Summary
- XLSmart and ZTE signed a memorandum of understanding at MWC Shanghai 2026 to deepen work on AI-powered networks, 5G and fixed wireless access.
- The intended operating surface spans radio access, transport, core infrastructure and network operations rather than one isolated optimisation tool.
- XLSmart is described as serving more than 69.4 million customers after the combination of XL Axiata, Smartfren Telecom and Smart Telecom.
- The companies cite an existing AI-enabled implementation that delivered energy savings above 10% while maintaining network performance and customer experience.
- That result is company-reported and comes without a disclosed baseline, duration, site count, absolute energy use, monetary value or independent validation.
The memorandum matters because it moves artificial intelligence from a single network feature toward an end-to-end operating proposition. It also creates a demanding burden of proof.
A model that adjusts radio power is not the same product as a control layer spanning RAN, transmission, core and operations. Once automation crosses those boundaries, its value depends on whether the operator can turn many local recommendations into lower total cost without weakening resilience or customer service.
XLSmart and ZTE have described the ambition. They have not yet disclosed the system-level result.
An end-to-end scope increases both leverage and risk
Radio optimisation can change when cells, carriers or antenna functions use power. Transport intelligence can reroute traffic or allocate capacity. Core-network tools can anticipate load, while operations systems can prioritise alarms and interventions.
Connecting those layers could remove duplicated forecasting and shorten the distance between a traffic signal and an operational response. It also increases dependency. A poor forecast can propagate from the radio layer into capacity planning, maintenance and customer policy.
The relevant question is therefore not whether each component contains AI. It is who controls the objectives, how conflicts are resolved and what happens when the model is wrong. An operator needs rollback, audit trails and human authority across every automated decision surface.
The 10% claim needs a denominator
ZTE says an existing AI-powered energy-efficiency implementation achieved savings above 10% while maintaining network performance and customer experience. That is useful directional evidence, but it is not yet a network-wide business case.
The announcement does not identify the number or type of sites, the pre-intervention baseline, the measurement period, seasonal effects, traffic growth, absolute kilowatt-hours or the value of the saved electricity. It does not disclose an independent assessor.
“More than 10%” could represent an important recurring saving across a large footprint, or a controlled result in a narrow cluster. Both are compatible with the published statement. Until the denominator is available, the figure should be treated as a company-reported implementation result, not as proof that the whole XLSmart network has become 10% more efficient.
Merger scale changes the integration problem
XLSmart serves more than 69.4 million customers following the merger of XL Axiata, Smartfren Telecom and Smart Telecom. That scale offers more traffic data and a larger cost base on which optimisation could matter.
It also means the network may contain different equipment generations, operating practices, coverage obligations and customer profiles. AI does not erase those differences. It needs consistent telemetry, permissions, asset records and performance definitions before it can optimise across them.
The most valuable near-term result may be operational consistency: a common way to detect congestion, classify faults and decide where engineers should intervene. That benefit would still need evidence such as fewer repeat incidents, shorter repair times and lower energy per carried unit.
FWA turns network optimisation into a retail test
The parties say they plan to broaden 5G coverage and expand 5G fixed-wireless home services during 2026. FWA converts spare mobile capacity into a household product, but it also gives the operator a new source of sustained evening traffic.
The economics depend on spectrum, coverage, indoor reception, customer equipment, installation, backhaul and the number of households sharing a sector. AI can help forecast demand or allocate resources; it cannot create capacity that is not present.
No rollout timetable by market, capital commitment, household target, speed tier, price or subscriber forecast appears in the announcement. The memorandum sets a direction. It does not establish that FWA will produce attractive unit economics everywhere XLSmart offers it.
The next disclosure should connect savings to service
Useful evidence would report energy per unit of traffic, the number and mix of sites, the measurement period, peak-hour performance, dropped sessions, fault rates and customer-experience indicators before and after automation. FWA reporting should add sector utilisation, install success, average busy-hour speed, churn and support cost.
Those measures would reveal whether efficiency comes from genuine optimisation rather than from reducing capacity or shifting cost elsewhere. They would also allow the operator to distinguish a repeatable operating model from a vendor demonstration.
XLSmart and ZTE have widened the control surface they intend to automate. The memorandum becomes consequential only when the denominator grows as precise as the ambition.