Nokia’s AI-RAN Trials Go Global: The Long Road from Lab to Live
In the annals of telecommunications, few technological shifts have promised as much and delivered as slowly as the convergence of artificial intelligence and radio access networks. For years, AI-RAN has been the industry’s favorite parlor trick—impressive in controlled demonstrations, tantalizing in vendor roadmaps, but perpetually just out of reach for commercial deployment. That narrative is now being rewritten. Nokia’s announcement this week that eight global operators are advancing AI-RAN trials using the NVIDIA Aerial RAN Computer represents more than just another press release; it signals a fundamental inflection point in how the industry moves from theoretical promise to practical reality.
The shift from early evaluation to live field trials by operators including A1 Group, Chunghwa Telecom, du, e&, Mobily, stc, TPG Telecom, and Zain Saudi is not merely a milestone—it is a validation of a decade-long journey that began with the first tentative experiments in software-defined networking and has now arrived at the doorstep of commercial viability. To understand why this moment matters, one must trace the arc of how we got here.
The Seeds of Intelligence: From 4G Optimization to AI-Native Ambitions
The story of AI in radio access networks begins not with grand visions of autonomous networks, but with the humble pursuit of efficiency. In the early 2010s, as 4G LTE deployments matured, operators faced a persistent challenge: how to squeeze more capacity from finite spectrum while managing escalating energy costs. The first forays into AI were modest—machine learning algorithms applied to traffic prediction and load balancing, often running in centralized data centers and feeding recommendations to human engineers.
These early systems were, by today’s standards, primitive. They could forecast traffic patterns with reasonable accuracy and suggest parameter adjustments, but they lacked the speed and autonomy to act in real time. The latency between data collection, analysis, and action was measured in minutes or hours, far too slow for the millisecond-scale decisions required in a live radio network.
The industry’s thinking began to shift with the advent of 5G and the parallel rise of Open RAN. The disaggregation of the radio access network into functional components—centralized units (CU), distributed units (DU), and radio units (RU)—created new interfaces and new opportunities. If the RAN could be disaggregated, it could also be virtualized. And if it could be virtualized, it could be infused with intelligence at every layer.
By the late 2010s, the concept of AI-RAN had crystallized: not just AI for network operations, but AI integrated directly into the RAN compute stack, making split-second decisions about beamforming, scheduling, and resource allocation. The vision was compelling, but the hardware wasn’t ready. General-purpose CPUs, even when augmented with specialized accelerators, struggled to meet the stringent latency and throughput requirements of real-time RAN processing.
The GPU Gambit: NVIDIA’s Entrance and the Hardware Revolution
The turning point came when NVIDIA, flush with success in the data center AI market, turned its attention to telecommunications. The company’s graphics processing units (GPUs), already the workhorses of AI training and inference in the cloud, offered a tantalizing proposition: massive parallel processing power that could handle both the computational demands of RAN signal processing and the neural network workloads of AI algorithms.
NVIDIA’s Aerial platform, a software development kit for building AI-on-5G applications, and its subsequent Aerial RAN Computer, represented a bold bet that the future of RAN would be GPU-accelerated. The pitch was straightforward: instead of dedicated ASICs that could only perform fixed functions, GPUs offered flexibility. They could run RAN workloads today and be reprogrammed for new AI algorithms tomorrow, future-proofing operator investments.
But skepticism abounded. Traditional RAN vendors argued that GPUs were power-hungry and expensive, ill-suited for the harsh economics of cell site deployment. The transmission time interval (TTI)—the 1 millisecond window in which a base station must complete critical tasks—imposed a hard limit on computational complexity. Any algorithm that missed the TTI was worthless, as radio conditions would have changed by the time the calculation completed. Ericsson’s head of RAN compute, Michael Begley, articulated this constraint bluntly at MWC Barcelona, noting that the physics of the radio interface would always constrain the size of AI models that could be deployed in real time.
Yet Nokia, under the leadership of CEO Justin Hotard, saw an opportunity. The company had been developing its AirScale portfolio, originally designed to support Marvell Technology’s silicon. The vision was to create a slimline GPU that could slot into existing AirScale chassis, operating within the same power constraints and at a comparable price point. The promise was compelling: operators could deploy AI-RAN software upgrades for years, reaping the benefits of ongoing enhancements without replacing hardware.
“What we anticipate with AI-RAN is that it gives customers an opportunity to reduce the ongoing hardware investment, or to elongate that,” Hotard explained in a call about first-half financial performance. “Then they can get the benefit of ongoing enhancements once they deploy AI-RAN software.”
From Trials to Field: The Eight-Operator Milestone
Nokia’s announcement this week marks the transition from laboratory to field for this vision. The eight operators now advancing trials represent a cross-section of the global telecom landscape: A1 Group in Austria, Chunghwa Telecom in Taiwan, du and e& in the UAE, Mobily and stc in Saudi Arabia, TPG Telecom in Australia, and Zain Saudi. Their collective move from early evaluation to live field trials is significant not just in scale but in geographic diversity, suggesting that AI-RAN’s appeal transcends market maturity and regulatory environments.
The trials are built on the NVIDIA Aerial RAN Computer, which integrates GPU acceleration with Nokia’s RAN software. Early results are promising: the platform delivers over 20% spectral efficiency improvements, a figure that, if sustained in commercial deployments, would translate into substantial capacity gains without additional spectrum. For operators facing relentless data growth and constrained capital budgets, such efficiency improvements are not merely nice-to-have—they are existential.
Nokia’s collaboration with NTT DOCOMO, which has been ongoing, continues to progress, adding depth to the trials. DOCOMO’s involvement is particularly noteworthy given its history of RAN innovation and its early embrace of multi-vendor Open RAN architectures. The operator’s participation lends credibility to Nokia’s approach and suggests that AI-RAN is being evaluated not in isolation but as part of a broader Open RAN strategy.
The financial markets have taken notice. Nokia’s stock rose nearly 6% to 17, implying substantial upside if the trials translate into commercial contracts.
The Open RAN Connection: Why AI-RAN Matters for Disaggregated Networks
For the Open RAN community, Nokia’s AI-RAN progress carries particular significance. The two movements—Open RAN and AI-RAN—have always been intertwined, sharing a common vision of disaggregated, software-driven, intelligent networks. Open RAN’s promise of interoperability and vendor diversity creates the ideal foundation for AI-RAN, where algorithms can be deployed as applications (dApps) on a multi-vendor RAN infrastructure.
Nokia’s strategy reflects this convergence. The company has been positioning its AI-RAN platform as compatible with Open RAN principles, emphasizing the E3 interface for dApp development and the potential for third-party innovation. The slimline GPU approach, designed to fit existing AirScale chassis, lowers the barrier to entry for operators who have already invested in Nokia’s RAN infrastructure, while the software-upgrade path aligns with Open RAN’s ethos of continuous improvement.
Yet challenges remain. The integration of AI into RAN raises questions about interoperability, testing, and security that the industry is only beginning to address. How do AI algorithms trained on one vendor’s data perform on another’s equipment? How can operators validate that AI-driven decisions meet regulatory requirements for spectrum usage and quality of service? These are not trivial concerns, and they will require sustained collaboration across the ecosystem.
The Road Ahead: From Field Trials to Commercial Reality
The transition from field trials to commercial deployment is rarely linear. Nokia’s eight operator partners will spend months—perhaps years—evaluating performance, refining algorithms, and building the operational processes needed to manage AI-driven networks. The 20% spectral efficiency improvement observed in trials must be replicated across diverse deployment scenarios, from dense urban canyons to suburban sprawl and rural expanses.
Moreover, the business case for AI-RAN must be proven not just in technical terms but in economic ones. Operators will need to see clear returns on their investment, whether through reduced energy costs, deferred capital expenditure, or new revenue-generating services. Nokia’s promise of elongating hardware investment cycles is appealing, but it must be weighed against the cost of GPUs and the complexity of managing a hybrid CPU-GPU RAN environment.
The competitive landscape is also evolving. Ericsson, Nokia’s traditional rival, has been pursuing its own AI-RAN strategy, emphasizing RAN compute platforms and subscription-based AI services. Samsung has demonstrated AI-driven RAN gains in live 5G networks with KDDI in Japan. The battle for AI-RAN supremacy will be fought not just on technology but on ecosystem, partnerships, and the ability to deliver tangible value to operators.
Conclusion: A Milestone, Not a Destination
Nokia’s announcement that eight global operators are advancing AI-RAN trials is a milestone worth celebrating, but it is not the finish line. The telecommunications industry has learned, often painfully, that the gap between demonstration and deployment can be vast. AI-RAN’s journey from concept to commercial reality has been long and winding, marked by false starts, technological hurdles, and shifting expectations.
What makes this moment different is the convergence of enabling technologies: powerful, energy-efficient GPUs; mature Open RAN interfaces; and a growing body of operational experience with AI in network management. The operators participating in Nokia’s trials are not naive optimists; they are hard-nosed businesses seeking competitive advantage in a market where efficiency and agility are paramount.
As the trials progress, the industry will watch closely. Success would validate the GPU-accelerated RAN vision and accelerate the shift toward AI-native networks. Failure would reinforce the skepticism that has long surrounded AI-RAN, pushing the dream of intelligent, autonomous networks further into the future. For now, the field trials represent a brave step forward—a bet that the long road from lab to live is finally nearing its destination.
Sources
- Nokia stock rises 6% as AI-RAN trials expand, defense deal signed
- Nokia to start dealing out GPUs in high-stakes AI-RAN gamble
- The week in AI-native telco (3-9 Oct 2026), The AI-Native Telco | TelecomTV
- MTN Opens 2026 Technology Hackathon to African Innovators and Employees - TechAfrica News
- AI will reshape networks as intelligence moves beyond DCs: says Deutsche Telekom
- Ericsson urges India operators to focus on 5G monetisation, FWA, AI services for future growth - Telecompaper