The telecommunications industry is currently gripped by a singular, transformative narrative: the integration of Artificial Intelligence into the Radio Access Network (AI-RAN). Proponents argue that by infusing AI into the backbone of mobile connectivity, operators can achieve unprecedented levels of spectral efficiency, automated optimization, and energy conservation. Yet, as the industry gathers at forums and conferences, a pragmatic, sobering question persists: Do the performance gains of today actually justify the staggering capital expenditure required to implement these technologies?
Rob Hughes, head of wireless marketing at 1Finity—a Fujitsu company—has emerged as a leading voice of caution, urging Mobile Network Operators (MNOs) to abandon the "wait and see" approach. His argument is not that the technology is failing, but that the sequencing is flawed. Rather than betting the house on speculative, long-term performance metrics, Hughes contends that operators should pivot toward "opportunity-driven" deployments—projects where an immediate, tangible business case and a paying customer already exist.
The Business Case Gap: Efficiency vs. Revenue
The AI-RAN Alliance, a cross-industry body, distinguishes between two fundamental modes of operation. The first, "AI for RAN," focuses on using machine learning to optimize the network’s internal mechanics—such as boosting cell-edge uplink throughput or dynamically adjusting power consumption. The second, "AI RAN," reimagines the RAN infrastructure as a multi-purpose platform that serves AI workloads for external enterprises, alongside its traditional telecommunications duties.
While 1Finity has spent years collaborating with ecosystem partners to refine RAN optimization, the results have been a mixed bag. "We have seen substantial improvements," Hughes concedes. "But operators are struggling to justify the investments based on the performance improvements that are available today."
The performance gains are real, but they are not yet large enough to carry the weight of a multi-billion dollar capital expenditure budget. The structural reality of telecommunications infrastructure provides the missing link. RAN networks are designed to handle peak capacity, meaning the compute power residing at the edge is significantly underutilized during off-peak hours. This represents a massive pool of "wasted headroom."
Hughes argues that by pooling and reselling this excess capacity, MNOs can generate a new revenue stream from infrastructure they are already powering and maintaining. The strategy is not to deploy GPUs everywhere on day one, but to deploy them where demand justifies the spend.
Orchestration, Not Hope: Managing Shared Infrastructure
If MNOs intend to run third-party enterprise AI workloads alongside their core network traffic, they face a critical technical hurdle: how to prevent enterprise applications from degrading the quality of service for the network’s primary users.
Hughes rejects the "hope-based" approach of simply dumping all workloads into a shared server pool. Instead, he advocates for policy-driven orchestration that treats the RAN as the priority workload at all times. This requires a robust system of continuous resource monitoring, reservation, and prioritization. The implementation is complex; it involves partitioning tenants, tracking consumption, and maintaining granular billing systems.
"You need both options to soak up the excess compute," Hughes notes. "Some clients will pay for dedicated, high-priority capacity, while others will opt for best-effort or off-peak access."
The Strategic Edge: Why MNOs Can Out-Compete Hyperscalers
The most provocative element of the 1Finity strategy is its direct challenge to the supremacy of hyperscalers like AWS, Google Cloud, and Microsoft Azure. While hyperscalers have dominated the AI narrative, Hughes believes MNOs hold a set of "home-court advantages" that tech giants simply cannot replicate:
- Last-Mile Connectivity: MNOs own the transport and the final connection to the device.
- Spectrum and Wireless: Exclusive control over the wireless medium is a barrier to entry that cloud providers lack.
- Mobility: MNOs manage the handoffs and the mobile nature of sensors and devices.
- Data Sovereignty: In an era of strict regional data regulations, MNOs provide a local, physical landing spot for data that cannot cross borders.
- Distributed Edge: MNOs possess thousands of existing real estate locations—towers and central offices—with established power and cooling, which are far more geographically granular than the centralized, regional cloud data centers used by hyperscalers.
"The hyperscalers can provide the AI services," Hughes asserts, "but MNOs can combine the AI services with the connectivity, the mobility, the spectrum, and the distributed edge." By bundling AI services with connectivity, operators can increase customer "stickiness," creating a service package that is fundamentally more valuable than raw cloud compute.
Overcoming the "Customization Trap"
A critical lesson from the failed promise of early Mobile Edge Computing (MEC) initiatives is that customization is the enemy of scale. Previous efforts required bespoke integration for every client, making deployments prohibitively expensive and time-consuming.
To avoid this, 1Finity has introduced a two-tiered approach. The first is "GPU-as-a-service," a straightforward, low-touch model that provides raw compute capacity. The second involves a partnership with Aible, utilizing a library of over 100 standard application agents. These agents are configurable through guided, no-code workflows, shifting the burden of customization from the operator to the end-user.
In a recent demonstration, 1Finity showcased a shoplifting detection system for a retail chain. The setup involved cameras connected to a Customer Premises Equipment (CPE) unit, linked to an Open RAN-compliant radio, and feeding into a cluster of NVIDIA GPUs. The servers managed the RAN stack, the AI application, and an additional "AI-as-a-service" workload simultaneously. The demonstration was purposefully vendor-neutral, utilizing Quanta servers to prove that the architecture can operate independently of the hardware provider.
Orchestration: The Secret Sauce
Underpinning these deployments is a sophisticated resource management layer provided by partners like Armada. Unlike standard cloud management, which only monitors GPU temperature or load, this orchestration layer reaches directly into the RAN’s Centralized Unit (CU) and Distributed Unit (DU).
It pulls specific Key Performance Indicators (KPIs), such as active User Equipment (UE) counts and throughput per UE. This "RAN-aware" monitoring allows the policy engine to anticipate when a new AI workload might impact a mobile subscriber’s experience, enabling the system to throttle or shift resources in real-time.
Implications for the Future of 5G
The industry is standing at a crossroads. The promise of "AI-native" networks is often framed as a distant, massive infrastructure upgrade that requires global standards and a complete overhaul of current hardware. Hughes’s perspective offers a more grounded, immediate alternative.
By focusing on "mass customization" and leveraging the infrastructure that is already in place, operators can build operational expertise today. This creates a bridge to the future: as the demand for AI grows and the technology matures, operators will already have the physical and software frameworks in place to host fully autonomous, AI-native networks.
For the mobile operator, the takeaway is clear: stop waiting for the perfect, standardized AI-RAN ecosystem to arrive. Instead, look for the first enterprise customer willing to pay for a localized AI solution, and use that as the foundation to transform the network from a simple connectivity pipe into a multi-faceted platform for the intelligent edge.
As Hughes concludes, the business case for AI-RAN does not require a nationwide build-out to be validated. It simply requires a clear-eyed understanding of the existing infrastructure and a single, paying customer to prove the model works. In the volatile world of telecommunications, that is as close to a sure bet as an operator can get.
