One of the key challenges addressed by the initiative is access to training data. NVIDIA noted that telecom operators often face restrictions on using network and customer information because of privacy concerns. To address this, the company is promoting synthetic data tools including NVIDIA NeMo Safe Synthesizer and NVIDIA NeMo Anonymizer.
SoftBank Corp. is using those technologies to create synthetic datasets that mirror the characteristics of real network performance and configuration data without exposing sensitive information. The company is applying those datasets to fine-tune its telecom-focused AI model and develop specialized network agents.
NVIDIA is also emphasizing the role of long-running AI agents capable of managing complex workflows under operational and regulatory requirements. Its NemoClaw blueprints and OpenShell secure runtime are designed to provide policy controls, governance mechanisms, and restricted access to telecom systems.
Several partners are demonstrating deployments built on those technologies. AdaptKey is testing security-focused agents for self-healing 5G operations that can identify network and security issues before submitting remediation requests through its KeySmith platform. Amdocs is exploring autonomous customer-care applications, including roaming support scenarios, while also using the technology for data-science agents that analyze customer accounts and migration readiness.
NTT DATA is combining NVIDIA Nemotron models with NemoClaw to develop agents that monitor long-term network performance and investigate signs of degradation. ServiceNow is extending its Project Arc platform into telecom operations, enabling agents to manage incident-response workflows by gathering information from multiple systems and coordinating actions under defined governance controls.
Tata Consultancy Services is developing what it describes as a multi-fidelity “AI sensor” architecture that uses long-running agents to identify network issues and trigger more detailed investigations when needed. According to the company, the approach is intended to accelerate the process from anomaly detection to resolution.
Alongside autonomous agents, NVIDIA and its partners are focusing on simulation technologies that allow operators to test AI-generated recommendations before applying changes to live networks. The company argues that GPU-accelerated simulation environments can provide a safer way to validate decisions and reduce operational risk.
Forsk has integrated an AI-based radio propagation model into its Naos RAN planning platform, creating a radio access network digital twin that supports near-real-time network optimization. VIAVI Solutions is using NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs to accelerate large-scale RAN simulations within its TeraVM AI RAN Scenario Generator and has also introduced an IP Network Configuration Blueprint for validating network changes before deployment.
KDDI and KDDI Research are collaborating with NVIDIA, Keysight, and Samsung Research America on a high-fidelity RAN digital twin built with NVIDIA Aerial Omniverse Digital Twin technology and Keysight’s emulation tools. The project is designed to allow multiple autonomous agents to simulate and assess future network scenarios, including traffic changes, radio conditions, and emerging AI-enabled network functions.
Through these demonstrations, NVIDIA and its partners are outlining a framework for telecom operators seeking to expand the use of AI from automation tools to more autonomous operational systems while maintaining governance and control over network decisions.
This analysis is based on reporting from NVIDIA.
Image courtesy of NVIDIA.
This article was generated with AI assistance and reviewed for accuracy and quality.
