Artificial intelligence is being deployed across telecommunications networks to cut energy consumption by double digits while keeping service speeds intact. Two major deployments show how machine learning can optimize power use in real time, delivering measurable environmental and cost benefits without compromising user experience. This emerging approach suggests that the solution to AI's energy hunger may lie not in building more power plants, but in making existing infrastructure smarter.

What Does AI-Powered Network Optimization Actually Do?

Ucell, one of Uzbekistan's largest mobile operators, recently completed a full commercial rollout of ZTE's RAN AI-powered energy-saving solution across its entire network. The system uses machine learning to analyze real-time traffic patterns and automatically adjust power consumption at individual cell sites. When traffic is light, the network enters energy-saving modes; when demand spikes, it scales back up seamlessly.

The results are concrete. The deployment increased Ucell's energy efficiency ratio by 10.6%, meaning the network now delivers 10.6% more data for every kilowatt-hour of electricity consumed. For a telecom operator managing thousands of base stations, that translates to significant reductions in both carbon emissions and operational costs.

The system works through a two-layer intelligence architecture. Network-level AI forecasts traffic patterns and orchestrates strategy, while base station-level AI executes decisions in real time and monitors performance. The technology implements multiple energy-saving techniques simultaneously, including symbol-level, channel-level, carrier-level, and equipment-level shutdowns, all continuously optimized through machine learning.

How Are Companies Ensuring AI Efficiency Doesn't Sacrifice Performance?

A critical challenge in energy optimization is avoiding the trap of saving power at the expense of user experience. Ucell's solution addresses this through continuous real-time monitoring. The system evaluates key performance indicators before, during, and after each energy-saving action, and automatically exits energy-saving mode if any performance thresholds are exceeded. This safeguard ensures that users never notice the optimization happening behind the scenes.

The approach reflects a broader industry shift toward what might be called "intelligent efficiency." Rather than simply turning off equipment or reducing capacity, these systems make granular, data-driven decisions about where and when to save power. The result is an optimal balance between environmental responsibility and service quality.

Steps to Deploy AI-Powered Energy Efficiency in Data Centers

• Assess Current Infrastructure: Evaluate existing cooling systems, power distribution, and workload patterns to identify where AI optimization can have the greatest impact on energy consumption.
• Implement Liquid Cooling Solutions: Deploy advanced cooling technologies like liquid cooling for high-density compute environments, which can significantly reduce energy waste compared to traditional air cooling methods.
• Integrate Renewable Energy Sources: Pair AI infrastructure with 100% renewable energy sources and design systems to capture and repurpose excess heat for secondary uses, such as district heating for residential areas.
• Monitor and Continuously Optimize: Use machine learning to continuously analyze performance metrics and automatically adjust resource allocation in real time, ensuring efficiency gains persist as workloads change.

What's Happening in the Broader AI Infrastructure Market?

The energy efficiency trend extends beyond telecommunications. Verda, a European AI cloud provider, recently partnered with Supermicro to deploy liquid-cooled NVIDIA Blackwell-accelerated systems across its data centers. Blackwell is NVIDIA's latest generation of graphics processing units (GPUs) designed for training large language models and running other compute-intensive AI workloads.

Verda's infrastructure operates on 100% renewable energy and is collaborating with local utilities to repurpose excess heat from its data centers to support residential heating for up to 15,000 homes. This approach transforms data centers from pure energy consumers into contributors to local heating infrastructure, a model that could reshape how we think about the environmental footprint of AI compute.

The deployment includes a range of advanced systems, including NVIDIA GB300 NVL72 rack-scale systems, NVIDIA HGX B300 and NVIDIA HGX B200 systems, and NVIDIA RTX PRO 6000 Blackwell Server Edition-accelerated systems. Supermicro's pre-tested and validated configurations enabled Verda to accelerate deployment timelines while reducing operational risk and optimizing system performance.

"Our mission is to empower pioneering teams across the globe with AI-native infrastructure. Partnering with Supermicro helps us deliver on that promise at scale," said Ruben Bryon, Founder and CEO of Verda.

Ruben Bryon, Founder and CEO of Verda

Why Does This Matter for the Future of AI?

As AI adoption accelerates globally, energy consumption has become a critical bottleneck. Data centers powering large language models and other AI systems consume enormous amounts of electricity. Rather than waiting for breakthrough technologies like fusion power or photonic chips, these deployments show that intelligent software can deliver meaningful efficiency gains today using existing hardware.

The Ucell deployment demonstrates that even mature infrastructure like cellular networks can be retrofitted with AI optimization. The 10.6% efficiency improvement across an entire national network suggests that similar gains could be replicated across other telecom operators, data center operators, and cloud providers worldwide.

"The 10.6% improvement in energy efficiency ratio demonstrates that AI-driven intelligence can deliver both environmental and economic value without compromising user experience," said Wang Guangdong, CEO of ZTE Uzbekistan.

Wang Guangdong, CEO of ZTE Uzbekistan

For enterprises and cloud providers, the message is clear: energy efficiency is no longer a trade-off with performance. By deploying machine learning to optimize resource allocation in real time, organizations can reduce their carbon footprint, lower operational costs, and maintain or even improve service quality. As AI workloads continue to grow, this approach may prove essential to making AI infrastructure sustainable at scale.