NVIDIA Launches Agentic AI Blueprint and Telecom Inference Model to Advance Autonomous Networks

Autonomous networks, which possess intelligent and self-management capabilities, are shifting from a future vision to a top strategic priority for current telecom operators. In NVIDIA’s latest report, “Current Status and Trends in AI Development for the Telecom Industry,” network automation has become the AI application scenario with the highest return on investment and the most attention.

It is important to emphasize that automation and autonomy are not the same concept. Autonomous networks can execute predefined workflows, but they also need to understand operator intent, weigh various factors, and make autonomous decisions about what actions to take. The key to achieving this transformation lies in reasoning models and AI agents fine-tuned based on telecom data.

To realize true network autonomy, an end-to-end intelligent agent system must be built, including key components such as telecom network models, communicative AI agents, and network simulation tools for validating operational effects.

On the eve of the Mobile World Congress (MWC 2026) in Barcelona, NVIDIA released an open large-scale telecom model based on NVIDIA Nemotron (LTM), a comprehensive guide for building network operation reasoning agents, and a new NVIDIA Blueprint for energy saving and network configuration. This blueprint, through multi-agent orchestration, helps operators move toward autonomous operation.

Additionally, as part of the GSMA’s newly launched Open Telco AI initiative, NVIDIA will release this new LTM model, implementation guide, and agent-based AI blueprint in open source format through the authoritative mobile industry organization GSMA for industry-wide use.

Open Nemotron 3 Large-Scale Telecom Model Introduces Reasoning Capabilities to the Telecom Field

To successfully deploy generative AI and agent-based AI at scale in telecom operations, AI models must understand the specialized language of telecom and perform logical reasoning on complex workflows. To this end, NVIDIA partnered with AdaptKey AI to launch a new open-source (300109) NVIDIA Nemotron LTM with 30 billion parameters, enabling global operators to build autonomous networks.

This model is based on the NVIDIA Nemotron 3 series foundational models and has been fine-tuned by AdaptKey AI using open telecom datasets, including industry-standard documents and synthetic logs, specifically optimized for understanding telecom terminology and reasoning about workflows such as fault isolation, repair planning, and change validation.

As an open model, Nemotron LTM allows telecom operators to fully understand the training methods and data used, enabling safe and rapid deployment in local network environments. Operators can also securely combine their own network and operational data to safely adjust and extend the model’s telecom reasoning capabilities, steadily progressing toward autonomous operation without sacrificing data control or security.

Teaching AI Agents to Think Like Network Engineers

NVIDIA and Tech Mahindra jointly released an open guide to help telecom operators fine-tune domain-specific reasoning models and build intelligent agents capable of safely executing Network Operations Center (NOC) workflows.

The guide proposes a training framework that enables models to “think like NOC engineers”: focusing on high-impact, frequently occurring fault types; transforming expert solutions into step-by-step operational procedures; and converting these procedures into structured “reasoning trajectories” that clearly document each step, tool invocation, results, and decision basis. These trajectories serve as “thinking examples” for the model to learn from, helping it not only know “what to do” but also understand “why these checks and fixes are safe and effective.”

Using NVIDIA NeMo-Skills workflows, operators can fine-tune models based on these reasoning trajectories, laying the foundation for building telecom-specific AI agents with problem-solving and reasoning abilities similar to network engineers.

Maximizing Energy Efficiency with a New Intent-Driven Blueprint

Autonomous networks operate in a closed loop: understanding network states through models, taking actions based on intent via intelligent agents, and feeding simulation results back into the system to verify and optimize decisions. NVIDIA’s new intent-driven RAN energy efficiency Blueprint integrates these elements to help operators systematically reduce 5G RAN energy consumption while ensuring service quality.

This blueprint incorporates VIAVI’s TeraVM AI RAN scenario generator (AI RSG), a leader in network testing and measurement, to generate synthetic network data such as cell utilization, user throughput, and other traffic patterns, converting them into concise, queryable formats.

Subsequently, an energy planning agent reasons based on this synthetic data to generate energy-saving strategies and performs simulation validation within the AI RSG. This allows operators to safely verify whether energy-saving strategies align with their intent in a closed-loop environment, without changing live network configurations or impacting user services.

Global Operators Begin Deploying NVIDIA Blueprint for Network Configuration

NVIDIA Blueprint for telecom network configuration is being adopted by multiple operators worldwide.

Cassava Technologies is using this blueprint to build the Cassava Autonomous Network platform, designed to optimize Africa’s diverse multi-vendor mobile network environment. The platform deploys three intelligent agents: one monitors the network and recommends configuration changes; another applies changes and automatically generates documentation and governance records; and a third assesses the impact of changes, with safe rollback if unexpected consequences occur.

NTT DATA, leveraging this blueprint, has deployed an intelligent traffic control system for a major Japanese operator. When network disruptions cause large-scale user reconnections and traffic surges, this system helps stabilize the network.

Specifically, an AI agent analyzes real-time network demand across the entire network and decides where, when, and how to admit new users. As network conditions stabilize, the agent dynamically adjusts decisions, transforming previously manual configuration processes into a data-driven optimization cycle, creating a more resilient mobile network.

Evolving Network Configuration with Multi-Agent Orchestration

To assist telecom operators in designing, observing, and optimizing complex agent-based workflows in RAN, NVIDIA is collaborating with BubbleRAN. Using NVIDIA NeMo intelligent agent toolkit (NAT) and BubbleRAN’s agent orchestration framework (BAT), they are enhancing the NVIDIA Blueprint for telecom network configuration.

BubbleRAN is integrating NAT and BAT into its Opti-Sphere platform to enable more flexible, containerized management of network monitoring, configuration, and validation agents, connecting them to tools that report network metrics and traffic status. This continuous process proposes and verifies configuration change suggestions.

Telenor Group will be the first telecom operator to adopt this blueprint in collaboration with BubbleRAN, aiming to improve the performance of its Telenor Maritime (a global maritime connectivity provider) 5G network.