Ucell has finished a nation‑wide rollout of ZTE’s AI‑driven RAN energy‑saving platform. Real‑time traffic forecasting and per‑cell power control boost the network’s data‑per‑kilowatt‑hour ratio by 10.6 %, trimming carbon output and OPEX while keeping LTE/5G performance steady.
Ucell and ZTE Deploy AI‑Powered Green RAN Across Uzbekistan, Cutting Energy Use by 10.6 %
Date: 26 May 2026 | Source: ZTE press release
Ucell, one of Uzbekistan’s largest mobile operators, has completed a full‑scale installation of ZTE’s AI‑enabled RAN energy‑saving solution. The deployment covers more than 4 500 base stations, spanning LTE and 5G sites in urban, suburban and rural zones. The system promises a measurable lift in the energy efficiency ratio (GB/kWh) and a proportional drop in carbon emissions and operating costs.
How the AI engine works
The platform uses a two‑layer intelligence model:
| Layer | Role | Typical latency |
|---|---|---|
| Network‑level AI | Forecast traffic for the next 15 min‑hour, generate site‑wide power‑saving policy | ~200 ms (cloud‑edge) |
| Base‑station AI | Execute per‑cell power mode changes, monitor KPI drift, abort if thresholds breached | <50 ms (on‑site) |
The network‑level model ingests historical traffic, weather, and event calendars. It outputs a site‑wide energy‑saving schedule that tells each base station which cells can enter low‑power states. The base‑station AI then applies symbol‑level muting, carrier‑level shutdown, and equipment‑level hibernation in real time, constantly checking throughput, latency, and RRC connection success rates.
If any KPI exceeds the pre‑set limits (e.g., latency > 30 ms, drop rate > 2 %), the local AI instantly restores full power, ensuring the user experience remains unchanged.
Measured impact
ZTE supplied the following figures after a 30‑day monitoring window:
| Metric | Before AI rollout | After AI rollout | Δ |
|---|---|---|---|
| Data traffic per kWh (GB/kWh) | 1.84 | 2.03 | +10.6 % |
| Network‑wide power draw (MW) | 112.5 | 106.2 | ‑5.6 % |
| CO₂ emissions (tCO₂/yr) | 45 800 | 43 200 | ‑2 600 |
| OPEX (energy cost, USD/yr) | 4.8 M | 4.5 M | ‑6 % |
The uplift in GB/kWh translates directly into lower carbon output: each megawatt‑hour saved avoids roughly 0.45 tCO₂, matching the reported reduction.
Compatibility checklist for homelab‑style replication
If you want to experiment with a similar AI‑driven energy‑saving stack in a test lab, here are the minimum components you’ll need:
- Radio Access Hardware – Any vendor’s LTE/5G eNodeB/gNodeB that exposes a programmable API (e.g., ZTE ZXA10, Nokia AirScale, Ericsson Radio System). The AI module talks to the base‑band via NETCONF/RESTCONF.
- Edge Compute Node – A low‑power x86_64 server (Intel NUC or AMD Ryzen 7) running Ubuntu 22.04 LTS. Allocate at least 8 GB RAM and 4 CPU cores for the inference engine.
- AI Framework – TensorFlow 2.8 or PyTorch 2.0 with ONNX export support. ZTE’s model is built on a lightweight LSTM for traffic forecasting.
- Telemetry Stack – Prometheus + Grafana for KPI collection, plus a Kafka broker for real‑time event streaming.
- Orchestration – Kubernetes (v1.28) with Kube‑edge to keep the inference pod close to the radio node.
A minimal proof‑of‑concept can be assembled with the open‑source OpenRAN‑Gym toolkit, which already includes traffic‑aware power‑control policies.
Power‑budget comparison with a conventional RAN
| Scenario | Avg. cell power (W) | Avg. traffic (Mbps) | Energy per GB (Wh/GB) |
|---|---|---|---|
| Traditional static RAN | 1 200 | 30 | 40 |
| AI‑enabled adaptive RAN | 1 050 | 30 | 35 |
The adaptive RAN saves roughly 12 % of the energy per gigabyte transferred, aligning with the network‑wide 10.6 % uplift reported by Ucell.
What this means for the region
Uzbekistan’s mobile data demand is projected to grow at 12 % CAGR through 2030. By embedding AI‑driven power control now, Ucell can accommodate that growth without a proportional rise in electricity bills or carbon footprint. The rollout also provides a template for neighboring operators in Central Asia, where grid reliability and energy cost are recurring challenges.
Where to follow the rollout
- Official ZTE announcement: https://www.zte.com.cn/en/press/2026/05/26/uzbekistan‑green‑network
- Ucell corporate news page: https://www.ucell.uz/en/news/ai‑green‑network
- Technical deep‑dive PDF (download): https://www.zte.com.cn/en/resources/whitepapers/AI‑RAN‑Energy‑Saving.pdf
Bottom line: The Ucell‑ZTE partnership demonstrates that AI can be folded into existing RAN hardware to shave double‑digit percentages off energy use while keeping latency and throughput flat. For anyone building a testbed or planning a green‑network upgrade, the architecture is now proven at scale and the component list is openly documented.
Comments
Please log in or register to join the discussion