The rising number of connected devices and data demands are putting a significant strain on network infrastructure. AI-native networks, especially those using agentic AI, are addressing these challenges by implementing fundamental architectural changes. Companies like SoftBank and Tech Mahindra have developed new LTMs and AI agents, while Amdocs, BubbleRAN, and ServiceNow are enhancing their network operations and optimization with new AI agents, all by utilizing NVIDIA AI Enterprise – https://blogs.nvidia.com/blog/telecom-agentic-ai-for-network-operations/
(Image Credit – Nvidia blog)
The Scale of Modern Telecommunications
Modern telecom networks manage an unprecedented scale of operations, handling millions of daily user connections and processing over 3,800 terabytes of data per minute. This massive data flow comes from diverse sources including base stations, routers, switches, and data centers, creating a complex ecosystem of unstructured information that traditional automation tools struggle to manage effectively.
At the GTC global AI conference, NVIDIA announced a significant advancement in telecommunications AI: industry partners are developing specialized Large Telco Models (LTMs) and AI agents specifically for telecommunications. These solutions are built using NVIDIA NIM and NeMo microservices within the NVIDIA AI Enterprise software platform, representing a new generation of AI-powered network operations.
Technical Architecture
Large Telco Models (LTMs) are AI systems designed to process and analyze telecommunications network data. Similar to how language models interpret text, LTMs interpret network signals, events, and metrics.
These models, developed using NVIDIA’s AI tools, offer several key functions:
- Network Analysis: LTMs can interpret real-time network events, predict potential failures, and suggest or implement solutions.
- Performance Optimization: By utilizing specialized microservices, these models are designed to handle telecom-specific tasks efficiently and with low latency.
- Adaptive Learning: LTMs can update their knowledge based on new network events and anomalies, continuously improving their performance.
Practical Applications
When implemented in network operations, LTMs can contribute to:
- Minimizing Network Downtime: By predicting potential failures before they occur.
- Improving Service Quality: Through faster problem resolution and network optimization.
- Enhancing Network Security: By continuously monitoring for and responding to potential threats.
These AI models represent an advancement in network management technology, providing telecommunications companies with new tools for network optimization and service improvement. The integration of LTMs into network operations allows for more data-driven and automated management of telecommunications infrastructure.
Industry Adoption
As announced at GTC 2025, leading telecommunications companies are already implementing these technologies:
- SoftBank and Tech Mahindra have developed new LTMs and AI agents
- Amdocs, BubbleRAN, and ServiceNow are deploying AI agents for network operations
- 40% of telecommunications companies surveyed are integrating AI into their network planning and operations
This transformation represents a significant shift in how telecommunications networks are managed and optimized, moving from traditional automation to AI-driven intelligent operations.
What are the key Telco problems that Agentic AI can solve
Network Complexity with Agentic AI
- Current networks struggle with device density and diverse traffic patterns: Network operations centers currently manage millions of concurrent connections with varying QoS requirements. Agentic AI systems can autonomously negotiate and optimize these connections based on learned behaviors and goals.
- Traditional AI systems cannot handle complex decision-making: While basic AI can handle predefined scenarios, agentic AI enables autonomous decision-making across multiple network domains, adapting to new situations without human intervention.
- Agentic AI enables self-organizing network management: Autonomous agents collaborate to optimize network resources, learning from each other’s experiences and adapting their strategies based on network performance goals.
- Real-time adaptation through multi-agent systems: Multiple AI agents work together to monitor and adjust network parameters, each specializing in different aspects of network management while coordinating their actions.
Performance Requirements with Agentic Systems
Technical capabilities include:
- Advanced spectral efficiency through collaborative AI agents: Multiple agents work together to optimize spectrum usage, with each agent responsible for different frequency bands or network segments, achieving up to 50% improvement in spectral efficiency.
- Predictive maintenance through agent-based monitoring: Autonomous agents continuously monitor network components, sharing information and collectively deciding on maintenance actions before failures occur.
- Dynamic network slice orchestration: Agentic AI systems negotiate and manage network slices autonomously, balancing competing demands and optimizing resource allocation in real-time.
- Goal-oriented QoS management: Agents pursue specific quality objectives, automatically negotiating and adjusting network parameters to meet service level agreements.
Edge Computing with Autonomous Agents
AI-native architecture enables:
- Distributed agent-based processing: Autonomous agents at edge nodes make independent decisions while coordinating with other agents to optimize overall network performance.
- Adaptive edge resource management: Agents learn from local conditions and adjust their behavior, optimizing resource allocation based on changing demands and conditions.
- Collaborative decision making: Multiple agents work together across edge nodes to make coordinated decisions, improving overall network efficiency.
- Agent-based service optimization: Autonomous agents continuously optimize service delivery based on learned patterns and real-time conditions.
Security Architecture with Agentic AI
AI components provide:
- Autonomous security agents: Self-directed security agents actively hunt for threats, coordinate responses, and adapt their strategies based on new attack patterns.
- Collaborative threat response: Multiple security agents work together to identify and respond to distributed attacks, sharing information and coordinating defensive actions.
- Self-evolving security policies: Agents continuously learn from security incidents and adapt policies autonomously, improving defense mechanisms over time.
- Multi-agent security coordination: Different security agents specialize in various aspects of network security while coordinating their actions for comprehensive protection.
Operational Impact of Agentic Systems
Agentic Systems can provide measurable improvements through:
- Autonomous operation optimization: Self-directing agents reduce manual operations by up to 85%, continuously improving their performance through learning.
- Agent-based resource management: Collaborative agents achieve up to 45% improvement in resource utilization through coordinated decision-making.
- Automated service evolution: Agents autonomously identify and implement service improvements based on usage patterns and performance metrics.
Technical Evolution with Agentic AI
The progression toward 6G requires:
- Multi-agent network architecture: Networks of autonomous agents work together across all network layers, each pursuing specific objectives while contributing to overall network goals.
- Emergent intelligence capabilities: Collective behavior of multiple agents leads to advanced network optimization beyond what individual AI systems can achieve.
- Self-evolving network functions: Agents continuously learn and adapt their strategies, improving network performance over time without human intervention.
- Collaborative spectrum optimization: Multiple agents work together to achieve up to 60% better spectrum utilization through coordinated management strategies.
Conclusion
The introduction of Large Telco Models represents an interesting development in telecommunications network management. These AI solutions directly address the challenges of processing 3,800+ terabytes of network data per minute that traditional automation tools cannot handle effectively. With 40% of telecom companies already implementing AI in their operations and major players developing custom LTMs, the industry is rapidly moving toward AI-driven network management. This transformation promises more efficient operations, improved network performance, and enhanced service delivery capabilities. As adoption increases, LTMs and AI agents will become standard components in modern telecommunications infrastructure, establishing new benchmarks for network operation and optimization.