The telecom sector is entering a more consequential phase of artificial intelligence adoption. The shift is no longer limited to customer care bots, analytics dashboards, or isolated automation tools. Major infrastructure vendors are now embedding AI more deeply into radio access networks, edge platforms, and software-defined network stacks, arguing that future mobile systems will need intelligence built into the network itself rather than added to it later. Huawei used Mobile World Congress 2026 to launch enhanced “AI-Centric Network” solutions tied to 5G-Advanced and the transition to 6G. Samsung has been advancing AI-RAN work with NVIDIA around software-based networks. Ericsson is framing AI RAN as a strategic layer for performance, automation, and energy efficiency. Nokia, working with Deutsche Telekom, is pushing AI-native RAN use cases across cloud, edge, and radio domains.
The Evolving Role of AI in Telecom Infrastructure
Huawei’s recent messaging is among the clearest examples of that repositioning. At MWC Barcelona 2026, the company said it was building “AI-centric networks and computing backbones” to help carriers respond to the AI era, while linking those systems to high-capacity, low-latency infrastructure for mobile AI applications. Huawei also said there are already 70 million 5G-Advanced users globally and described the next several years as a window for large-scale 5G-Advanced deployment and eventual 6G evolution. Whatever one’s view of Huawei’s market position across different geographies, the company’s language is notable in that it treats AI not as an operational feature but as a design principle for the network.
Samsung is advancing a similar thesis from a more software-centric angle. Its March 2025 announcement with NVIDIA said Samsung had demonstrated interoperability between its O-RAN-compliant virtualized RAN and NVIDIA accelerated computing, with the stated goal of making AI-RAN deployable on commercial-off-the-shelf server infrastructure. Samsung’s broader network AI material further extends that message, arguing that cloud-native, software-based telecom platforms lay the foundation for domain-specific AI integration across the RAN, core, transport, and service layers. The operational promise is straightforward: more predictive maintenance, faster anomaly detection, better resource allocation, and a tighter feedback loop between network conditions and real-time decisions.
Ericsson has drawn one of the sharpest distinctions in this market by separating “AI for RAN,” “AI in RAN,” and “AI on RAN.” That framework matters because it helps explain the commercial stakes. “AI for RAN” refers to intelligence operating around the radio stack, often through orchestration and optimization layers. “AI in RAN” points to models embedded directly in the radio software stack for functions such as beamforming and link adaptation. “AI on RAN” goes a step further by using distributed network infrastructure to support low-latency multimodal services for devices such as robots, smart glasses, and extended reality endpoints. In other words, the network is being positioned not only as something AI manages, but as something AI workloads can run on and through.
NVIDIA’s telecom strategy helps explain why so many vendors are converging around this architecture. Its AI-RAN positioning is explicit: a software-defined platform that fuses radio and AI workloads on shared infrastructure. NVIDIA says the model supports voice, data, video, and growing “AI traffic” from devices such as cameras, robots, drones, and AI agents, while enabling operators to run 5G functions and AI applications on a common accelerated platform. The company also claims that dynamically allocating 5G and AI workloads on the same GPU can increase capacity utilization by two to three times while improving energy efficiency. Those claims will need to be proven in commercial deployments at scale, but they show where vendor roadmaps are heading: fewer siloed stacks, more pooled compute, and tighter integration between connectivity and inference.
Nokia’s collaboration with Deutsche Telekom points to the same end state, but from the operator modernization side. The companies said in March 2026 that they would work on AI-powered receivers, channel estimation and prediction, adaptive beamforming, and predictive network optimization, while also pushing standards-aligned, interoperable interfaces. That is significant because the AI debate in telecom is increasingly about operational control. Operators want better automation, but many also want vendor-agnostic management layers, open interfaces, and flexibility across cloud, edge, and radio domains. AI-native infrastructure will be more attractive if it lowers lock-in rather than deepens it.
The industry ecosystem around this shift is expanding quickly. The AI-RAN Alliance said in February 2026 that it had reached 132 members worldwide and was showcasing 33 innovation demos tied to AI-native networks. That does not mean the architecture is settled, but it does suggest that AI-RAN has moved beyond a small group of lab experiments. In parallel, the GSMA and the World Economic Forum have both emphasized that telecom operators are under pressure to use AI to improve efficiency, create new revenue opportunities, and support AI-driven services. The result is a two-sided market logic: telecom is using AI to run networks better, and it is also trying to become the infrastructure on which AI services are delivered.
Evaluating the Impact of Transitioning to AI-Native Networks
AI-centric telecom infrastructure is no longer a futuristic concept. It is becoming the organizing principle behind how major vendors describe 5G-Advanced, edge computing, and the path to 6G. Huawei is framing AI as a network design model. Samsung is tying AI-RAN to software-based deployment and multi-domain orchestration. Ericsson is defining the technical layers of AI integration inside and around the RAN. Nokia is aligning AI-native RAN with open, programmable network management. NVIDIA is pushing shared AI-and-RAN compute as the commercial substrate beneath it all. The industry’s direction is increasingly clear: the next phase of telecom competition will hinge not only on coverage and capacity, but on how intelligently networks can sense, decide, adapt, and host new AI-era services.
