Hello,

AI moved from slideware into live network control this week.
The real story is not “AI in telecom.” It is operators taking back control of the network layer.

Let’s dive in.

Ericsson and T-Mobile announced on 12 May 2026 that T-Mobile’s 5G Advanced network had moved AI-native Scheduler with Link Adaptation into large-scale commercial trials on live network traffic. Ericsson said the software runs on its hardware, uses a neural network to predict radio conditions in real time, and achieved close to a 10% rise in spectral efficiency and up to a 15% boost in downlink throughput versus legacy rule-based methods.

One day earlier, Ericsson and KDDI announced a large-scale AI uplink optimisation field trial in Japan. The trial covered about 1,500 5G cells and 1,300 4G cells, using Ericsson’s Uplink Interference Optimizer rApp on Ericsson Intelligent Automation Platform, alongside a third-party rApp from Japan-based FYRA. Ericsson reported 9.6% average 4G throughput improvement, 3.1% average 5G throughput improvement, and a 27% improvement in 5G SINR.

The stronger signal is not the headline percentage. It is where AI is being placed. T-Mobile’s case sits inside the scheduler and link adaptation layer. KDDI’s case sits inside uplink interference optimisation, tied to rApps and live O-RAN-aligned automation. These are not customer-care bots. They are radio decisions, live traffic and measurable performance.

Two operators. Two countries. Two different network problems. Same direction.

T-Mobile is using AI to make downlink performance more efficient under changing radio conditions. KDDI is using AI to improve uplink quality, which matters for XR, industrial devices, video upload, remote work and physical AI use cases. The common thread is operational control. AI is being embedded close to where the network makes decisions.

The KDDI result is more interesting because it also proves third-party rApp integration. Ericsson said the field trial validated EIAP support for Ericsson and non-Ericsson rApps, which matters for operators trying to avoid closed AI control loops.

RAN AI is no longer just about better KPIs. It is becoming a control-layer question. Operators should now ask vendors where AI runs, who can audit it, whether third-party rApps can plug in, and what happens when two AI control loops disagree.

Vodafone announced on 5 May 2026 that it had agreed to buy CK Hutchison’s 49% stake in VodafoneThree for £4.3 billion. Once complete, Vodafone will own 100% of the UK’s largest mobile operator and one of the country’s fastest-growing broadband providers. Vodafone said the deal is expected to complete in the second half of 2026, subject to UK National Security and Investment Act approvals.

A week later, on 12 May 2026, Vodafone published its FY26 preliminary results for the year ended 31 March 2026. The official release itself points investors to the full results package, while Reuters reported Vodafone expects core earnings of €11.9 billion to €12.2 billion for the current year after refocusing around Germany, Britain and Africa.

Vodafone’s own release says the VodafoneThree integration has already delivered network quality improvements ahead of schedule, with better coverage, speeds and reliability. It also expects £700 million in annual cost and capex synergies by FY30.

This is not directly an AI deployment story. It is still an AI-in-telecom story because control of the operating base decides how fast AI can be shipped.

A 51% owner can influence the network. A 100% owner can force the roadmap. That matters when AI in telecom depends on data access, RAN upgrades, OSS/BSS integration, cloud capacity, field operations and investment timing. VodafoneThree gives Vodafone one UK operating engine rather than a negotiated joint venture.

The bet is simple: consolidate first, automate faster after. That is cleaner than trying to layer AI over fragmented ownership, duplicated systems and slow joint-governance decisions.

Watch VodafoneThree less as a merger story and more as an execution platform. If Vodafone can turn UK scale into faster 5G rollout, FWA growth and network automation, other European operators will face pressure to simplify before they automate.

AI-heavy telecom networks need compute, clean data and automation. In many African markets, the more immediate constraint is still energy.

AP reported two weeks ago that African telecom tower operators are accelerating the shift from diesel to solar and hybrid power as fuel costs rise. The article cited roughly 500,000 towers across Africa and said energy can account for up to 60% of tower operating expenses in off-grid regions. It also reported that MTN in South Sudan has cut fuel costs by 30%, while Airtel in Zambia and Congo has halved diesel usage through solar and hybrid systems.

This is not separate from telecom AI. Remote towers with weak energy economics are harder to monitor, harder to optimise and harder to turn into edge nodes. If operators want AI-driven planning, predictive maintenance and autonomous network operations, they need stable infrastructure under the model.

Africa will not copy Europe’s AI network path. The business case starts from resilience and cost control, not only performance uplift.

Developed markets talk about AI-RAN, programmable networks and autonomous operations. Emerging markets often need a more basic sequence: reduce diesel exposure, improve uptime, stabilise site economics, then automate.

That does not make the African case less advanced. It makes it more honest. AI in telecom has to meet the real operating model. In markets where energy is the largest tower cost, the best AI use case may be predicting battery drain, fuel logistics, solar output, theft risk and site maintenance before it optimises radio parameters.

The winner in these markets will not be the operator with the flashiest AI demo. It will be the one that connects power, towers, network data and field operations into one decision loop.

Emerging-market AI strategy should start with site economics. If energy, uptime and field visits are the biggest cost centres, automate those first. RAN AI matters, but unreliable infrastructure will eat the business case before the model reaches production.

A new paper published on 15 May 2026 argues that 6G should move away from vendor-led infrastructure cycles and toward operator-controlled digital services. The paper frames the issue around five priorities: control first, customer first, business first, operations first and technology last. It also proposes a “6G Control Compact” and a “Guarantee Economy” model, linking 6G architecture to enforceable service outcomes rather than generic connectivity.

This lands at the same time as AI-RAN economics are becoming more serious. A March 2026 paper on programmable AI-RAN argues that GPU-heavy RAN deployments could create returns from leasing surplus compute capacity to AI tenants, with modeled ROI up to 8x across certain scenarios.

The key shift is commercial, not academic. If RAN hardware, edge compute and AI workloads begin to sit on shared infrastructure, operators must decide what they own, what they rent and what they expose.

6G will not be won by a better air-interface story alone. Operators already learned from 5G that technical upgrades without clear monetisation create pressure on capex and returns.

The new 6G control argument fits what is happening in live networks this week. T-Mobile and KDDI are embedding AI into RAN behaviour. Vodafone is tightening control over its UK operating base. African operators are reworking tower economics. All four point to the same operating question: who controls the layer where decisions are made?

Vendors will still matter. But operator-side control, auditability and commercial ownership are becoming harder to ignore.

Start treating 6G strategy as an ownership design problem. The right question is not only which vendor has the best roadmap. It is which parts of the AI-native network the operator must own, federate or consume.