Why Telecoms Are Betting on Open AI Models and NVIDIA Nemotron 3 LTM
Telecom operators embrace open AI models for control and customization as NVIDIA launches Nemotron 3 LTM and tools for secure local deployment at scale.
Summary
On October 6, 2026, NVIDIA said 89% of respondents to its latest State of AI in Telecommunications survey considered open source models and software important to their AI strategies. Operators gain lower-cost access to increasingly competitive intelligence while reserving closed models for selected workloads. Artificial Analysis Intelligence Index v4.3.2 shows leading open models improving in reasoning, coding, science and agentic tasks. Open weights and recipes also support telecom-specific tuning, model inspection and governance, deployment across public clouds, private infrastructure and edge environments, and locally adapted services for enterprise and government customers.
NVIDIA’s Nemotron family offers open weights, training data and recipes, agentic reasoning and speech capabilities. Its 30-billion-parameter Nemotron 3 Large Telco Model, fine-tuned by AdaptKey on open source telecom datasets, targets network configuration and customer incident triage; NVIDIA also released an end-to-end NVIDIA NeMo recipe for tuning models with operator data. SoftBank Corp. is using Nemotron and other foundations in its continuously developed SoftBank Large Telecom Model for network operations, design and management, Rajeev Koodli, SoftBank principal fellow and SB Telecom America senior vice president, said. AT&T chief data and AI officer Andy Markus said NVIDIA helps match workloads by performance, cost and control while providing scale, reliability and governance. NVIDIA AI Enterprise and NVIDIA Agent Toolkit cover protected data pipelines, synthetic data, agent orchestration, secure runtimes and simulation. Indosat Ooredoo Hutchison chief data and AI officer Chirag Sukhadia said Sahabat-AI adapts open models to Indonesia’s language, culture, data and applications.
Positives
- 89% of telecom survey respondents consider open source models and software important to their companies’ AI strategies.
- Nemotron 3 LTM uses 30 billion parameters and telecom datasets to improve network configuration and customer incident triage.
- NVIDIA released a complete NeMo recipe for adapting open models to each operator’s networks, customers and procedures.
- SoftBank Corp. is continuously advancing its Large Telecom Model for network operations, design and management.
- Sahabat-AI gives Indosat Ooredoo Hutchison an open foundation tailored to Indonesian language, culture, data and applications.
- Open models can run across public clouds, private infrastructure and edge environments while supporting locally governed services.
Risks & concerns
- Open models alone cannot deliver safe autonomous telecom operations without protected data pipelines, governance, orchestration, secure runtimes and simulation.
- Fine-tuning requires operators to anonymize sensitive records or generate privacy-preserving synthetic datasets before using operational data.
- Model choice adds operational complexity because each workload must be matched against performance, cost and control requirements.
- Locally adapted services must satisfy different language, industry, regulatory and data governance requirements across countries.