AI-RAN delivers operational ROI for telcos by moving beyond theoretical concepts into practical energy and spectral efficiency gains. As operators face flatlining revenues and spiralling complexity, the focus has shifted from technological novelty to bottom-line impact through energy savings and automation.

This shift links directly to Open RAN (O-RAN) principles, specifically openness and disaggregation. Industry sentiment backs this coupling. A Heavy Reading survey indicates that 30 percent of operators believe AI-RAN and O-RAN will be “very closely linked” in the next five years. Among those deploying O-RAN by 2025, that figure rises to 47 percent.

For telecoms execs, however, the priority remains immediate value generation rather than long-term architectural theory.

Energy optimisation currently offers the most tangible business case for AI-RAN. Radio networks consume vast amounts of power, and managing this without degrading user experience remains a primary challenge.

Ericsson reports that 10 to 15 percent power savings have been demonstrated in multiple trials and similar quantitative gains in live networks. Using an “AI native link adaptation” feature tested with Bell Canada, the trial demonstrated up to a 20 percent increase in downlink throughput and a 10 percent increase in spectral efficiency compared to baselines. For network operators, getting more capacity out of existing spectrum directly improves capital efficiency.

Infrastructure strategy: integration over isolation

Building purpose-built infrastructure for AI workloads separate from the standard network stack is a strategic misstep. The consensus among technical architects is to run AI on the cloud-native application platforms operators already use for microservices.

Fatih Nar, Distinguished Chief Architect in the CTO Office at Red Hat, warns against building “another platform yet for AI only”. Instead, operators should enrich their current platforms with AI capabilities, allowing workloads to move between on-premise and hybrid cloud environments without fragmentation.

This logic extends to hardware. Kanika Atri, Senior Director of Telecoms Marketing at NVIDIA, addresses persistent concerns regarding the cost and power of accelerated computing. She argues the industry must move past the assumption that GPUs are inherently too expensive for the edge.

“We would not be doing this if we cannot meet the baseline stringent requirements for telcos radio rollouts,” Atri states.

Solutions must meet the site’s cost and power envelope as a starting point. Atri cites the ‘ArcPro’ platform, which fits a “tiny form factor” under 300 watts while running 5G capability sets, as proof that accelerated computing can align with telco constraints.

Toward intent-based automation

Beyond efficiency, AI-RAN enables a shift toward “intents” (i.e. allowing operators to define what the network should achieve rather than how to configure it.)

In scenarios like a major concert requiring immediate high-throughput slices, intent-based systems allow the network to configure itself rapidly, reducing manual engineering burdens.

However, executives should essentially temper expectations regarding full autonomy. The transition is gradual. Nar cautions that even hyperscalers like AWS and Google have not achieved fully autonomous operations despite their resources.

“You think more complicated access networks such as radio… [will] fully be autonomous? Not likely,” Nar warns, suggesting that humans will remain in the loop to manage physical complexities like weather and interference.

The human capital challenge of implementing AI-RAN

Implementing AI-RAN requires organisational adjustment. A frequent error is segregating AI talent from network engineering teams.

“Please do not segregate your AI and data scientists from the domain expert,” Nar advises. He suggests avoiding a detached “Centre of Excellence” in favour of a “Community of Practice” that embeds AI knowledge within radio access network operations teams.

Domain expertise is non-negotiable because telco AI models are quite different from the Large Language Models (LLMs) dominating headlines. They require an understanding of the “language of the wireless environments,” such as RF propagation, which generalist AI practitioners often lack.

The path forward involves pragmatic steps rather than wholesale infrastructure replacement. Operators and enterprises should:

• Leverage O-RAN principles: Even if not fully deploying O-RAN, aligning architecture with its principles creates the necessary data portability for AI integration.

• Start with energy: Focus initial pilots on energy management and sleep modes, where ROI is quantifiable and verifiable.

• Unify platforms: Reject separate AI silos. Ensure AI workloads run alongside existing network functions on a common cloud-native platform.

• Integrate teams: Embed data scientists within network engineering units to ensure models reflect physical network realities.

As the industry moves toward 6G, the foundational work done today in AI-RAN – establishing data pipelines, refining models, and proving commercial viability – will determine which operators successfully transition to high-performance, programmable networks.

See also: 6G networks will host AI agents to automate enterprise workflows

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events. Click here for more information.

Telecoms is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.