Summary

• O2 and Nokia completed a live deployment of AI-powered radio network optimisation at O2's Paddington Giga Site.
• The companies report spectral efficiency improvements of up to approximately 20%, alongside better throughput during busy periods.
• The trial forms part of O2's £700 million network investment programme, but no nationwide rollout timetable has been confirmed.

UK mobile operator O2 and Nokia have completed a live trial of AI-powered radio network optimisation in London Paddington, reporting spectral efficiency improvements of up to approximately 20% while the site continued serving ordinary mobile customers.

The technology was deployed at O2’s Giga Site near one of London’s busiest transport hubs, where changing passenger numbers and mobile traffic create demands that can vary considerably throughout the day.

Nokia’s software uses predictive AI to assess radio conditions and anticipated traffic before adjusting how available network resources are allocated. The approach is intended to improve capacity without requiring an equivalent increase in spectrum or physical infrastructure.

O2 announced the results on 26 October as part of its £700 million Mobile Transformation Plan. The companies describe the trial as the UK’s first live deployment of this particular AI-powered radio access network feature, although they have not published detailed independent test data or a timetable for wider commercial deployment.

Making more use of existing spectrum

Mobile operators can increase network capacity by acquiring additional spectrum, upgrading equipment or installing more sites, but each option involves commercial and technical constraints.

Radio spectrum is finite and subject to licensing arrangements, while new infrastructure may require suitable locations, planning approvals and substantial investment.

Improving how efficiently existing frequencies are used offers another route to carrying additional traffic, particularly in congested environments where demand changes frequently.

Spectral efficiency measures the amount of information that can be transmitted through a given amount of radio spectrum. Better efficiency allows an operator to carry more data using the same frequencies, provided the remainder of the network can support the additional traffic.

At Paddington, O2 and Nokia tested predictive resource allocation on a live 5G installation rather than relying exclusively on a laboratory environment.

The companies say the feature produced spectral efficiency gains of up to approximately 20% during the trial, with additional improvements in throughput when the network was busy.

Those measurements are company-reported results from a particular site. They should not be interpreted as evidence that every customer received a 20% faster connection or that the same improvement would occur across the operator’s entire network.

Individual download speeds depend on signal conditions, device capabilities, available capacity, traffic congestion and other technical factors.

Nor does a percentage improvement in spectral efficiency directly establish financial savings. The economic value depends on the additional traffic carried, implementation expenditure and whether the technology reduces or delays the need for other capacity investments.

Prediction replaces some reactive adjustments

Radio networks already contain automated processes that allocate resources according to changes in demand and signal quality. Operators have also used AI techniques in planning, fault management and performance analysis for several years.

The Paddington development applies predictive software more directly to resource management within a live radio access network.

According to the companies, the system continuously assesses network conditions and anticipates demand before adjusting resource allocation, reducing the need for manual intervention during normal operation.

That may be particularly useful where traffic patterns change rapidly, as can happen around transport interchanges, major events and busy commercial areas.

However, a predictive system can also make incorrect assumptions. If traffic changes unexpectedly or a model misinterprets network conditions, its chosen allocation may be less effective than anticipated.

Operators therefore need controls that keep automated adjustments within suitable technical limits and allow engineers to investigate unusual behaviour.

The live trial provides evidence that the software can operate while customers are connected, but does not establish how it will behave across equipment generations, different geographical conditions or extended periods of sustained operation.

Monitoring, software maintenance and change management remain important because the network must continue functioning reliably as models and surrounding infrastructure are updated.

Live deployment differs from full commercial rollout

Nokia is participating in several European initiatives examining how artificial intelligence can improve radio network performance and support future communications services.

Some projects investigate the integration of radio processing with accelerated computing infrastructure, while others focus on improving particular network functions through machine learning.

These approaches overlap but are not interchangeable. A feature that improves spectrum allocation does not establish that a telecommunications operator has deployed every element of a broader AI-native architecture.

Earlier in October, Techopia examined Nokia’s wider AI radio network programme, including work involving Elisa, Deutsche Telekom and other European operators.

The O2 trial adds a separate result from a completed live deployment at a specific UK site, rather than repeating those earlier programme announcements.

It also differs from laboratory performance claims because customers continued using the installation during the test.

However, wider adoption will require evaluation across multiple network environments. Performance improvements achieved at a busy transport hub may differ from those available at suburban sites, industrial campuses or rural installations.

The business case will depend partly on where traffic demand is sufficiently high or variable to justify the additional software and operational complexity.

Software adds another investment layer

O2’s Mobile Transformation Plan includes substantial upgrades to its existing network, with separate multi-year arrangements involving Nokia and Ericsson to modernise radio equipment and improve coverage and capacity.

The AI trial sits within that wider investment programme, rather than representing a standalone commitment of £700 million to the Paddington feature.

Software-based optimisation may help operators make better use of upgraded infrastructure, but adoption still requires integration, testing and support.

Network engineers must determine how the feature interacts with established operational systems and whether the performance benefits remain consistent as demand and equipment configurations change.

Those requirements can become more demanding where multiple suppliers contribute to the same network and software changes must be coordinated across different hardware platforms.

For commercial operators, a successful feature should ideally improve performance without introducing unnecessary dependence on specialist engineering resources or creating new failure points.

O2 and Nokia say the Paddington deployment maintained stable network accessibility and reliability during the trial. The companies have not disclosed detailed reliability measurements or independent comparisons covering extended operation.

More testing will be needed to establish how effectively the system can be expanded and which types of sites offer the greatest benefits.

The project demonstrates that predictive radio management can improve efficiency under live customer conditions, but its long-term commercial contribution will depend on repeatable results and deployment costs across a wider network estate.