Rakuten Mobile reaches Level 4 autonomy for RAN energy efficiency

Tens of thousands of base stations, scattered across diverse landscapes; operating a national mobile network has historically been a mammoth undertaking of physical infrastructure. The task of maintaining peak performance to meet the fluctuating demands of users typically requires an army of engineers across the country, operating around the clock.

As the industry prepares for 6G and beyond, the complexity of mobile networks has begun to outpace the capacity of traditional human management. Forward-looking carriers are seeking ways to make their networks self-aware and capable of managing their own health and energy consumption.

In February 2026, Rakuten Mobile announced a significant breakthrough on this journey: Japan’s newest carrier has officially achieved Level 4 autonomy for its Radio Access Network (RAN) energy efficiency optimization.

This milestone represents a shift from a network that only follows human instructions, to one that thinks for itself. Leveraging advanced AI, Rakuten Mobile has moved beyond simple automation to create a system that can independently adjust power consumption based on real-time traffic, achieving a 20% reduction in energy usage.

How can a network be autonomous?

What is the significance of Level 4 autonomy? TM Forum, the global industry alliance for telco and tech companies, defines a six-stage path (Level 0-5) toward full self-driving networks:

• Level 0-1: Manual operations or basic, isolated task automation using simple scripts.
• Level 2: Process automation where multiple tasks are chained into workflows (though humans still handle all exceptions).
• Level 3: Orchestrated, proactive automation that can detect and resolve known issues (but still requires human lead for novel situations).
• Level 4: Cognitive automation where the network uses AI and machine learning to adapt to unknown conditions and optimize itself independently.
• Level 5: The system has closed-loop automation capabilities across multiple services, multiple domains and the entire lifecycle via cognitive self-adaptation.

While many carriers already use software to automate repetitive tasks, these systems are generally rules-based and can only handle scenarios that a human engineer has already anticipated and coded for. Level 4 autonomy, meanwhile, introduces the ability for the network to adapt to the unknown.

Andy Tiller, EVP of Member Products at TM Forum, says that it’s important to distinguish between traditional network automation and true autonomy.

“Autonomy is a measure of how far the network can run without human intervention – for example, diagnosing and fixing faults, or tuning the network to find the best balance between performance and energy consumption,” he explains. “This can be done with rules-based automation where humans anticipate what might happen and then code sophisticated rules to deal with these situations, but Level 4 autonomy can handle situations that humans have not anticipated – this requires AI.”

This means not just new software, but a fundamental change in the role of human engineers. “Level 4 also requires a mindset change for telecoms professionals; they need to get good at tuning and optimizing the AI, rather than tuning and optimizing the network directly.”

Precision efficiency in a live environment

For Rakuten Mobile, this level of autonomy was made possible by its embrace of Open RAN architecture. The traditional closed networks employed by most carriers often create data silos that AI can’t peer into. Rakuten Mobile’s network architecture is open, cloud-native, and entirely software-driven – allowing AI to sit at the very heart of operations.

Rakuten Mobile’s Director of Network Operations, Shailesh Gupta, oversees a team of 400 engineers tasked with managing this infrastructure. He describes the jump to Level 4 as a transition from proof of concept to a rigorous operational discipline.

“We have developed software called the Rakuten Intelligent Controller,” he explains. “This is based on AI and machine learning models where we have different types of use cases to save energy.”

The system works dynamically to allocate the appropriate amount of power to different parts of the network at any given moment.

“To give a few examples: if customers are not at the site or not using the network – perhaps at night – we switch off the power. If we are radiating [radio frequency signals] on a high level but the customer is not there, we can reduce the power to save energy. All of these different use cases together have resulted in energy savings of 20%,” Gupta continues. “This is all AI-based; it is not a static rule. If a customer comes online, the power is immediately on. If traffic goes to zero, it adjusts.”

Gupta highlights three core principles that allow the network to maintain this level of precision. The first is predictive learning, in which the AI analyzer platform looks at user mobility and traffic loads to anticipate issues before they occur.

The second involves deep data analysis to uncover hidden network patterns, and the third is a real-time safety net to detect anomalies.

“We have an anomaly detection system that acts as our safety net, tracking performance metrics like latency, throughput and session stability in real time, so that any anomaly is immediately detected and corrected before it gives a bad impression to the customer.”

Balancing performance with sustainability

As mobile network traffic continues to grow with the proliferation of 5G, autonomous energy optimization is becoming less of a neat technical feat, and more of a core necessity to mitigate the industry’s carbon footprint and manage rising energy costs. TM Forum’s Andy Tiller sees autonomous networks as the only viable solution to this growth.

“Traffic on our communications networks is growing rapidly – much of it now generated by AI workloads. This growth will only accelerate as robots, cars, drones and other connected physical AI becomes prevalent,” he warns. “The corresponding increase in energy consumption and carbon footprint could have profound consequences for the planet, so it’s fitting that AI can help address the challenge it has created, reducing energy consumption by tuning networks far better than any human can.”

On the user end, the benefits of an autonomous network are felt through consistency. Gupta notes that a network that can self-heal and self-optimize is a network that can stay online.

“It ensures better network uptime, more stable performance and better overall quality,” he lists. “We ensure there is no customer impact even while saving these energy costs. Issues are always part and parcel of the network, but Level 4 automation changes how the issues are handled. Instead of reacting after the customer complains, the network can detect, predict and resolve problems before they occur.”

Cost, too, is another area Rakuten Mobile subscribers are feeling the benefits – something Rakuten Group Chairman and CEO Mickey Mikitani highlighted in an announcement late last year that the carrier would double down on customer value.

“From the cost perspective, Rakuten Mobile’s network cost is one of the lowest in the industry,” Gupta adds. “Whatever costs we are ultimately saving, we are passing on to the customer.”

The road to Level 5

While most wireless carriers remain at Level 2 or 3, Rakuten Mobile’s Level 4 validation puts it in an advantageous position.

“There are other carriers pursuing AI deployment aggressively, but Rakuten Mobile makes its mark because we are the first player in Japan to get a Level 4 autonomous certification,” Gupta says. “Globally, a few operators got this certification before us, but in the Open RAN space worldwide, we are the first. This is where we are really competitive; efficiency becomes a core pillar whether it is optimizing energy consumption or minimizing operational overhead.”

Even with Level 4 achieved, however, Rakuten’s journey toward a completely autonomous network is far from complete. Having proven the concept in RAN energy efficiency, the company is now focused on expanding that same level of intelligence across every domain of its network, from the core to the edge.

“We have achieved Level 4 in one domain, and we are working with TM Forum to achieve it in others.” Gupta explains. “Our vision is to create a network that runs itself end-to-end: self-optimizing, self-healing and self-decisive.”

Level 5 automation, with no human intervention required, remains the industry’s north star, and one that no carrier has yet reached. But with its cloud-native architecture and AI already making real-time decisions at scale, Rakuten Mobile may be better positioned than most to get there first.