Japan’s Rakuten Mobile has expanded its partnership with Intel to embed Artificial Intelligence (AI) into virtualized radio access networks (vRAN), signaling a broader industry shift toward AI-first mobile infrastructure aimed at improving performance, automation and energy efficiency.
Rakuten Mobile said it will work with Intel to develop and deploy AI-native capabilities across the RAN stack, moving beyond basic virtualization toward networks that can dynamically optimize themselves in real time. The initiative builds on the companies’ existing vRAN collaboration based on Intel Xeon processors and FlexRAN software and is positioned as a step toward autonomous and self-optimizing mobile networks.
The companies said their joint work will focus on four technical priorities: improving spectral efficiency, automating RAN operations, optimizing resource allocation and reducing energy consumption. AI models will be integrated directly into Layer 1 and Layer 2 RAN functions as well as network operations and management platforms, with an emphasis on meeting carrier-grade latency and reliability requirements.
Rakuten Mobile, widely seen as one of the most aggressive adopters of cloud-native and Open RAN architectures, is using the collaboration to push its software-centric network model further. By embedding AI workloads into the vRAN software stack rather than treating AI as an external analytics layer, the operator aims to enable real-time decisioning inside the RAN itself.
Sharad Sriwastawa, co-CEO and CTO of Rakuten Mobile, said the partnership is focused on validating “transformative AI-driven innovations” and demonstrating how AI can be integrated efficiently into existing vRAN stacks to improve both performance and operational efficiency.
From Intel’s side, the partnership is also strategically important as chipmakers compete to position their processors as the foundation for AI-enabled telecom infrastructure. Intel said the joint effort uses its vRAN AI Development Kit, FlexRAN reference software and AI libraries running on Xeon 6 system-on-chip platforms, which include built-in acceleration features such as AVX512/VNNI and AMX designed to support low-latency AI inference.
Kevork Kechichian, executive vice president and general manager of Intel’s Data Center Group, said the companies are demonstrating how AI benefits can be realized in production vRAN environments and noted that Intel processors already power the majority of commercial vRAN deployments globally.
Industry context: AI moves into the RAN layer
The collaboration reflects a growing industry trend: AI is moving from network analytics and planning tools into real-time network control loops. Operators and vendors are increasingly exploring AI-driven radio optimization, traffic prediction, interference management and automated healing — all functions that traditionally relied on static rules and manual tuning.
AI-native RAN could become especially important as networks grow more complex with 5G standalone cores, network slicing, private networks and, eventually, early 6G features. Embedding AI directly into the RAN layer allows faster response times than cloud-only AI control systems and may help operators manage rising traffic loads without proportional increases in operating cost.
Energy efficiency is another driver. RAN remains the most power-hungry part of mobile networks, and AI-based optimization of radio resources and hardware utilization is emerging as a key lever for reducing energy per bit — a metric increasingly scrutinized by both regulators and investors.
For Rakuten Mobile, the move also reinforces its positioning as a testbed for next-generation, software-driven telecom architectures that can later be exported through its global platform and partner ecosystem. For Intel, it strengthens its role in the telecom AI infrastructure stack at a time when operators are reassessing silicon and platform choices for AI-era networks.
The companies said testing and validation are already underway, with the goal of enabling software-upgradable, AI-powered RAN capabilities on open, general-purpose compute platforms.
