AI-RAN momentum builds and it might be time to pay attention
Artificial Intelligence (AI) has triggered a global supercycle of compute demand, and in the years ahead, networks may struggle to keep up. As AI workloads begin moving closer to users and devices, it may no longer be enough for mobile networks to merely transport AI-generated traffic; they will need to become active participants in the intelligence loop.
This necessity has brought AI-RAN to the forefront, positioning it as a critical step toward future-proof network design. AI-RAN spans both AI-for-RAN — where AI enhances network performance and efficiency — and AI-on-RAN, where the network itself enables AI applications at the edge. In this paradigm, the base station is reimagined as a computing node capable of running diverse workloads that blend AI and RAN functions, powered by accelerated hardware and the growing softwarization of network functionalities. Early signs of this shift are already visible, as the industry converges on the idea that the base station of the future will be as much a computing platform as it is a radio node.
A Pivotal Investment: NVIDIA and Nokia
The $1 billion investment by NVIDIA in Nokia, announced in October 2025 at NVIDIA’s GTC in Washington D.C., marked a pivotal moment. For the first time, an AI company placed a billion-dollar bet on the future of telecom infrastructure; not to build faster chips or larger data centers, but to transform the mobile network itself into an AI platform.
The partnership between Nokia and NVIDIA, supported by T-Mobile U.S. and Dell Technologies, fuses the economics of accelerated computing with the operational scale of telecom networks.
Nokia, with decades of leadership in radio and network systems, brings deep “network DNA” and best-in-class expertise in building carrier-grade infrastructure. NVIDIA contributes its unmatched strength in AI and accelerated computing. Together, they form a uniquely complementary partnership.
As part of Nokia’s anyRAN strategy, this collaboration expands its portfolio with a new AI-RAN architecture in which the NVIDIA GPU becomes an integral part of the radio baseband replacing traditional ASIC-only designs and allowing 5G and future 6G RAN software to run on a programmable, commercial-off-the-shelf (COTS) Aerial RAN Computer Pro (ARC-Pro) platform. For decades, mobile networks relied on custom-built processors — efficient but rigid and costly. AI-RAN breaks that pattern. In this model, the same GPUs that power virtualized radio functions can also execute AI inference when traffic loads are low, transforming what was once a fixed cost into a shared, revenue-generating resource. Each base station effectively becomes a miniature AI data center capable of serving nearby devices, vehicles, and enterprises. This convergence could finally make the long-discussed edge computing model economically viable enabling networks to generate value beyond connectivity.
The AI-RAN Alliance: Shaping the Future
The AI-RAN Alliance, which now counts more than 115 member organizations, has become the platform for shaping this direction. Unlike a traditional standards organization, the Alliance focuses on building implementation blueprints and practical frameworks, enabling members to co-develop and test interoperable AI-RAN solutions. This approach allows faster experimentation and iteration — making it an especially meaningful platform for academic partners such as the University of Oulu.
At its most recent meeting in Boston, hosted by Northeastern University — one of the leading academic centers in AI-RAN research — the AI-RAN Alliance convened to look back on the progress and chart the next phase of its roadmap. Discussions focused on finalizing the AI-RAN reference architecture, reviewing progress across ongoing work items, and defining concrete next steps for evolution and benchmarking. While the two flagship initiatives DATA4AI and Test Methodology continue to gather momentum, a new task group on agentic AI was also established, exploring how autonomous AI-driven agents can manage network behavior dynamically and intelligently. T he meeting was complemented by an impressive showcase of AI-RAN-related demonstrations at Northeastern, illustrating how academia can meaningfully contribute to the shared industry vision. T he ambition is bold yet clear: to transform the telecommunications network into the world’s distributed inference engine, where intelligence is embedded across every layer of connectivity.
6G Flagship’s Role in Advancing AI-RAN
Amid this momentum, 6G Flagship continues research toward AI-RAN. From its inception, 6G Flagship recognized AI as a key enabler for next-generation wireless systems, and today it stands among the most active academic contributors within the Alliance leading the task group on a survey of the state of the art in AI-for-RAN.
As part of its ongoing work within the AI-RAN Alliance, the AI-RAN team at 6GFlagship has been advancing research on multimodal, sensor-driven digital twins for AI-RAN in collaboration with NVIDIA, aimed at enabling more intelligent networks. The work focuses on how networks can perceive and model the physical world around them, closing the loop between sensing, simulation, and optimization, enabling real-time digital twins. T he team presented its latest developments at both NVIDIA GTC in Washington D.C., at the AI-RAN Alliance booth, and at the Brooklyn 6G Summit together with NVIDIA. The demonstration showcased how real-time digital twins can dynamically generate channel information by combining LiDAR and depth sensing, enabled by a ray tracer operating directly on point clouds (NimbusRT). This highlighted the potential of multimodal perception to provide real-time environmental awareness that can be leveraged to optimize network operation and lay the groundwork for future AI-native RAN deployments.
One of the highlights at the Boston meeting was that the AIRAN team from 6G Flagship was among the ten recipients of the AI-RAN Innovation Challenge Award for its proposal envisioning a framework for multimodal, sensor-driven digital twins for AI-RAN.
This aligns with the AI-RAN Alliance’s vision of a unified communications–sensing–control ecosystem:
- AI-for-RAN: Using AI and sensor data to optimize radio operations.
- AI-with-RAN: Sharing compute resources between AI, digital twin processing, and RAN functions at the network edge.
- AI-on-RAN: Enabling new AI-driven services powered by the digital twin utilizing the RAN itself.
This holistic approach supports the transition from connectivity to intelligence as a service — a transformation particularly relevant to enterprise and industrial networks. Smart factories, hospitals, and logistics hubs increasingly depend on both reliable connectivity and environmental awareness.
From Connectivity to Intelligence
As AI continues to redefine both computing and connectivity, AI-RAN is emerging not merely as a technical upgrade, but as a necessary leap toward networks that learn, adapt, and think alongside the world they connect.
Realizing this vision will require deep collaboration across industry and academia. Within this global effort, 6G Flagship continues to play an active role through ongoing research and new initiatives that will soon unfold.
One of the ongoing is the SNS JU funded EU–ROK consortium project 6GARROW, which brings AI natively into radio access networks to enable smarter, more efficient and flexible wireless connectivity. By combining AI-driven optimisation with advanced device–network integration, 6GARROW supports seamless interoperability and fosters new 6G innovations. Oulu contributes by advancing energy-efficient terminal technologies: using generative AI to reduce transmitted data, developing power efficient constellation techniques, and applying AI to mitigate hardware impairments for better device performance.
The shift now underway will determine how the next era of mobile innovation is shaped. Those who understand that connectivity itself is becoming intelligent — and act on it — will shape the foundations of the 6G era. For everyone else, the AI-RAN transformation may not wait.
This article originally appeared in the December 2025 issue of 6G Waves Magazine.
About the author
AI-RAN Research Manager