
VodafoneThree plans to test Nokia’s AI-native anyRAN software running on the NVIDIA Aerial RAN computer, first in a controlled lab environment and then in a selected-site field trial in the UK, to measure whether AI in the radio access network can improve capacity, reliability and spectral efficiency. The UK operator, formed from the merger of Vodafone UK and Three UK in June 2025, described the program as planned when it announced it on November 18, 2026. It gave no start date, duration, site location or timetable for moving from the lab to the field.
The first use cases are AI-for-RAN functions, which apply machine learning to how the radio network itself operates: machine learning-based channel estimation and link adaptation. VodafoneThree says these could help the network assess changing radio conditions more accurately and adapt how data is transmitted in real time. It will benchmark the results against its existing network performance, measuring capacity, reliability and spectral efficiency, and will study how the capabilities could be scaled across the wider network. The operator has also listed a wider set of potential AI-RAN use cases: radio optimisation, mobility, energy efficiency, anomaly detection, coverage and capacity optimisation, Quality of Service (QoS) assurance, and sharing accelerated compute between RAN and AI workloads.
The proposed architecture combines Nokia’s anyRAN software with Red Hat OpenShift and NVIDIA accelerated computing as a common platform for both RAN and AI workloads. VodafoneThree ties the program to differentiated connectivity services, citing its Vodafone SuperMobile plan, which it says offers speeds up to four times faster than standard plans in busy areas and guaranteed minimum speeds of 15 Mbps, and to future services over terrestrial and non-terrestrial networks. The announcement does not name the spectrum bands, radio units, number of GPU servers or sites involved, and it does not mention any commercial deployment decision.
“AI-RAN represents an exciting next step in the evolution of our network, using AI to make connectivity more intelligent, responsive and efficient. Working with Nokia, we’ll test these capabilities to understand how AI can improve network performance and, ultimately, the experience we deliver for our customers. As our 5G Standalone network evolves, we see significant potential for AI-RAN to help us respond more dynamically to different customer needs and unlock new, differentiated connectivity experiences, while laying the foundations for the next generation of intelligent, AI-native networks,” said Iain Milligan, Network Development and Infrastructure Director, VodafoneThree.
| Key Points: VodafoneThree AI-RAN Trial |
| THE TRIAL• Operator: VodafoneThree (UK), with Nokia and NVIDIA • Planned lab trial, then a selected-site field trial • Announced November 18, 2026; no dates given for either phase |
| THE PLATFORM• Nokia AI-native anyRAN software • NVIDIA Aerial RAN computer and accelerated computing • Red Hat OpenShift • One platform for RAN and AI workloads |
| INITIAL USE CASES• Machine learning-based channel estimation • Machine learning-based link adaptation • Benchmarked against existing network on capacity, reliability and spectral efficiency |
| WIDER AGENDA• Radio optimisation, mobility, energy efficiency, anomaly detection • Coverage and capacity optimisation, QoS assurance • Shared accelerated compute for RAN and AI; differentiated services such as Vodafone SuperMobile |
| NOT DISCLOSED• Start dates, duration and field site location • Spectrum bands, radios and server counts • Commercial deployment plans |
Analysis
VodafoneThree is joining Nokia’s AI-RAN program at an early stage. Nokia launched its commercial AI-RAN platform with NVIDIA in July 2026, with pilot deployments scheduled for later in 2026 and commercial availability planned for 2027, and in September it added eight more operators to proofs of concept, lab evaluations and live trials. Among operators in Converge Digest’s coverage, T-Mobile is furthest along: it completed an AI-RAN call in October on its commercial 5G Standalone network in Bellevue, Washington, with Nokia’s anyRAN software on NVIDIA Aerial RAN Computer servers. In Europe, Deutsche Telekom validated AI-enhanced uplink channel estimation in labs in Bonn and Oulu ahead of 2027 field trials, and BT plans live pilots in Birmingham in 2027. VodafoneThree, which has yet to start its lab phase, is behind both.
The benchmark results will carry the most weight. Nokia says its AI radio algorithms have shown more than 20% spectral-efficiency improvement and that its roadmap targets 50% by 2027 and more than 100% by 2028, but it has not disclosed the test conditions. VodafoneThree plans to measure capacity, reliability and spectral efficiency against its own network, which would provide an operator-run comparison. Its announced metrics do not include GPU utilization, power consumption per site or cost, which together decide whether a GPU-based RAN node pays for itself, although energy efficiency and shared RAN and AI compute are on its list of potential use cases.
The trial also sits within a network where Ericsson is the main supplier. In September 2025, Ericsson signed an SEK 12.5 billion (about $1.13 billion), eight-year agreement to build the majority of VodafoneThree’s radio access network and all of its core network. The announcement does not say how a Nokia field site would connect into that network or whether a commercial award for Nokia could follow. For Nokia, the UK trial adds another large European operator to an AI-RAN strategy backed by NVIDIA’s $1 billion equity investment, which gave NVIDIA a 2.9% stake in the company in October 2025.
| Company Profile: VodafoneThree |
| OVERVIEW• UK mobile network operator serving fixed and mobile markets; registered office Newbury, Berkshire • Formed from the merger of Vodafone UK and Three UK in June 2025 • Brands include Vodafone UK, Three UK, VOXI Mobile, SMARTY and Talkmobile • Iain Milligan, Network Development and Infrastructure Director |
| NETWORK PLAN• £11 billion network investment program • Target: 99% 5G Standalone population coverage by 2030 and 99.96% by 2034 • Ericsson: majority of RAN and all of core under an eight-year deal (Sept 2025) |
| PARTNER: NOKIA• Headquarters Espoo, Finland; President and CEO Justin Hotard • Emma Falck, President of Mobile Infrastructure • AI-native anyRAN software on NVIDIA Aerial RAN Computer • NVIDIA holds a 2.9% stake after a $1 billion investment (Oct 2025) |
| WEBSITES• vodafonethree.com • nokia.com |
We’re tracking AI-RAN as operators move GPU-accelerated radio software from labs to field trials and commercial networks. See Converge Digest AI-RAN coverage.
