Telcos are known for their heavy energy use, but so is AI. Could one, ironically, help the other?
Telecom operators rank among the most power-hungry companies on the planet, consuming roughly 1-2% of global electricity demand. As 5G networks expand and mobile data traffic continues its upward trajectory, that consumption keeps climbing. The GSMA has pushed telecom operators toward Net Zero commitments by 2050 — a deadline that’s approaching fast while the industry’s appetite for electricity shows no signs of slowing down.
Sustainability has, at least nominally, moved from afterthought to a little more important for some major telecom operators. Whether this reflects genuine commitment or clever positioning is harder to say. The tech industry’s track record on environmental promises doesn’t necessarily inspire confidence. That said, the economics might finally be pushing in the same direction as the environmental imperatives. Energy costs are squeezing margins, and regulators are paying closer attention to corporate carbon footprints than ever before.
AI has entered the conversation as a potential solution, with promises to optimize network operations and cut consumption without degrading performance. Will these efficiency gains actually show up at scale, or will AI just layer additional energy demand onto an already power-intensive industry?
RAN Optimization
The Radio Access Network sits at the center of any serious energy reduction effort, typically accounting for around 75%of the power consumed at a mobile site. The inefficiency is hard to ignore. Traffic volume can plummet between peak hours and the middle of the night, yet energy consumption only drops a little. Cell towers and base stations keep drawing power whether they’re managing thousands of simultaneous connections or sitting mostly idle at 3 AM.
AI systems can monitor traffic patterns in real time and automatically shift different parts of the RAN into sleep mode, or shut them down completely, when demand falls off. Reported power savings range from 6-7% on the conservative end to as high as 20%, depending on how the systems are implemented and how aggressively operators are willing to deploy them.
Operators have historically been reluctant to put cell towers into sleep mode. The fear of service disruptions and angry customers has kept many playing it safe. According to Ladan Pickering, Principal System Architect at 1Finity, a Fujitsu company, modern AI is changing that calculus: “Network resources can be safely and efficiently managed by leveraging the latest AI-driven data analysis, optimizing network configuration parameters for automated maintenance and orchestration. This enables improved energy efficiency with increased performance, without impacting quality of experience.”
The technical implementation matters more than it might seem. Pickering points out that “in the typical RAN, radios consume the majority of power, despite the fact they are not fully utilized 24/7. Building and using AI specialized models specific to different segments of the RAN can optimize power savings and reduce power consumption substantially.” These targeted machine learning applications look nothing like the energy-hungry large language models dominating AI headlines — a distinction that matters enormously when asking whether AI-driven efficiency gains can outpace the energy costs of running AI systems in the first place.
Dynamic Energy Management
AI-powered platforms are finding applications beyond the RAN, managing energy consumption across broader telecom infrastructure. Cloud-native AI systems can adjust the power state of servers based on real-time workload conditions, scaling capacity to match actual demand instead of maintaining constant readiness for peak loads that may never materialize.
Predictive maintenance could also help with energy reduction, though it rarely gets much attention in green AI discussions. By anticipating network outages before they happen, AI systems could enable proactive maintenance that eliminates unnecessary truck rolls — dispatching service vehicles to check on or repair equipment. Every avoided truck roll means fuel saved and emissions avoided, and those savings compound across networks dealing with millions of potential failure points.
Renewable energy integration, of course, could play a significant role. The core challenge with wind and solar is their inherent variability, but that could become more manageable when AI systems can forecast energy demand and schedule operations to align with renewable supply. This capability could prove critical as operators face growing pressure to show real progress on decarbonization.
Cooling operations and water consumption present additional optimization targets. Data centers and network facilities need substantial cooling to keep equipment within operating temperatures, and AI can consolidate network functions into fewer locations while automatically tuning cooling systems to reduce water use. Whether operators will prioritize these less visible efficiency gains remains uncertain, especially when water savings don’t hit the bottom line as directly as electricity reductions.
Documented Savings
Early deployment results offer both encouragement and grounds for skepticism. Pilots have shown energy savings in the 6-10% range, with some vendors claiming potential reductions up to 25% with no impact on customer experience.
Pickering points to more verifiable outcomes: “Today’s intelligent applications are able to leverage user equipment data and traffic estimates powered by AI and ML to switch network capacity on or off as needed while maintaining service continuity, demonstrating confirmed power savings of more than 20 percent compared to conventional methods of estimating communication traffic for individual base stations.”
Financial structures are adapting to support these investments. Some telecom markets are tapping green bonds and sustainable financing vehicles to fund AI-driven efficiency upgrades. These instruments help operators invest in new technology while signaling sustainability commitments, though it’s worth noting that green bonds generate favorable publicity whether or not the promised environmental benefits actually materialize.
The business case is becoming harder to dismiss though, environmental considerations aside. Younger consumers increasingly prefer eco-friendly telecom options, and investment communities are paying more attention to long-term environmental impact. When sustainability starts aligning with shareholder value, even skeptics might expect something meaningful to happen.
Conclusions
Deploying AI-driven efficiency measures demands substantial upfront capital — a barrier that has consistently slowed adoption of technologies promising long-term savings at the expense of immediate spending. Telecom operators, already facing intense competitive pressure and often carrying significant debt, may be slow to prioritize investments with multi-year payback horizons.
Measuring the full lifecycle environmental impact of these systems remains a challenge, of course. The energy and materials consumed in manufacturing AI hardware, rolling out systems across networks, and eventually retiring obsolete equipment all need to factor into any honest assessment of net environmental benefit. Robust metrics and transparent reporting are still very much works in progress.
Most critically, any conversation about “Green AI” needs to separate AI systems that reduce energy consumption elsewhere from the energy burden of AI infrastructure itself. According to Pickering, “unlike public AI applications like chatbots, where high levels of frequent inferencing can consume as much energy as building the models, the quantity of inferencing in mobile network applications is far less and will consume dramatically less energy than building the models.” The specialized, narrow AI applications driving telecom efficiency gains are fundamentally different from the massive general-purpose models grabbing headlines.
Operators that delay green transformation face mounting competitive risks and regulatory pressure. Whether this translates to genuine progress or just more sophisticated greenwashing depends largely on forces outside the industry’s direct control, like carbon pricing mechanisms, regulatory enforcement, and whether consumers and investors keep the pressure on. The technology to meaningfully reduce telecom energy consumption increasingly exists. The question is whether an industry built around growth will embrace the discipline required to actually use it.
