0

At RCR Tech’s recent Intelligent RAN Forum (available on-demand here), operators converged on the need for measurable returns from AI investments, even as their approaches to network architecture, automation and future services diverge

AI promises to fundamentally transform the radio access network, from improving spectral efficiency and automating operations to enabling distributed computing platforms that support entirely new applications and business models. But for mobile network operators, the path from what’s technologically possible to what’s commercially practical is becoming increasingly pragmatic.

That was a recurring theme in conversations with executives from Orange, Optus, Rakuten Mobile, SK Telecom, Verizon, Turkcell and Deutsche Telekom. They described different approaches to AI-RAN, but demonstrated substantial alignment around the need for tangible operational and economic outcomes.

The emerging consensus was less about which AI-RAN architecture will prevail and more about sequencing investments in a manner that captures measurable efficiencies today, builds toward more autonomous operations, and makes significant new infrastructure commitments when demand justifies them.

 Improving the network operators already have

The most immediate AI-RAN opportunity is making existing networks perform better, operate more efficiently and consume fewer resources. Optus, working with Ericsson, detailed progress across AI-driven link adaptation, coverage prediction and automated coverage compensation. The Australian operator reported spectral efficiency improvements reaching 10% across trial sites and 25% under certain conditions.

Sriharan Amirthalingam, Optus chief technology officer for networks, described these as practical opportunities while the industry develops more ambitious AI capabilities. “Those are three examples where we are trying to collaborate and bring in the third category of AI into the RAN,” he said. “To me, those are bird-in-hand things we can do now while we work for a longer-term orchestrated AI engine.” 

Rakuten Mobile is pursuing similar economic outcomes through a different architectural foundation. Its fully virtualized, cloud-native Open RAN network has enabled rapid deployment of AI-driven energy optimization, with the operator reporting savings exceeding 20% and achieving TM Forum Level 4 autonomous network validation for the energy-efficiency use case. In a forum presentation, Sudhakar Pandey, Rakuten Mobile’s head of RAN, explained how its software-defined infrastructure supports incremental intelligence without requiring wholesale architectural reinvention.

The financial implications extend beyond operating expenses. As Turkcell associate R&D director Güneş Kesik explained, “better use of capacity of our existing infrastructure can delay capex investments.” Improved spectral efficiency can effectively create additional network capacity without incremental spectrum acquisition or site construction. AI-driven energy management reduces recurring costs, while automated troubleshooting and optimization can increase workforce productivity.

Taken together, these benefits suggest the near-term economic value of AI-RAN is largely focused on improving returns on billions of dollars already invested in network infrastructure.

Architectural flexibility meets operational discipline

While operators share these objectives, their architectural strategies reflect different starting points. Rakuten Mobile is extending intelligence across an existing cloud-native, disaggregated network. Orange, conversely, is evaluating how to introduce AI capabilities across established RAN infrastructure without introducing unnecessary complexity or operational risk.

Working with Nokia, Orange plans to test AI-RAN capabilities in its live network, evaluating performance, energy consumption and total cost of ownership before scaling deployment. Laurent Leboucher, Orange’s group CTO, emphasized the need to demonstrate sustained benefits under real-world operating conditions. “Capacity, customer experience and economics are part of the same value chain,” he said.

Orange is considering multiple deployment models, including upgrades to existing Nokia AirScale infrastructure, dedicated AI-RAN nodes and cloud-native implementations. The objective is not architectural uniformity but operational consistency across different environments. This reflects a broader lesson from previous network transformation cycles. Flexibility is valuable only when it translates into improved economics rather than increased integration burden.

SK Telecom raised similar concerns around interoperability, particularly as operators seek to scale AI capabilities across heterogeneous, multi-vendor networks. Proprietary implementations that perform well in controlled trials may prove difficult to operationalize across an entire network. Turkcell’s Kesik offered a useful standard for evaluating this tradeoff. “The complexity we remove through automation should be greater than the complexity we introduce with AI.”

For operators, the relevant question isn’t whether AI can improve RAN performance; clearly it can and is. The real hinge is whether those improvements justify the compute, integration, lifecycle management and operational costs necessary to sustain them.

From automation to autonomy

Beyond immediate optimization, operators are increasingly focused on the transition from deterministic automation to autonomous network operations.

Verizon’s approach illustrates the distinction. During a forum interview, Anil Guntupalli, Verizon senior vice president of technology and product development, described a future in which AI agents reason across network domains, exchange contextual information and coordinate responses to emerging problems.

“This is where the agentic world kicks in,” Guntupalli said, describing the need for dynamic, cross-domain action rather than isolated, scripted responses. But more sophisticated intelligence also raises questions about control, accountability and the relationship between operators and their technology suppliers.

Vendor-provided AI agents may optimize individual network functions, but operators possess the broader understanding of network topology, operational dependencies, customer requirements and business priorities necessary to coordinate their actions. “The intent and outcome is something Verizon wants to own,” Guntupalli said.

That position suggests an important evolution in operator differentiation. As AI capabilities become more widely available, competitive advantage may increasingly reside in the orchestration layer that governs them, rather than in individual optimization algorithms. Autonomy, moreover, does not necessarily mean removing humans from network operations. Both Verizon and Rakuten emphasized human-defined intent, guardrails and escalation processes as fundamental to deploying AI safely across critical infrastructure.

Monetization remains the unanswered question

If operators largely agree on the value of AI-for-RAN, the investment case for AI-and-RAN and AI-on-RAN remains considerably less certain.

The AI-RAN Alliance distinguishes network optimization from shared AI/RAN compute infrastructure and the delivery of new AI applications at the network edge. The latter two concepts could transform telecom infrastructure into a distributed AI platform, potentially opening new revenue opportunities beyond connectivity.

But those opportunities remain dependent on the emergence of applications that actually require the capabilities operators would be investing to provide. SK Telecom’s Dongwook Kim delivered perhaps the forum’s clearest warning. “The biggest risk [with] AI RAN is not technical failure, but repeating the mistakes [of] the early 5G era.”

His concern was that operators could again prioritize advanced network capabilities without sufficiently validating the applications and commercial demand required to monetize them.

Optus offered a more expansive view of what could eventually change that equation. Amirthalingam anticipates AI agents communicating with other agents, more uplink-intensive multimodal traffic, and distributed inference workloads that alter network traffic patterns and infrastructure requirements.

Those developments could create meaningful opportunities for mobile operators. But the relationship between new workloads, network requirements and operator revenue remains unresolved. Deutsche Telekom’s Gabriel Ionita brought the discussion back to fundamental commercial discipline. “We need a real customer problem.”

The message is not that operators should abandon longer-term AI infrastructure ambitions. Rather, investment in shared accelerated compute, edge inference and new services must be evaluated against identifiable demand, competitive alternatives and credible paths to profitability.

A disciplined transition to AI-native networks

The Intelligent RAN Forum revealed an industry moving beyond abstract promises of AI transformation toward increasingly specific operational and economic decisions. Operators differ in their architectures, technological maturity and appetite for experimentation. But their priorities are converging on extracting greater value from existing network assets, introducing intelligence without compromising reliability, preserving control over increasingly autonomous operations, and aligning future infrastructure investments with validated commercial demand.

The most consequential development is that AI-RAN is beginning to shift from a technology-led proposition to an operator-led investment discipline. That discipline will determine whether incremental gains in network performance and efficiency become the foundation for a fundamentally different network operating model, and ultimately whether AI-RAN succeeds in creating the new businesses that previous generations of wireless technology struggled to deliver.

Be sure to check out the full Intelligent RAN Forum.