For years, AI-native telecom networks were a familiar Mobile World Congress promise. MWC 2026 changed the tone. Field trials, commercial hardware, open-source tools and a broad operator coalition showed that AI-RAN is moving from research agenda to deployment choice.
The shift matters beyond telecom. Networks built to run radio functions and AI workloads on shared infrastructure could place inference closer to where data is generated, while software-defined operation could make connectivity evolve more like cloud infrastructure.
Two competing routes to an AI-native network
Nvidia secured commitments from more than a dozen operators and technology companies—including BT Group, Deutsche Telekom, Ericsson, Nokia, SK Telecom, SoftBank, T-Mobile and Cisco—to pursue open, secure, software-defined 6G foundations. It also released a 30-billion-parameter Nemotron Large Telco Model, guidance for building network-operations agents and blueprints for energy efficiency and configuration.
Nokia’s partnership with Nvidia supplied important live evidence. Its anyRAN software was tested over the air with T-Mobile US, Indosat Ooredoo Hutchison and SoftBank. At T-Mobile’s Seattle innovation centre, a single Nvidia Grace Hopper 200 server handled commercial 5G alongside video streaming, generative AI queries and AI video captioning. IOH completed what it called Southeast Asia’s first AI-RAN-powered Layer 3 5G call.
Ericsson is taking a different path. Its ten new AI-ready radios use neural-network accelerators embedded in purpose-built silicon rather than external Nvidia GPUs. The company argues that custom hardware can offer stronger power efficiency, total cost of ownership and supply-chain independence. A wider collaboration with Intel will span compute, cloud and AI-driven RAN and packet-core uses.
Live trials reveal new operating models
The demonstrations were not only about faster radio performance. SoftBank showed its AITRAS Orchestrator identifying spare compute capacity that could run third-party AI workloads. That points toward a new business model in which operators monetise distributed computing at network sites as well as connectivity.
SoftBank also demonstrated AgentRAN, an autonomous system that translates natural-language goals into real-time 5G and 6G configurations. SK Telecom described an AI-native rebuild extending from the network core to customer service, including automated wireless quality management, traffic control and equipment operations.
The surrounding hardware ecosystem is expanding quickly. Quanta, Supermicro, MSI and Lanner announced systems designed to combine AI inference, radio functions and packet processing. AMD promoted its EPYC edge platform and Open Telco AI initiative as another route from pilot to production. The range of suppliers suggests AI-RAN is becoming a procurement category, not merely a vendor demonstration.
Enterprise implications reach beyond the radio
For enterprise technology leaders, the architectural question is where inference should run, on what hardware and at what cost. GPU-based platforms may offer flexibility and shared capacity. Custom accelerators may deliver efficiency and tighter optimisation. MWC did not settle that contest, but it made the trade-off concrete.
AI at the network edge could support workloads that benefit from low latency and proximity to local data. Software-led upgrades could also reduce dependence on lengthy hardware refresh cycles. At the same time, combining network and AI workloads raises new expectations for orchestration, security and energy management.
Nvidia’s telecom report cited in the source found that 77% of respondents expect AI-native wireless architecture to deploy significantly faster than previous network generations. Forecasts are not guarantees, but the activity in Barcelona provides substance behind the optimism.
MWC 2026 proved that the components are arriving together: tested radio software, shipping hardware, operator demand and emerging operational tools. The open question is no longer whether AI will enter telecom infrastructure. It is which architecture can deliver intelligence, resilience and acceptable economics at network scale.
The answer will emerge in live networks, where energy use, reliability and revenue matter more than the most impressive conference demonstration.
