--- title: "Private 5G with inference in the radio | Nexus Observatory" description: "A private cellular network gives a site its own coverage, its own security and its own priorities. AI-RAN adds a second idea…" url: https://observatory.global/stories/private-5g-ai-ran/ publisher: Nexus Observatory Inc. updated: 2026-10-31 --- [Stories](https://observatory.global/stories/)System: Private 5G and AI-RAN # Private 5G with the inference in the radio A private cellular network gives a site its own coverage, its own security and its own priorities. AI-RAN adds a second idea: the accelerators that run the radio can also run the site’s models, and machine learning can tune the radio itself12. **Systems**: Open RAN, Shared accelerators, Edge inference **Toward 2030**: AI in the air interface, 6G **Figures**: 4, field scene, systems diagram, data story, timeline to 2030 Go to the references **01** Field scene ## A private cell on an industrial site. A tower with three sectors, an edge rack beside it and the machines and sensors it serves. Shared spectrum frameworks let enterprises license a band locally, such as the Citizens Broadband Radio Service in the United States3 and local licences in the United Kingdom4. Figure as text: An open-pit mine with a private cell, GNSS monitoring and edge An open pit carved into real terrain with benches, three GNSS monitoring points, a seismic station, a private 5G tower and an edge node. - Private 5G - Edge rack - Slope monitoring - Vehicle **02** Systems diagram ## Where the radio runs. Open RAN splits the base station into a radio unit at the mast, a distributed unit and a central unit, joined by open interfaces, with intelligent controllers that host applications5. When the distributed unit runs on a general accelerator, the same hardware can serve the site’s inference, the idea the AI-RAN working groups call AI-and-RAN1. Figure as text: An Open RAN with a shared accelerator for radio and inference A four-tier diagram: devices and sensors, radio units, a shared accelerator hosting the distributed unit and inference, and the core with the site applications. - **Devices** : Camera, Vehicle, Sensors - **Radio units** : Radio unit 1, Radio unit 2 - **Shared accelerator** : DU and inference, RAN controller - **Core and applications** : 5G core, Site applications 10 links. Scenarios: Normal operation; Radio unit 1 down. **03** Data story ## Why the band decides the site count. Free-space loss rises with frequency: at a kilometre, every doubling of frequency costs six decibels before walls, foliage or rain6. That is the physics behind the move from low bands for coverage to mid and high bands for capacity. Figure as text: Free-space path loss at 1 km by band A bar chart of free-space path loss at one kilometre for six bands from 700 MHz to 100 GHz, rising from about 89 to about 132 decibels. | Band | Loss | | --- | --- | | 700 MHz | 89.3 | | 1.8 GHz | 97.5 | | 3.5 GHz | 103.3 | | 6 GHz | 108 | | 26 GHz | 120.7 | | 100 GHz | 132.4 | **04** Timeline to 2030 ## Machine learning moves into the standard. 3GPP studied machine learning for beam management, positioning and channel state feedback in Release 182 and has continued that work in later releases on the road to 6G7. Figure as text: Open and intelligent RAN, 2018 to 2030 A timeline from 2018 to 2030 with the O-RAN Alliance, shared spectrum, the AI-RAN Alliance, the 3GPP machine learning study and planned 6G releases. 1. **2018-02** O-RAN Alliance formed. Open interfaces and intelligent controllers for the radio access network. 5 2. **2019** Local licensing for shared spectrum. Enterprises license mid-band spectrum for their own sites. 4 3. **2023-12** Machine learning for the air interface studied (3GPP TR 38.843). Beam management, positioning and channel feedback use cases. 2 4. **2024-02** AI-RAN Alliance formed. AI-for-RAN, AI-and-RAN and AI-on-RAN. 1 5. **2029** First 6G specifications planned. Release 21, in 3GPP’s published timeline. 7 Figure 1 Field scene Figure 1. An open-pit mine with a private cell, GNSS monitoring and edge. An open pit carved into real terrain with benches, three GNSS monitoring points, a seismic station, a private 5G tower and an edge node. ## References 1. [AI-RAN Alliance: AI-for-RAN, AI-and-RAN and AI-on-RAN working groups](https://ai-ran.org) . AI-RAN Alliance, 2024. 2. [3GPP TR 38.843: Study on artificial intelligence and machine learning for the NR air interface (Release 18)](https://www.3gpp.org/DynaReport/38843.htm) . 3rd Generation Partnership Project, 2023. 3. [47 CFR Part 96: Citizens Broadband Radio Service (3550–3700 MHz)](https://www.ecfr.gov/current/title-47/chapter-I/subchapter-D/part-96) . United States Federal Communications Commission, 2015. 4. Enabling wireless innovation through local licensing: shared access to spectrum supporting mobile technology. Ofcom, 2019. 5. [O-RAN Architecture Description and the O-RAN specifications](https://www.o-ran.org/specifications) . O-RAN Alliance, 2024. 6. [Recommendation ITU-R P.525: Calculation of free-space attenuation](https://www.itu.int/rec/R-REC-P.525/en) . International Telecommunication Union, Radiocommunication Sector, 2019. 7. [Releases: the timeline of 3GPP Releases 17 to 21](https://www.3gpp.org/specifications-technologies/releases) . 3rd Generation Partnership Project, 2025. ## Related stories [System: Edge AI inference **Deciding at the edge** A model that detects smoke, flooding or a fault is most useful where the frame is captured. Running inference at the edge cuts the time to a decision and…](https://observatory.global/stories/edge-ai/) [Horizon 2030: 6G and sensing **Networks that sense** The framework for 6G names integrated sensing and communication as one of its usage scenarios: the same radio that carries data also measures range,…](https://observatory.global/stories/sensing-networks-6g/) [Place: Coast and port **The port that sees the approach** A port observes three things at once: the ships coming in, the sea they come through and the cranes that receive them. Ship identity travels on a radio…](https://observatory.global/stories/coast-port/) ## Where this applies [Connectivity](https://observatory.global/connectivity/)[Edge compute & AI](https://observatory.global/edge-ai/)[Software](https://observatory.global/software/)[Hardware & power](https://observatory.global/hardware/)[Test and qualification](https://observatory.global/#lines)[Industry](https://observatory.global/industries/industry/)[Mining](https://observatory.global/industries/mining/)[Transport](https://observatory.global/industries/transport/)[Telecom](https://observatory.global/industries/telecom/)[Events](https://observatory.global/industries/events/)[Field tour: open pit mine](https://observatory.global/environments/open-pit-mine/)[Field tour: coastal port](https://observatory.global/environments/coastal-port/) © 2026 Nexus Observatory Inc. **Legal** [Privacy notice](https://observatory.global/legal/privacy/) · [Terms of use](https://observatory.global/legal/terms/) · [Cookie policy](https://observatory.global/legal/cookies/) · [Accessibility](https://observatory.global/legal/accessibility/) · [Official languages](https://observatory.global/official-languages/) · [Notices](https://observatory.global/notices/) **Company** [About](https://observatory.global/about/) · [Open governance](https://observatory.global/consortium/) · [Contact](https://observatory.global/contact/)