NVIDIA Unveils Open Telco AI Model

Open-source telco AI model, a practical implementation guide and multi-agent blueprints released via GSMA’s Open Telco AI aim to help operators train domain-specific agents, reduce energy use and automate network configuration.

Autonomous networks — intelligent, self-managing telecoms operations — are shifting from long-term aspiration to near-term priority. According to NVIDIA’s latest State of AI in Telecommunications report, network automation ranks as the leading AI use case for both investment and return on investment.

True autonomy goes beyond scripted automation: systems must grasp operator intent, weigh trade-offs and choose actions. Achieving that requires an end-to-end agentic approach combining telco-tuned reasoning models, collaborating AI agents and simulation to verify outcomes before changes are applied.

Ahead of Mobile World Congress Barcelona, NVIDIA introduced an open Nemotron-based Large Telco Model (LTM), a detailed playbook for building reasoning agents for network operations, and new NVIDIA Blueprints covering energy saving and network configuration with multi-agent orchestration. As part of GSMA’s new Open Telco AI initiative — launching tomorrow — NVIDIA will release the open-source LTM, the implementation guide and the agentic AI blueprints as open resources through the industry body.

Developed with AdaptKey AI, the 30‑billion‑parameter Nemotron 3 LTM is fine-tuned on open telecom datasets — including industry standards and synthetic logs — to understand sector terminology and reason through tasks such as fault isolation, remediation planning and change validation. As an open model, it provides transparency into training methods and data sources, supports secure, rapid on‑premises deployment, and enables operators to extend telco‑specific reasoning using their own network and operational data without compromising control or security.

In parallel, NVIDIA and Tech Mahindra have published an open-source guide showing how operators can fine-tune domain-specific reasoning models and build agents that safely carry out network operations centre (NOC) workflows. The framework prioritises high‑impact, high‑frequency incident types, converts expert fixes into step‑by‑step procedures and records structured reasoning traces for each action, tool call, outcome and decision — examples the model learns from to understand both what to do and why. Using the NVIDIA NeMo‑Skills pipeline, teams can fine-tune a reasoning model on these traces to form the basis of telco-specialised agents.

To tackle energy consumption, the new NVIDIA Blueprint for intent-driven RAN energy efficiency links models that understand the network with agents that act on operator intent and a simulator that feeds results back in a closed loop. It integrates VIAVI’s TeraVM AI RAN Scenario Generator (AI RSG) to produce synthetic data — such as cell utilisation and user throughput — in a simple, queryable format. An energy-planning agent then reasons over this data to propose energy‑saving policies, which are validated in AI RSG so operators can meet objectives without altering live configurations or affecting subscribers.

Operators are already adopting the NVIDIA Blueprint for telco network configuration. Cassava Technologies is using it to build Cassava Autonomous Network, an agentic platform tailored to Africa’s heterogeneous, multi‑vendor mobile landscape. Its three-agent design monitors the network and recommends changes, applies approved updates with documentation and governance, and evaluates the impact — rolling back safely if needed.

NTT DATA is applying the blueprint to add intelligence to traffic regulation, helping a tier 1 operator in Japan manage surges when users reconnect after outages. An AI agent assesses real-time demand across cells to decide when and how to admit new users, then adapts as conditions stabilise — turning manual tuning into a data-led optimisation cycle that strengthens network resilience.

NVIDIA and BubbleRAN are also extending the configuration blueprint with the NVIDIA NeMo Agent Toolkit (NAT) and the BubbleRAN Agentic Toolkit (BAT) to orchestrate complex multi‑agent workflows across the RAN. BubbleRAN is integrating NAT and BAT into its Opti‑Sphere platform to flexibly manage monitoring, configuration and validation agents across containers and workloads, and to connect them to tools reporting network metrics and traffic status so they can continuously propose and verify configuration changes.

Telenor Group will be the first operator to adopt the blueprint with BubbleRAN to enhance its 5G network for Telenor Maritime, the group’s global provider of connectivity at sea.

Further details on these agentic AI advances for telecoms will be showcased at Mobile World Congress in Barcelona from 2–5 March. A software product information notice is available on NVIDIA’s website.

www.nvidia.com/en-us/