Nokia is attempting to move mobile-network economics closer to the software industry, introducing an AI accelerated radio platform that operators can deploy through upgrades, standalone nodes, or cloud infrastructure and purchase through a continuing subscription.

Nokia says the platform will combine its anyRAN software with Nvidia’s accelerated computing technology to increase the traffic operators can carry over existing spectrum and radio estates. Pilot deployments are expected towards the end of 2026, followed by commercial availability in 2027.

Three routes are planned. Existing AirScale customers will be able to add a GPU powered capacity unit, operators can install standalone AI-RAN nodes, and cloud-oriented providers will be offered systems running on commercial accelerated servers supplied through partners.

Nokia says its present technology has demonstrated spectral-efficiency improvements exceeding 20%, with a target of 50% during 2027 and more than 100% by 2028. Those figures remain company claims rather than results from broad commercial deployment, although the AI-RAN product material gives operators a timetable against which to test them.

More capacity without another spectrum purchase

Mobile operators face a persistent economic problem because traffic continues to grow while spectrum acquisition and radio replacement require substantial capital. Software and accelerated computing could allow them to extract more capacity from installed assets, defer parts of a hardware refresh, or improve service in dense areas without purchasing an equivalent amount of additional spectrum.

That capacity will not be free. GPUs use electricity, subscriptions create recurring expenditure, and higher-performance baseband systems may require changes to cooling, network sites, transport links, and operations. Any comparison with conventional equipment needs to include energy, licensing, integration, maintenance, and hardware across the whole lifecycle.

The model also changes the cadence of network development. Traditional radio equipment advances through long hardware cycles, whereas software based infrastructure can receive algorithms and performance improvements more frequently. Nokia wants operators to pay for that continuing development through subscriptions, moving some expenditure from periodic capital purchases into operating budgets.

Telecommunications companies have heard similar claims during previous network-virtualisation programmes, many of which progressed more slowly than suppliers expected. Radio workloads require predictable performance, low latency, reliability, and efficient specialist processing, which makes them less forgiving than conventional cloud applications.

Open interfaces can still create concentrated dependency

Nokia says the platform will comply with Open RAN standards and work with compatible radio units, allowing customers to combine suppliers. An open interface does not automatically remove dependency when software, accelerator architecture, optimisation tools, and the commercial roadmap remain closely connected.

Nvidia’s involvement adds another concentration question. General accelerated computing can make the radio network more programmable, but it may also extend the influence of a dominant AI-computing supplier into telecommunications infrastructure. Operators will need to examine portability, software support, hardware availability, security, and long-term pricing alongside radio performance.

Network skills will change as radio engineering absorbs more software development, model optimisation, cloud operations, and data management. Knowledge of spectrum and antennas remains indispensable, although it will sit alongside faster release cycles and closer monitoring of algorithmic behaviour.

Security responsibilities will expand as well, because a programmable radio estate introduces more code, interfaces, update mechanisms, and supply relationships. Operators need assurance that performance improvements cannot compromise isolation, resilience, emergency-service availability, or recovery from a faulty release.

Pilot customers will be able to test whether Nokia’s claimed gains survive real traffic patterns, diverse sites, mixed equipment, and national operational requirements. They will also determine whether subscription pricing leaves enough of the economic value with the operator after hardware and energy costs have been included.

Nokia has not yet demonstrated that AI-RAN will double usable capacity across commercial networks or turn 6G into a software update. It has, however, moved the discussion into product selection and procurement, where measured performance, energy consumption, integration cost, and supplier dependency will decide whether the model progresses beyond controlled trials.