What are VodafoneThree’s AI-RAN trials with Nokia and NVIDIA?
VodafoneThree will test Nokia‘s AI-native anyRAN software running on the NVIDIA Aerial RAN Computer, starting in a controlled lab and then moving to selected live sites in the UK. The aim is better spectrum efficiency, a network that reacts faster to changing radio conditions, and shared computing for RAN and AI services. With this announcement, all three UK mobile network operators have AI-RAN work with Nokia either completed or planned.
The announcement on 18 November 2026 came a day after parent company Vodafone Group opened a multi-vendor AI-RAN test hub in Málaga, Spain. Together, the two moves show Vodafone running a two-track approach: a UK trial programme tied to a specific vendor platform, and a group-level lab designed to compare chips, software and architectures before any wide deployment decision.
Why VodafoneThree’s AI-RAN trials matter
AI-RAN has moved quickly from conference slides to operator networks during 2026. What has been missing is evidence from live networks at scale. The UK is now becoming one of the clearest places to watch that evidence build, because all three national operators are testing Nokia’s AI-RAN platform at different stages, with different goals and different hardware.
For VodafoneThree, the trials sit inside a much larger network programme. The operator has committed £11 billion to building what it calls the UK’s best network, and it has split its radio estate between Ericsson and Nokia. That means it is now testing AI in the RAN with both of its main radio vendors at the same time, each using a different approach to computing.
Three things make this announcement more significant than another trial press release:
- Spectrum is the constraint. Operators cannot buy their way out of capacity limits easily, so software that extracts more throughput from existing spectrum has a direct business case. Nokia has set a target of doubling spectral efficiency by 2028.
- Compute is becoming shared infrastructure. VodafoneThree will look at whether the same accelerated computing that runs the radio network can also host AI services, which changes how operators might value cell-site investment.
- Products need proof. VodafoneThree links the work to differentiated connectivity services, citing gaming, streaming, working and AI applications on the move, building on its Vodafone SuperMobile plan.
What the VodafoneThree AI-RAN trials will test
The programme combines Nokia’s RAN software, NVIDIA’s Aerial RAN Computer and Red Hat OpenShift as the cloud platform layer. Testing begins in a lab before moving to a selected-site field trial on VodafoneThree’s network. VodafoneThree has not published a timetable, but the trial work is expected to run during 2027.
The operator has outlined three areas of focus:
- AI for RAN. Machine learning applied to how the radio network operates, such as channel estimation and link adaptation. One example the operator gives is spotting a weakening connection and automatically adjusting how data is sent to keep the customer connected.
- Network performance and spectrum use. Results will be benchmarked against the commercial network for capacity, reliability and spectral efficiency.
- AI and RAN on shared compute. Whether the same accelerated computing platform can run both the mobile network and AI applications, creating a foundation for future edge AI services.
Iain Milligan, Network Development and Infrastructure Director at VodafoneThree, described AI-RAN as an exciting next step in the evolution of the network and said the operator sees significant potential for it to help respond more dynamically to different customer needs. Emma Falck, President of Mobile Infrastructure at Nokia, called the collaboration an important step in moving AI-RAN from industry ambition toward commercial reality, and said the approach protects existing network investments while creating a software-driven foundation for edge AI services and AI-native 6G.
Vodafone’s Málaga lab and the wider AI-RAN validation effort
Vodafone Group opened its AI-RAN lab in Málaga on 17 November 2026. It builds on the site’s Open RAN work and is designed as a multi-vendor innovation and validation hub, producing repeatable benchmarks that show where AI adds value and guide deployment decisions for Vodafone and its partners. Nokia and NVIDIA are named partners, alongside academic partners, other equipment vendors and software developers.
The lab is deliberately chip-agnostic. It is equipped with the NVIDIA Aerial RAN Computer MGX platform on QCT servers and Intel Xeon 6 (Granite Rapids) servers, and it will also test the purpose-built telecom chips and x86 processors already used in Open RAN. That allows Vodafone to compare GPU-accelerated RAN with alternatives on the same test bed before committing capital.
Vodafone has organised the work into three programmes:
| Programme | Question it answers | Planned milestones |
| AI-for-RAN | Can AI improve beamforming, resource allocation, energy use and spectral efficiency? | Validation of AI radio algorithms and rApps, xApps and dApps |
| AI-and-RAN | Can RAN functions and AI applications share the same computing infrastructure? | Silicon benchmarks (SPEC, MLPerf, AI-RAN Alliance); first lab 5G massive MIMO call in December 2026, then field trials |
| AI-on-RAN | Can the network host AI applications close to users? | Physical AI, vision AI and robotics use cases; integrated sensing and communications assessment |
Kyriakos Exadaktylos, Head of RAN Evolution in Vodafone Group Technology, said the lab will help turn AI-RAN ambitions into measurable results and lay the foundations for 6G.
Vodafone describes lab work and field trials over this year and next as the way to test customer benefits and the business case before wider deployment. In other words, the Málaga results are intended to decide where and how AI-RAN scales, not simply to showcase it.
How VodafoneThree’s AI-RAN plans compare with O2 and BT
The three UK operators are at different points on the path from lab to live network. The comparison below separates what has been completed from what is planned:
| Operator | AI-RAN partners | Status | Reported result or target |
| Virgin Media O2 | Nokia | Completed: live AI-powered RAN feature at the O2 Giga Site in London Paddington (October 2026) | Up to about 20% higher spectral efficiency (operator-reported, single site) |
| BT / EE | Nokia, NVIDIA | Planned: first AI-RAN sites on the live EE network in Birmingham in 2027 | Aims to test whether lab gains of more than 20% (Nokia-reported) hold in a live network |
| VodafoneThree | Nokia, NVIDIA (Red Hat platform) | Planned: lab testing, then a selected-site field trial | No targets published; benchmarked against the commercial network |
| VodafoneThree | Ericsson | Under way: AI-ready AIR 3266 massive MIMO rollout in London (November 2026) | AI scheduler trial before end of financial year; nationwide rollout planned for 2027 subject to validation |
Two patterns stand out. First, the O2 result came from a software feature on existing hardware, while BT and VodafoneThree are testing NVIDIA-accelerated platforms. The UK will therefore produce evidence on both routes to AI in the RAN. Second, VodafoneThree is running parallel AI-RAN tracks with both of its main radio vendors, one built on Ericsson’s own silicon and one on NVIDIA GPUs. Whatever its trials show, it will hold comparative data on both approaches from the same national network.
What VodafoneThree’s AI-RAN trials mean for operators and vendors
For operators, the most useful outcome of 2027 will not be a headline efficiency figure. It will be clarity on where AI-RAN pays back. A gain in spectral efficiency at a busy transport hub is valuable, but the business case changes depending on whether the gain needs new accelerated hardware at every site or can run as software on existing equipment. Vodafone’s decision to benchmark GPUs, CPUs and purpose-built chips side by side reflects exactly that question.
For Nokia, the UK is turning into a showcase market. Having all three national operators testing its AI-RAN software gives it a rare chance to show results across different network designs, traffic patterns and hardware choices within one country. Nokia and NVIDIA have said their first commercial AI-RAN release is planned for 2027, so the UK trials line up closely with that product timeline.
For NVIDIA, VodafoneThree adds another European operator testing the Aerial RAN Computer in a live environment, following trial commitments from BT and others. The AI-and-RAN programme, which explores running AI services on the same compute as the radio network, is central to NVIDIA’s argument that cell sites can become distributed AI infrastructure.
For enterprises, the link is less direct but worth watching. Differentiated connectivity, guaranteed minimum speeds and edge AI hosting are the services AI-RAN is meant to enable. Vodafone SuperMobile in the UK already promises speeds up to four times faster than standard plans in busy areas and a guaranteed minimum of 15 Mbps. If AI-RAN delivers, those service guarantees could become easier to offer more widely.
What comes next for VodafoneThree’s AI-RAN programme
Several milestones over the next 12 months will show how far the work has progressed:
- Vodafone’s first lab-based 5G massive MIMO call on shared AI-and-RAN infrastructure in Málaga, planned for December 2026.
- Publication of silicon benchmark results comparing GPU, CPU and purpose-built chip performance for RAN and AI workloads.
- The move from lab testing to VodafoneThree’s selected-site field trial in the UK, and whether the operator publishes KPIs from it.
- BT’s first AI-RAN sites in Birmingham in 2027, which will give the UK a second live-network data point alongside O2’s Paddington result.
- VodafoneThree’s parallel Ericsson AI scheduler trial and its planned nationwide rollout, subject to validation.
Today, the evidence base is still thin. One operator has published a live result from a single site, and the rest is planned work. What makes VodafoneThree’s announcement worth following is the structure behind it: a vendor trial in the UK, a chip-neutral lab in Spain and a parallel programme with a second vendor. By the end of 2027, that combination should give the industry some of the clearest operator-led evidence yet on whether AI-RAN is ready to scale.
For more analysis like this, subscribe to the TeckNexus Newsletter, explore our 2027 programs, or partner with us to share your perspective and generate leads.











