NVIDIA announced at GTC that partners are building large telco models (LTMs) and AI agents specifically for the telecommunications industry.
Telecom networks process an enormous volume of data each day, supporting millions of connections and generating thousands of terabytes every minute. This inflow—covering traffic records, performance metrics, configuration details, and topology information—is complex and often unstructured, presenting significant operational challenges for traditional automation tools.
To address these challenges, vendors are adopting NVIDIA’s NIM and NeMo microservices within the NVIDIA AI Enterprise software platform to build LTMs and AI agents. These solutions are intended to automate decision-making, improve operational efficiency, increase staff productivity, and boost overall network performance.
LTMs are specialised, multimodal large language models trained on telecom-specific network data. They provide a foundation for AI agents that can interpret network states, predict issues, and take corrective actions automatically—helping operators manage complex, real-time network workloads more effectively.
Several industry players have already introduced LTMs and AI agents. SoftBank and Tech Mahindra are among those developing bespoke models and agent-based tools. Companies such as Amdocs, BubbleRAN, and ServiceNow are expanding their network operations and optimisation capabilities with AI agents built on NVIDIA AI Enterprise.
A recent NVIDIA telecom survey found that 40% of respondents are actively rolling out AI within network planning and operations—indicating growing industry adoption.
LTMs: Interpreting the language of networks
Just as large language models understand and generate human language, LTMs are designed to comprehend the “language” of telecom networks. They interpret signals, alerts, and telemetry to provide actionable insights and automated responses.
Key advantages of these LTMs include:
- Specialised network intelligence: LTMs can understand real-time events, anticipate likely failures, and trigger remediation steps.
- Optimised for telecom workloads: Using NVIDIA NIM microservices, LTMs are tuned for efficiency, accuracy, and low latency—qualities essential for telecom environments.
- Continuous learning and adaptation: With post-training scalability through NVIDIA NeMo, LTMs can learn from new incidents, alerts, and anomalies to improve over time.
NVIDIA AI Enterprise also supplies tools and blueprints to accelerate AI agent development. These agents are built to streamline operations, reduce costs, and raise key performance indicators (KPIs) such as:
- Reduced downtime: Predictive capabilities help avoid incidents before they disrupt service, strengthening network resilience.
- Improved customer experience: AI-driven optimisations can yield faster, more reliable connectivity and fewer outages.
- Enhanced security: Continuous monitoring and threat detection allow networks to respond to risks in real time, improving overall security posture.
Telecom giants pioneer LTMs and AI agents with NVIDIA
Major telecom companies are already using NVIDIA AI Enterprise to advance their network automation and assurance efforts.
SoftBank developed an LTM derived from a large-scale base model and trained on its proprietary network data. Initially targeting network configuration, this LTM—delivered as an NVIDIA NIM microservice—can automatically reconfigure network resources to handle traffic surges at large events, such as stadiums or other high-demand venues.
Tech Mahindra built an LTM using NVIDIA’s agentic AI tooling to support network operations. Its Adaptive Network Insights Studio offers comprehensive visibility into network issues, generating automated reports with multiple detail levels to assist IT teams, engineers, and executives. Tech Mahindra’s Proactive Network Anomaly Resolution Hub uses the LTM to automatically resolve a large share of network events, reducing engineers’ manual workload and improving productivity.
Amdocs’ Network Assurance Agent, powered by the amAIz Agents framework, automates repetitive but essential tasks such as fault prediction, impact analysis, and recommended preventative steps. It also provides guided remediation to help engineers address problems efficiently. Amdocs’ Network Deployment Agent streamlines adoption of open radio access network (RAN) technologies by automating integration, deployment, and interoperability testing while supplying actionable insights to network teams.
BubbleRAN is developing an autonomous, cloud-native multi-agent RAN intelligence platform that leverages LTMs to monitor network state, configuration, availability, and KPIs—helping to simplify monitoring and troubleshooting.
ServiceNow’s telecom-focused AI agents, built on NVIDIA AI Enterprise and deployed on NVIDIA DGX Cloud, are designed to improve productivity by automatically generating resolution playbooks and predicting disruptions before they occur. These agents can reduce resolution times, improve customer satisfaction, and analyse incidents to identify root causes and prevent repeat outages.
Through collaboration with partners, NVIDIA is helping to introduce a new generation of intelligent telecom networks. As data demand grows, these AI-driven innovations will be essential for operators to manage increasingly complex infrastructures efficiently and reliably.
(Image credit: NVIDIA)
Want to learn more about AI and big data from industry leaders? Consider attending the AI & Big Data Expo, held in Amsterdam, California, and London, which is co-located with events such as the Intelligent Automation Conference, BlockX, Digital Transformation Week, and Cyber Security & Cloud Expo.
Explore other upcoming enterprise technology events and webinars hosted by TechForge.