Telecom operators are increasingly turning to artificial intelligence to automate network operations, but true autonomy requires more than just automation—it demands reasoning and decision-making capabilities.
NVIDIA is addressing this need with a new open-source large telco model (LTM) based on its Nemotron 3 foundation models. Developed in collaboration with AdaptKey AI, the LTM is fine-tuned on telecom datasets to understand industry terminology and reason through workflows like fault isolation and remediation planning. Unlike traditional automation tools, this model can interpret operator intent, weigh tradeoffs, and execute actions without predefined scripts.
The LTM joins a suite of new NVIDIA Blueprints designed for energy efficiency and network configuration. These blueprints integrate multi-agent systems to simulate and validate decisions before they are applied in live networks, ensuring safety and performance. One such blueprint, developed with VIAVI, focuses on reducing power consumption in 5G radio access networks (RAN) by generating synthetic data to test energy-saving policies without disrupting service.
NVIDIA is also releasing an open-source guide for fine-tuning reasoning models, built on the NeMo-Skills pipeline. This framework helps operators train AI agents to mimic network engineers’ problem-solving processes, using structured traces of actions and outcomes as learning examples. The goal is to create telco-specialized AI that can handle high-impact incidents with precision.
These tools are being adopted by major telecom players, including Cassava Technologies, which uses the network configuration blueprint to optimize Africa’s multi-vendor mobile networks, and NTT DATA, which deploys it in Japan for traffic regulation. BubbleRAN is enhancing the blueprint with its Agentic Toolkit (BAT) and NVIDIA’s NeMo Agent Toolkit (NAT) to improve multi-agent orchestration, with Telenor Group set to be the first adopter.
NVIDIA’s advancements are part of a broader push toward autonomous networks, where AI agents work collaboratively to manage complex telecom operations. While the technology is still evolving, these open-source resources aim to accelerate adoption by providing telcos with the flexibility to customize models for their specific needs—without compromising data control or security.
