, which serves as the base for SoftBank’s AI-RAN solution.
- Workload Distribution:
The system manages three types of workloads:- RAN Workloads: Traditional radio access network functions.
- Internal AI Workloads: AI tasks managed within the SoftBank network.
- External AI Workloads: AI tasks from external sources.
- SoftBank E2E Orchestrator:
- This layer manages the overall system, coordinating between different components and workloads.
- It integrates with the Wireless Network and External Cloud.
- The orchestrator incorporates NVIDIA AI Enterprise Serverless API for AI task management.
- AI Supply and Demand: The system balances AI supply (available computing resources) with AI demand (incoming AI workloads).
- External Integration:
- NVIDIA AI Enterprise Serverless API connects the system to external AI applications and NVIDIA NIM (NVIDIA Inference Microservices).
- It supports various AI models including Llama 3, NVIDIA NIM, and Mistral AI.
- Network Connectivity: The architecture connects to the Wireless Network for RAN functions and to External Cloud for additional resources and services.
- AI Application Support: The system can run various AI applications, leveraging NVIDIA NIM and supporting models like Llama 3 and Mistral AI.
Hardware Implementation:
- NVIDIA GH200 Grace Hopper Superchip integration
- NVIDIA Bluefield-3 networking
- Spectrum-X fronthaul/backhaul connectivity
- 20 radio units with 5G core network integration
Performance Metrics:
- 20 5G cells per GH200 server (100-MHz bandwidth)
- 1.3 Gbps peak downlink (ideal conditions)
- 816Mbps in real-world deployment
- Utilization increase from 33% to nearly 100%
Economic Impact:
- 5x revenue generation per CapEx dollar
- 219% ROI over 5-year period (AI-heavy scenarios)
- 40% power reduction vs. traditional RAN
- 60% power savings vs. x86-based vRAN
Future Outlook:
So the future outlook includes NVIDIA’s next-gen capabilities roadmap on the below:
- 2x AI-RAN compute capacity
- 5x Llama-3 inferencing improvement
- 18x data processing enhancement
- 9x vector database search improvement
Conclusion
This convergence of AI and RAN technology represents more than just technical advancement—it’s a strategic inflection point for the telecommunications industry. The success of SoftBank’s implementation provides a blueprint for other telcos to avoid the mistakes of the edge computing era and successfully position themselves in the AI economy. The key question remains: Will the telecommunications industry learn from its edge computing experience and successfully capitalize on this AI inferencing opportunity, or will it once again be relegated to providing basic connectivity services?
Featured Image by aopsan on Freepik