Rakuten Symphony Achieves TM Forum Level 4 Autonomy in AI-Native Open RAN Energy Optimization

Introduction: Autonomy at the Edge of the RAN

As Open RAN architectures mature beyond disaggregation and multi-vendor interoperability, the focus has shifted to embedding intelligence directly into the radio access network. On its recently updated AI-RAN platform page, Rakuten Symphony announced a world-first achievement: TM Forum Level 4 autonomy certification for RAN energy efficiency optimization in an Open RAN environment. This milestone, validated through live deployments at Rakuten Mobile serving over 10 million subscribers, represents a significant step toward truly autonomous networks that operate with minimal human intervention.

Level 4 autonomy, in the TM Forum Autonomous Networks framework, implies closed-loop operations where the system detects issues, decides on actions, and executes them with high confidence, requiring only occasional human oversight for exceptions. For energy optimization specifically, this means AI models that predict traffic lulls, dynamically adjust power states across radios and baseband units, and maintain service-level agreements without manual tuning. In an Open RAN context, this capability rides on standardized interfaces, making the intelligence portable across vendors.

The Role of the RAN Intelligent Controller (RIC) in Delivering Autonomy

At the heart of Rakuten Symphony’s implementation sits the RAN Intelligent Controller (RIC). The company claims the industry’s first nationwide deployment of third-party xApps and rApps fully aligned with O-RAN Alliance standards. Unlike proprietary controllers, the RIC here functions as an open platform for intent-based policy enforcement.

In practice, the Non-RT RIC handles longer-term policy creation and model training (granularity >1 second), while the Near-RT RIC executes real-time closed-loop automation on timescales between 10 ms and 1 second. This separation enables granular adjustments such as dynamic beamforming, load balancing, and—critically for this certification—energy state management. xApps running on the Near-RT RIC can subscribe to standardized E2 interface data (KPIs, measurements) and issue control actions back to the RAN nodes via the same interface.

Rakuten Symphony’s deployment demonstrates how third-party applications can coexist on a shared RIC without vendor lock-in. Operators gain the ability to mix energy-saving rApps from one vendor with traffic-prediction xApps from another, all while maintaining O-RAN compliance. This multi-app ecosystem is what differentiates a true AI-native Open RAN from siloed vendor solutions.

Technical Integration: Leveraging Existing vRAN Investments

A key differentiator highlighted in the platform documentation is the avoidance of rip-and-replace strategies. Through validated integration with Intel, AI workloads are layered onto existing virtualized RAN (vRAN) software stacks. This approach preserves capital already deployed in CPU-based servers and avoids the immediate need for GPU-heavy infrastructure.

The architecture supports O-RAN Split 7.2 functional splits, where the Distributed Unit (DU) handles lower-layer processing. AI models for energy optimization run as containerized workloads alongside the DU/CU functions, consuming telemetry from the RIC and feeding decisions back through standardized interfaces. Performance claims include more reliable uplink detection (reducing packet loss), fewer retransmissions, and measurable spectral efficiency gains through real-time parameter tuning.

For operators concerned about hardware refresh cycles, this Intel-validated path means AI-RAN capabilities can be introduced via software upgrades on current-generation servers. The platform emphasizes cloud-native design: fully virtualized, container-orchestrated, and designed for continuous integration of new AI models without disrupting live traffic.

Energy Optimization Use Case: From Prediction to Closed-Loop Action

The certified energy efficiency application exemplifies the closed-loop workflow. AI models ingest historical and real-time traffic patterns, predict periods of low utilization, and proactively transition radio units and baseband processors into lower-power states. Because the optimization occurs within the RIC framework, decisions respect O-RAN policies for handover, slicing, and QoS.

Rakuten Mobile’s live network provided the proving ground. The system achieved Level 4 certification by demonstrating autonomous operation across thousands of sites, with energy savings realized without measurable impact on user experience metrics. Complementary capabilities include predictive capacity planning—where models forecast demand spikes and pre-allocate resources—and autonomous site management using computer vision to audit construction and maintenance activities remotely.

These use cases illustrate how AI-RAN moves beyond reactive optimization to proactive, intent-driven management. An operator can express high-level intents (“minimize energy while maintaining 99.9% coverage”) and have the RIC translate them into enforceable policies executed at the edge.

Deployment Velocity and Operational Impact

Beyond performance metrics, the platform reports up to 40% faster large-scale rollouts. AI-powered computer vision automates site audits, comparing as-built conditions against design specifications and flagging deviations without requiring field engineers to visit every location. Combined with the RIC’s ability to push consistent configurations across multi-vendor environments, this reduces both time-to-market and ongoing OPEX.

For a technical audience, the implication is clear: Open RAN’s disaggregated nature, when paired with a standards-aligned RIC, creates a programmable substrate for AI. Instead of waiting for monolithic vendor roadmaps, operators can deploy targeted intelligence apps and iterate rapidly.

Broader Implications for AI-Native Networks

Rakuten Symphony’s achievement underscores a maturing narrative in the industry. While GPU-accelerated baseband processing garners headlines for Layer 1 acceleration, many near-term AI-RAN gains come from higher-layer intelligence running on existing infrastructure. The Level 4 energy certification proves that meaningful autonomy is achievable today in production Open RAN networks.

Looking ahead, the same RIC platform that delivers energy optimization can host models for spectral efficiency, mobility management, and even emerging agentic AI workflows. Because the architecture is software-upgradable and standards-based, new capabilities—such as integration with generative AI for operational insights—can be added without hardware changes.

Operators evaluating AI-RAN strategies should note the emphasis on leveraging current vRAN investments. This pragmatic path lowers the barrier to entry and accelerates the shift from manual operations to autonomous networks.

Conclusion

Rakuten Symphony’s TM Forum Level 4 certification for Open RAN energy optimization marks a concrete milestone in the evolution of AI-native networks. By combining a standards-compliant RIC, third-party app ecosystem, and seamless integration with existing vRAN stacks, the company demonstrates that high levels of autonomy are not theoretical but operational at national scale. As more operators adopt similar architectures, the industry moves closer to networks that self-optimize in real time, reduce energy footprints, and free engineering resources for higher-value activities.

Sources

• https://symphony.rakuten.com/ai-ran