VIDIA introduced a new open large telco model and a set of agentic AI blueprints aimed at accelerating autonomous network operations, as telecom operators increase investment in AI-driven automation. In its latest State of AI in Telecommunications report, NVIDIA identified network automation as the top AI use case for both investment and return. The company outlined a shift from basic automation toward autonomous networks capable of reasoning over operator intent, evaluating tradeoffs, and executing validated actions across complex telecom environments.
At Mobile World Congress Barcelona, NVIDIA unveiled an open 30-billion-parameter large telco model (LTM) based on its Nemotron 3 foundation model family, along with implementation guidance for reasoning agents and new NVIDIA Blueprints for RAN energy efficiency and network configuration. The initiatives align with the GSMA’s Open Telco AI effort, under which NVIDIA is releasing the model, implementation guide, and blueprints as open resources. The company collaborated with AdaptKey AI to fine-tune the Nemotron-based LTM using open telecom datasets, industry standards, and synthetic logs, optimizing the model to understand telecom terminology and reason through workflows such as fault isolation, remediation planning, and change validation.
NVIDIA also worked with Tech Mahindra on an open guide that shows operators how to fine-tune domain-specific reasoning models using structured “reasoning traces” derived from network operations center (NOC) procedures. Using the NVIDIA NeMo-Skills pipeline, operators can train models to replicate the step-by-step logic of experienced network engineers. New blueprints further extend this approach to closed-loop RAN energy optimization and multi-agent network configuration, integrating tools such as VIAVI’s TeraVM AI RAN Scenario Generator to simulate traffic patterns and validate policy changes before deployment in live networks.
• Open 30B-parameter Nemotron-based Large Telco Model, fine-tuned on telecom datasets and synthetic logs
• Designed for workflows including fault isolation, remediation planning, and configuration validation
• On-premises deployment support with full transparency into training data and tuning
• Open implementation guide (with Tech Mahindra) for building reasoning agents trained on structured NOC procedures
• Blueprint for intent-driven RAN energy efficiency using closed-loop simulation with VIAVI AI RSG
• Blueprint for multi-agent network configuration adopted by Cassava Technologies and NTT DATA
• Multi-agent orchestration enhancements using NVIDIA NeMo Agent Toolkit and BubbleRAN Agentic Toolkit
• Initial deployment by Telenor Group for maritime 5G connectivity via Telenor Maritime
Cassava Technologies is using the network configuration blueprint to build its Cassava Autonomous Network platform, which deploys monitoring, configuration, and validation agents to manage multi-vendor environments across Africa. NTT DATA is implementing the framework with a Tier 1 operator in Japan to automate traffic regulation during demand surges. Telenor Group will adopt the enhanced blueprint with BubbleRAN to support 5G services for maritime connectivity.
“By combining open telecom reasoning models with multi-agent blueprints, we’re giving operators the tools to safely train AI agents on their own data and move toward fully autonomous networks,” said NVIDIA.
🌐 Analysis: NVIDIA’s move extends its AI infrastructure strategy beyond data centers into telecom operations, positioning Nemotron models as domain-specialized reasoning engines for vertical markets. The open model approach, aligned with GSMA’s Open Telco AI initiative, may accelerate ecosystem adoption as operators seek alternatives to proprietary AI stacks while competitors such as Ericsson, Nokia, and hyperscalers advance AI-native RAN and network automation frameworks.
