On the floors of MWC Barcelona and Nvidia’s GTC events, you could feel it: AI-RAN is quickly rising as the most talked-about subject in telecom. Major industry leaders are now concentrating on creating innovative approaches to both “AI for Networks” and “Networks for AI”.
What was a loose cluster of whitepapers and reference architectures two years ago is now a billion-dollar battleground. Demos, PoCs, announcements – the pace has shifted and is shaping up as the future of telcos.
What AI-RAN really is – and why it matters now?
So, what is AI-RAN? It sits on foundations that have been maturing for more than a decade:
- NFV and SDN that led to Cloud Native architectures on the infrastructure level
- vRAN and O-RAN with a concept of moving away from purpose-built RAN solutions towards virtualized / cloudified, disaggregated and open RAN deployable on standard COTS servers
- Multi-access Edge Computing (MEC) network architecture that brings cloud computing resources closer to users. To significantly reduce latency, congestion and enhance security.
Distributed networks of near-user servers – typically underutilized because of how RAN capacity is engineered – suddenly look like strategic assets. If RAN workloads can coexist with third-party applications, then MNOs have something rare in telecom: a chance to create new revenue streams.
By boosting servers’ computing power through enhanced AI inference capabilities, MNOs can leverage high-performance 5G/6G AI-improved RAN processing solutions and establish a distributed AI grid to support Edge AI applications. In addition, MNOs are well positioned to maintain sovereignty over technology, data storage, and operations. With latency requirements below 10 milliseconds, AI-RAN is anticipated to become the primary network architecture for Physical AI.
The AI-RAN Alliance
The largest organization supporting the concept is the AI-RAN Alliance, revealed at MWC 2024, now counting more than 130 members. Although the industry’s biggest players aren’t all inside its tent, many of those outside still back the idea of AI in the RAN and the coexistence of the RAN and Edge AI workloads.
For Unikie, joining the AI-RAN Alliance this year is a major step as we move forward with our physical AI solutions. For over 10 years, Unikie has been developing the enabling technologies for physical AI, including sensoring, CPU/GPU algorithms, UxVs, autonomous vehicles, and all the major connectivity and mobile communication stacks. Today, Unikie’s expertise covers full device, far/near edge to cloud continuum – intersecting directly with the AI-RAN agenda.
Three fronts of AI-RAN (and where Unikie plays)
The AI-RAN developments worldwide are focused around three domains. Find the AI-RAN Alliance’s own nomenclature in brackets.
Applying AI in RAN (AI-for-RAN)
Scope: AI leveraged optimization of RAN performance including radio signal processing, network operation and automation.
Unikie’s focus:
- AI Agent supported CPU/GPU RAN SW Algorithms optimization
- AI-assisted RAN SW porting to targeted HW architectures
- RF sensing, 6G ISAC, Simultaneous Localization and Mapping (SLAM)
- x/rApps optimisation and commercialization (w/ Partner)
Infrastructure sharing (AI-and-RAN)
Scope: Workload orchestration, co-existence and isolation on shared compute infrastructure based on cost and performance targets.
Unikie’s focus:
- Zero trust containerized architecture following highest security standards
- Policy based workload placement (w/ Partner)
Edge AI Applications (AI-on-RAN)
Scope: Enabling advanced applications at the network edge with bringing AI capabilities closer to the user for low-latency applications and real-time AI inference.
Unikie’s focus:
- Commercialization of Edge AI solutions with examples of Autonomous Vehicle Marshalling, Drones Swarms Mission Control and Planning, SLAM for Defence and Public Safety, Predictive Maintenance with Composite AI
- CPU/GPU Algorithms optimization
- CUDA Engineering Agent for Edge AI Apps
- Multi-stream data integration, assessment and control
- Device to edge processing offloading
The Physical Limits of Cloud Are Coming
The more AI becomes embodied in vehicles, robots, drones and industrial solutions, the less viable cloud computing becomes. These complex systems will create demand for high bandwidth and low latency processing with reliable, predictable, no-jitter connectivity and low time-to-first-token (TTFT) variance. And that, ultimately, is why the industry is treating AI-RAN not as another architectural option but as the enabler for Physical AI.
In the end, AI‑RAN will rise or fall on execution, not enthusiasm. The physics of the edge, the economics of shared compute, and the stubborn realities of RAN integration will decide how quickly Physical AI moves from keynote slides into everyday networks. For Unikie, physical AI has been part of our DNA since the beginning. For over a decade, our teams have collaborated with industry-leading clients and deployed hundreds of commercial products leveraging the enabling technologies of physical AI.
If AI‑RAN becomes the architecture that finally lets autonomous machines, sensing networks and real‑time AI operate at human speeds, we intend to be among the companies proving it works.
Read more about Unikie’s expertise for Physical AI
Mariusz Rudnicki
Telecom & Edge AI, Director of Business Development
+48 600 200 496 | mariusz.rudnicki@unikie.com
The author is a senior telecom and edge computing leader with close to 25 years of international experience spanning 2G–6G RAN, Cloud and Edge Computing, and AI-driven telecom solutions. Mariusz has held global sales, business development, and delivery leadership roles, working closely with Tier-1 operators, network equipment providers, and silicon vendors across Europe, North America, and APAC.
Mariusz is deeply involved in the evolution of Telecom, with hands-on experience across virtualized and cloud-native RAN, xPU-accelerated architectures, and emerging Edge AI business models. He actively collaborates with the broader ecosystem to bridge advanced network technologies with real-world deployment and commercialization. At Unikie, his focus is on translating Physical AI concepts into scalable, deployable solutions at the network edge.
