Artificial intelligence is moving from telecom customer service and network planning directly into the radio access network, creating a potentially important new investment cycle for mobile operators. The business case for AI-RAN is straightforward: use AI to extract more capacity from expensive spectrum, automate network operations, improve customer experience and potentially turn mobile infrastructure into distributed AI computing capacity.
The opportunity is becoming substantial. Dell’Oro Group forecasts cumulative AI-RAN revenue of $35 billion between 2026 and 2030, although it does not expect AI-RAN to materially expand the overall RAN market. Worldwide RAN revenue is forecast to grow at only around a 1 percent CAGR through 2030. This suggests AI-RAN spending will initially replace or enhance conventional network investment rather than create an entirely new RAN spending pool.
The Dell’Oro AI-RAN market forecast raises an important question for mobile operators: can investment in AI generate enough capacity and operational savings to justify additional software, computing and energy costs?
SoftBank Achieves Up to 25% Spectral Efficiency Gain
SoftBank has provided one of the strongest real-world examples of AI improving an existing 5G network.
In August 2026, SoftBank and Ericsson tested an AI-native scheduler on the Japanese operator’s 5G network. Compared with conventional technology, the trial achieved up to approximately 25 percent higher spectral efficiency and up to approximately 50 percent higher downlink user throughput.
More importantly, across all evaluated locations, both spectral efficiency and downlink throughput improved by approximately 10 percent.
The SoftBank-Ericsson commercial 5G AI trial demonstrates why AI-for-RAN could have an attractive near-term business case.
Spectrum represents one of the largest investments made by mobile operators. If AI enables an operator to carry more traffic over the same frequencies and existing radio equipment, it could potentially delay capacity upgrades in congested areas and lower cost per bit.
Nokia Targets More Than 100% Spectrum Efficiency Improvement
The longer-term capacity claims are even more aggressive.
Nokia launched its AI-native RAN platform in July 2026, combining Nokia anyRAN software with NVIDIA’s Aerial AI-RAN platform. Nokia says its AI-driven radio technologies have demonstrated more than 20 percent improvement in spectral efficiency.
The company is targeting a 50 percent gain by 2027 and more than 100 percent by 2028, effectively aiming to double the capacity available from operators’ existing spectrum assets.
The Nokia AI-native RAN platform supports multiple migration routes, including accelerated computing for existing AirScale infrastructure, standalone GPU-powered AI-RAN nodes and cloud-native deployments.
If operators can achieve these gains economically in commercial networks, AI could alter the traditional relationship between traffic growth and network Capex. Instead of responding to every capacity problem by adding spectrum, radios or sites, operators could increasingly improve capacity through software and AI processing.
However, telecom operators will need to compare the value of that additional capacity against GPU costs, software subscriptions and higher computing power requirements.
NTT DOCOMO Uses AI-RAN to Improve Customer Experience
NTT DOCOMO and Samsung are examining AI-RAN at the individual subscriber level.
Their technology uses AI to understand each user’s radio conditions, movement and service requirements, allowing the network to predict potential degradation and automatically select a better network configuration.
In a January 2026 validation using network data and a local 5G trial environment, NTT DOCOMO and Samsung reduced the frequency of communication-speed degradation from 13.1 percent to 7.2 percent, an improvement of 5.9 percentage points.
The DOCOMO-Samsung user-level AI-RAN project demonstrates that the return from AI-RAN does not necessarily need to come from higher total network capacity.
Better quality can reduce congestion-related customer problems while supporting demanding services such as video, gaming and video conferencing. Predictive optimization could also become increasingly important as operators move toward autonomous 5G-Advanced and eventually 6G networks.
Deutsche Telekom Cuts Network Response From Hours to One Minute
Operational expenditure presents another major AI business case.
Deutsche Telekom’s RAN Guardian Agent, developed with Google Cloud technology, uses agentic AI to identify traffic events, detect potential network problems and automatically adjust network parameters.
Since its launch in November 2025, RAN Guardian has reduced the time required to manage major events from hours to around one minute — an improvement of more than 95 percent. It also autonomously initiated more than 100 remediation actions during its first month.
For 2026, the system identified 237,000 events. During Germany’s Carnival season, it detected around 130 events and parades, each expected to attract more than 10,000 people, and pre-checked 611 mobile sites. Only five sites experienced peak-load conditions requiring optimization.
Following deployment in Germany, Deutsche Telekom is expanding RAN Guardian into European markets, starting with the Czech Republic and Croatia.
The Deutsche Telekom autonomous RAN initiative illustrates an important AI-RAN economic opportunity: operators could manage increasingly complex networks without proportionally increasing engineering resources.
T-Mobile Sees AI-RAN Creating New Revenue
T-Mobile US is looking beyond cost reduction.
T-Mobile US, in association with NVIDIA, Ericsson and Nokia, is developing AI-RAN technologies that combine radio processing with accelerated computing. T-Mobile says AI can optimize spectral efficiency and capacity in real time while anticipating congestion and proactively allocating network resources.
The longer-term opportunity is to use the same distributed infrastructure for third-party AI workloads.
Potential applications include large language models, vision-language models, industrial digital twins, robot fleet orchestration, spatial computing and interactive digital avatars.
If telecom operators can sell spare AI computing capacity at the network edge, AI-RAN could potentially create revenue in addition to reducing network costs. This model remains less proven than AI-based network optimization, but it represents an important strategic difference between AI-RAN and traditional RAN upgrades.
Energy and GPU Costs Could Challenge the Business Case
The biggest question is whether additional AI computing costs offset the savings.
GPU-based RAN could provide greater programmability and allow operators to share infrastructure between connectivity and AI workloads. But deploying accelerated computing throughout mobile networks could also increase equipment and electricity costs.
That is why the first large-scale opportunity may be AI-for-RAN rather than complete GPU-based AI-RAN replacement.
Operators can initially introduce AI into scheduling, interference management, traffic prediction, fault detection, energy optimization and capacity planning while retaining much of their existing radio and baseband infrastructure.
Dell’Oro expects near-term adoption to be dominated by AI-for-RAN, single-purpose applications, distributed RAN, 5G and non-GPU architectures. AI-RAN therefore appears more likely to evolve gradually through software upgrades and targeted accelerated-computing deployments than through wholesale network replacement.
AI-RAN Economics Are Moving From Theory to Real Networks
The operator evidence emerging in 2026 points to four different business cases.
SoftBank is demonstrating capacity, achieving up to 25 percent higher spectral efficiency and 50 percent higher downlink throughput.
NTT DOCOMO is demonstrating customer-experience gains, cutting communication-speed degradation from 13.1 percent to 7.2 percent.
Deutsche Telekom is demonstrating operational savings, reducing network-event management from hours to approximately one minute.
T-Mobile US is exploring new revenue, where AI-RAN infrastructure could eventually support both mobile connectivity and enterprise AI workloads.
Nokia’s target of more than 100 percent improvement in spectral efficiency by 2028 adds a potentially transformative fifth element: extracting dramatically more value from spectrum operators already own.
With AI-RAN revenue forecast to reach $35 billion cumulatively through 2030, the technology is moving toward commercial scale. But operators are unlikely to adopt AI-RAN simply because it introduces GPUs or AI into mobile infrastructure.
The winning architecture will be the one that demonstrates a measurable reduction in cost per bit.
For global mobile operators, the AI-RAN investment decision will ultimately depend on whether higher spectrum efficiency, delayed network expansion, lower operating costs, improved customer experience and new edge-AI revenue can exceed the additional cost of computing, software and energy.
BABURAJAN KIZHAKEDATH
