NVIDIA: Open AI Models & Blueprints Power Autonomous Telecom Networks

Autonomous Networks Gain Ground with NVIDIA’s New Tools

The shift toward intelligent, self-managing telecommunications networks – known as autonomous networks – is accelerating, with telecom operators increasingly prioritizing investment in this area. A recent NVIDIA report highlighted network automation as the top AI utilize case for return on investment, signaling a move beyond simply automating predefined tasks to building networks capable of reasoning and independent decision-making. NVIDIA is responding with new tools, including an open large telco model (LTM) based on Nemotron, blueprints for energy savings, and multi-agent orchestration capabilities, unveiled ahead of Mobile World Congress Barcelona.

Beyond Automation: The Need for Reasoning

While automation executes pre-set workflows, true autonomy requires networks to understand operator intent, weigh different options, and decide on the best course of action. This necessitates reasoning models and AI agents specifically trained on telecom data. NVIDIA’s approach focuses on creating an complete-to-end agentic system, encompassing telco network models, AI agents, and network simulation tools for validating actions before implementation.

Open Nemotron 3 Large Telco Model: Understanding the Language of Telecom

A key component of this push is the release of an open-source, 30-billion-parameter NVIDIA Nemotron LTM. Developed in collaboration with AdaptKey AI, this model is designed to understand the specific terminology and workflows within the telecommunications industry. Built upon the NVIDIA Nemotron 3 family of foundation models, the LTM was fine-tuned using both industry standards and synthetically generated telecom datasets. This optimization allows it to reason through tasks like fault isolation, remediation planning, and change validation.

The open-source nature of the Nemotron LTM is a significant aspect. It provides telcos with full transparency into the training data and processes, enabling secure, on-premises deployment and the ability to adapt and extend the model with their own network and operational data. This control is crucial for maintaining data security and avoiding vendor lock-in.

Teaching AI to Think Like a Network Engineer

NVIDIA, working with Tech Mahindra, has published a guide detailing how to fine-tune domain-specific reasoning models and build agents capable of executing network operations center (NOC) workflows safely. The core principle outlined in the guide is to focus on high-impact, frequent incident categories and translate expert resolutions into step-by-step procedures. These procedures are then converted into structured reasoning traces, capturing each action, tool used, outcome, and decision. These traces serve as “thinking examples” for the model, enabling it to understand not just *what* to do, but *why* a particular sequence of actions is effective, and safe.

Leveraging the NVIDIA NeMo-Skills pipeline, operators can fine-tune reasoning models on these traces, creating specialized AI agents that can solve problems with a level of expertise comparable to a human network engineer.

Energy Efficiency and Network Configuration: Practical Blueprints

NVIDIA is also offering practical blueprints for applying these advancements. The new Blueprint for intent-driven RAN energy efficiency aims to reduce power consumption in 5G radio access networks (RAN) without compromising service quality. This blueprint integrates network test and measurement technology from VIAVI, using synthetic network data to generate energy-saving policies. These policies are then simulated to validate their effectiveness before being implemented in live networks.

the NVIDIA Blueprint for telco network configuration is gaining traction with operators globally. Cassava Technologies is utilizing the blueprint to build Cassava Autonomous Network, a platform designed to optimize mobile networks across Africa. This platform employs three agents: one for monitoring and recommending changes, one for applying those changes with proper documentation, and one for assessing impact and rolling back changes if necessary.

NTT DATA is implementing the blueprint to intelligently regulate traffic, managing surges during network outages, and is currently deploying it with a major operator in Japan. The system uses an AI agent to analyze real-time demand and dynamically adjust user admission policies, optimizing network resilience.

Multi-Agent Orchestration for Complex Workflows

To manage the complexity of agentic workflows across RANs, NVIDIA and BubbleRAN are enhancing the network configuration blueprint with the NVIDIA NeMo Agent Toolkit (NAT) and BubbleRAN Agentic Toolkit (BAT). These complementary frameworks facilitate multi-agent orchestration, allowing for more flexible management of network monitoring, configuration, and validation agents. BubbleRAN’s Opti-Sphere platform integrates NAT and BAT to connect agents to tools that provide network metrics and traffic status, enabling continuous proposal and validation of configuration changes.

Telenor Group will be the first telco to adopt this enhanced blueprint with BubbleRAN, aiming to improve its 5G network for Telenor Maritime, its global connectivity provider for maritime services.

These developments represent a significant step toward realizing the potential of autonomous networks, offering telcos the opportunity to improve efficiency, reduce costs, and enhance network performance. The open-source approach and focus on practical blueprints are likely to accelerate adoption and innovation in this rapidly evolving field. Further advancements will likely focus on refining reasoning models, expanding agent capabilities, and developing more sophisticated orchestration frameworks to manage increasingly complex network environments. The launch of these tools at Mobile World Congress underscores the growing importance of AI and automation in the future of telecommunications.