The telecommunications industry stands at a pivotal junction. As 5G matures and the industry looks toward the horizon of 6G, the traditional Radio Access Network (RAN)—once a rigid, hardware-heavy fortress—is undergoing a radical metamorphosis. At the heart of this shift is the promise of "AI-native" RAN, a concept championed by Nokia and currently being stress-tested in the real-world environments of Orange.
While the technical promises are staggering—Nokia projects spectral efficiency gains of 20% today, scaling to over 100% by 2028—the transition from laboratory theory to live network reality is fraught with operational complexities.
Main Facts: The Convergence of Connectivity and Compute
The Intelligent RAN Forum recently served as the backdrop for a critical discussion between Nokia’s Chief Technology and AI Officer, Pallavi Mahajan, and Orange’s EVP of Networks and Group CTO, Laurent Leboucher. The central theme was clear: Nokia is attempting to redefine the RAN as a programmable platform, shifting away from the generational, hardware-centric cycles that have defined mobile networks for decades.
By decoupling software from proprietary hardware and integrating Nvidia’s accelerated compute platforms, Nokia is enabling machine learning algorithms to operate directly within the baseband. This allows for continuous, real-time optimization of radio resources. For operators like Orange, this represents a fundamental shift in business logic. The goal is no longer just "more bandwidth," but "smarter capacity" that can be managed through subscription-based software updates rather than costly, time-consuming hardware refreshes.
Chronology: From Proprietary Silos to the Software-Defined Era
The evolution of the RAN has followed a predictable, yet rigid, path from 2G to 5G. Historically, every generational leap required massive capital expenditure (CAPEX) to deploy custom silicon and proprietary hardware.
- The Legacy Era: For decades, RAN architecture was a "fixed stack," where hardware and software were tightly coupled. Upgrades were slow, and innovation cycles were tied to the decade-long rollout of new radio standards.
- The Disaggregation Phase (2020–2024): With the rise of Open RAN and cloud-native principles, the industry began separating the control plane from the data plane. Nokia’s
anyRANstrategy marks the next logical step in this journey, treating the network as a cloud-like environment. - The AI-Native Present: Nokia has introduced a three-tiered deployment model to facilitate this transition:
- Brownfield Upgrades: New capacity cards compatible with existing AirScale chassis.
- Cloud AI-RAN: Deployment on commercial off-the-shelf (COTS) servers.
- Standalone Nodes: Dedicated AI-accelerated units for ultra-high capacity environments.
- The Live Testing Phase (Late 2024 and beyond): Orange has now committed to moving these models out of the lab and into live, high-traffic network environments to validate performance, energy efficiency, and total cost of ownership (TCO).
Supporting Data: The Metrics of Efficiency
Nokia’s aggressive performance projections are the cornerstone of their market pitch. The numbers are not merely incremental; they are transformational:
- 20% Spectral Efficiency: Already demonstrated in testing environments.
- 50% Spectral Efficiency: Targeted for achievement by mid-2027.
- 100%+ Spectral Efficiency: The long-term benchmark set for 2028.
However, these figures carry a caveat. As Laurent Leboucher of Orange pointed out, the real-world value of these percentages depends entirely on their consistency. A gain that exists in a controlled lab environment but falters during peak traffic or in challenging radio conditions is of little use to an operator maintaining "five-nines" (99.999%) reliability.
Nokia’s approach addresses this by utilizing "glass-box" principles—ensuring that all AI-driven decisions are traceable, explainable, and, crucially, reversible. Should an AI model behave unexpectedly, the system is designed to roll back to conventional, deterministic algorithms, ensuring that innovation never comes at the cost of network stability.
Official Responses: The Operator’s Prudence
The dialogue between Nokia’s Mahajan and Orange’s Leboucher reflects a healthy tension between innovation and operational reality.
Nokia’s Perspective:
Pallavi Mahajan emphasized the concept of "fallow compute"—the idea that modern RAN hardware often sits with untapped potential that can be repurposed for AI workloads. By running AI at the edge (the cell tower), Nokia aims to minimize latency. "Not every workload has to be running on a cell tower," Mahajan noted, but for the latency-sensitive tasks of 6G, the intelligence must move closer to the radio edge.
Orange’s Perspective:
Laurent Leboucher, while supportive of the technological vision, remains focused on the "learning journey." For Orange, the priority is not a wholesale rip-and-replace strategy, but a pragmatic, site-by-site evolution. "The real test is whether these gains can be delivered consistently in a live network," Leboucher stated. He emphasized that for Orange, the primary goal is to avoid operational fragmentation, ensuring that whichever deployment model is chosen—be it brownfield, cloud, or standalone—it fits into a common software foundation with unified lifecycle management.
Implications: A New Operating Model for Telcos
The transition to an AI-native RAN carries profound implications for the future of telecommunications, extending far beyond simple capacity gains.
1. The Shift to Autonomous Operations
As the network becomes more autonomous, the role of network operations staff is undergoing a radical change. Engineers are transitioning from manual configuration tasks to "intent-based" management. They will act as supervisors of an autonomous system, setting high-level goals and monitoring the AI’s adherence to security and performance guardrails.
2. Monetizing the Network as a Platform
One of the most significant implications is the opening of the RAN to third-party applications. By using the E3 interface, operators can host Distributed Apps (DApps) that perform tasks previously impossible or too slow to handle in the central cloud. Leboucher envisions a future where the network provides more than just data—it provides services.
- Integrated Sensing: The network could detect drones or monitor industrial safety using radio waves.
- Precise Geolocation: Enabling location-based services that are more accurate than GPS in urban canyons.
- Radio Optimization: Replacing legacy SON (Self-Organizing Network) platforms with software-defined, AI-driven management that adapts to traffic patterns in real-time.
3. The End of "Generational" Upgrades
Perhaps the most disruptive implication is the move toward a continuous software-defined roadmap. If Nokia’s claim holds—that current AI-RAN deployments provide a viable path to 6G through software updates alone—then the massive infrastructure investment cycles of the past may be coming to an end. Operators will no longer be forced to tear down and rebuild their RAN every ten years to support the next generation of standards. Instead, they will subscribe to a continuously evolving platform.
4. Economic Viability and TCO
Ultimately, the success of AI-RAN rests on the balance sheet. Leboucher’s focus on TCO is indicative of a broader industry trend. For AI-RAN to succeed, the energy savings and the reduction in manual labor must outweigh the costs of deploying high-performance compute nodes. The move toward "fallow compute" is a strategic play to maximize the utility of existing assets, making the financial case for AI integration much more compelling.
Conclusion
The partnership between Nokia and Orange represents a critical litmus test for the telecommunications sector. By moving the AI-RAN debate from the forum floor to the live network, these two entities are attempting to solve the "last mile" of network intelligence.
If the gains in spectral efficiency can be maintained in the chaotic, high-traffic reality of a major European carrier, the industry will have successfully navigated one of its most difficult transitions. The RAN is no longer just a conduit for bits and bytes; it is evolving into a sophisticated, programmable, and autonomous engine for digital innovation. The question is no longer whether AI has a place in the RAN, but how quickly operators can master the complexity of managing it at scale.
