NVIDIA Unveils Open 30B Telco Reasoning Model and Agentic AI Blueprints for Autonomous Networks

What NVIDIA announced ahead of MWC

On March 1, 2026, ahead of Mobile World Congress Barcelona, NVIDIA introduced a new package for telecom network autonomy: an open NVIDIA Nemotron-based Large Telco Model (LTM), a practical implementation guide for building reasoning agents in network operations, and updated NVIDIA Blueprints focused on energy optimization and network configuration workflows.

NVIDIA framed the launch as part of GSMA's Open Telco AI initiative, with key assets released as open resources. The strategic point is straightforward: telecom operators need domain-specialized models and orchestrated agents, not only general-purpose chat systems, if they want to automate real network operations safely.

Why the model design is notable

The new LTM is described as a 30-billion-parameter open model developed with AdaptKey AI. NVIDIA says it is built on the Nemotron 3 family and fine-tuned using open telecom datasets, industry standards, and synthetic logs. The target use cases are core NOC tasks such as fault isolation, remediation planning, and change validation, where traceable reasoning and operational reliability matter more than broad consumer-style fluency.

The company also highlights transparency and deployment control: because the model is open, operators can inspect training context and adapt the model with their own operational data for on-premises workflows. That matters for telecom environments where data residency, regulatory boundaries, and uptime requirements often rule out black-box deployment approaches.

From reference architecture to field adoption

In parallel, NVIDIA and Tech Mahindra published an open guide using the NeMo-Skills pipeline to fine-tune reasoning models from structured "thinking traces" created from expert incident procedures. This is intended to teach agents not only what fix to apply, but why a sequence is safe and effective.

• Cassava Technologies is building a three-agent autonomous network workflow for monitoring, applying changes, and rollback assessment.
• NTT DATA is applying the blueprint to traffic regulation scenarios and deploying with a Tier 1 operator in Japan.
• Telenor Group plans to use the blueprint with BubbleRAN for Telenor Maritime 5G operations, with NAT and BAT for multi-agent orchestration.

The broader signal is that telecom AI is shifting from demo-stage copilots to production-grade, auditable agent systems. NVIDIA's open-model and open-blueprint approach is designed to reduce adoption friction while preserving operator control over data, security posture, and rollout pace.