Why Telecom Operators Are Building Their AI Strategy on Open Models

According to NVIDIA's latest State of AI in Telecommunications report, 89% of respondents said open source models and software are important to their company's AI strategy. The company argues the appeal goes past cost: carriers want to inspect, fine-tune and govern the models running their networks and customer care.

Alongside that, NVIDIA announced a 30-billion-parameter Nemotron 3 Large Telco Model, fine-tuned by AdaptKey on open telecom datasets, and released a full recipe for operators to adapt open models to their own networks using its NeMo libraries.

We think the core argument holds up. Telecom networks are critical infrastructure. An operator that can see a model's weights, train it on its own network and customer data, and run it on private or edge hardware has more control than one renting a black box. Matching every workload to the right mix of performance, cost and control is how AT&T's chief data and AI officer describes the approach, and that's a mature way to think about it.

The local-language angle is the most interesting part. Indosat Ooredoo Hutchison's Sahabat-AI models are built to reflect Indonesian language and culture. That's open models doing something closed ones rarely prioritize: serving a market on its own terms. SoftBank describes building its own large telecom model on open foundations, including Nemotron.

Now the caveat. This is NVIDIA's blog, about an NVIDIA survey, promoting NVIDIA models, recipes and an end-to-end NVIDIA platform. 'Open' here sits on top of a stack that is very much one company's. Operators should welcome open weights and still watch how much of the surrounding tooling ties them to a single vendor's hardware and software.

There's also real risk in the autonomous network vision. Fine-tuning on customer records requires serious privacy work, which NVIDIA acknowledges with talk of anonymization and synthetic data. And agents that configure networks or triage incidents can fail at scale in ways a chatbot can't. If this spreads quickly, the safeguards and audit trails have to keep pace.

Carriers are also positioning themselves as hosts for national AI efforts, fine-tuning open models for local languages, industries and data rules and offering them to enterprise and government customers.