Telecom operators turn to open AI models, Nvidia says

Telecom operators are increasingly basing their artificial intelligence strategies on open models, according to Nvidia, which cited survey findings showing broad support for open source models and software in telecom AI plans.

Its latest State of AI in Telecommunications report found that 89% of respondents viewed open source models and software as important to their company's AI strategy. Nvidia said operators are using open models not only to manage costs, but also to gain more control over how AI is trained, adapted and deployed across network operations and customer service.

The trend reflects a wider shift in telecom as operators try to apply generative AI and agent-based systems to core operational tasks. These include network design, incident triage, autonomous operations and customer interactions, all of which require models that can be adapted to sector-specific data and rules.

Nvidia said open models are becoming more competitive with proprietary alternatives in reasoning, coding, scientific and agentic workloads. Open weights and training methods also let telecom groups fine-tune models with their own network, customer and industry data, while giving them greater visibility into system behaviour for compliance and governance.

Open models can also be deployed across public cloud, private infrastructure and edge environments, an important feature for telecom operators that often run a mix of systems. Some carriers are also considering them as a way to offer localised AI services to enterprise and government customers.

Operator use

SoftBank is among the operators using open models as a base for telecom-specific AI development. The Japanese group is applying them to build what it calls a large telecom model for network operations, design and management.

"Open models allow SoftBank Corp. to build on the rapid progress of global foundation models while applying the network knowledge and operational expertise we have accumulated over many years," said Rajeev Koodli, Principal Fellow of SoftBank Corp. and Senior Vice President of SB Telecom America.

"We are using open foundations extensively, including NVIDIA Nemotron models and others, in developing our SoftBank Large Telecom Model, continuously advancing it for telecom-specific use cases such as network operations, design and overall management," Koodli said.

Nvidia also highlighted work around its Nemotron family of open models, which it said are aimed at reasoning tasks and voice applications. A 30-billion-parameter Nemotron 3 Large Telco Model was fine-tuned by AdaptKey on open source telecom datasets to improve accuracy in telecom-specific tasks.

According to Nvidia, the model is intended to give operators an open baseline that understands telecom terminology and can reason through workflows such as network configuration and customer incident handling. Nvidia also said it has released a full fine-tuning recipe through its NeMo open libraries to help operators adapt such models to their own networks, procedures and customer environments.

Model choice

AT&T framed the issue as selecting the right model for the right task rather than backing a single approach. That reflects a common concern among large operators as AI systems move from trials into live operations and they try to balance performance, spending and oversight.

"At AT&T, we believe the future of AI is not about choosing a single model; it's about intelligently matching every workload to the right combination of performance, cost and control," said Andy Markus, Chief Data and AI Officer of AT&T.

"Open models are essential to that approach, and with NVIDIA, we can bring that model-choice strategy into production with the scale, reliability and governance our business requires," Markus said.

Nvidia said production deployment requires more than the underlying model. Operators also need data pipelines that can anonymise sensitive records, produce synthetic datasets and support governed workflows for autonomous agents.

Its software stack and tools are designed to cover those steps, from data preparation and model tuning to orchestration, secure runtime and simulation. Nvidia also pointed to partner involvement across different parts of the process as operators seek to move AI systems into routine use.

Local language

Another theme is the use of open models to support national and local AI agendas. Telecom operators in several markets are building infrastructure intended to meet domestic language, regulatory and data governance requirements, and open models are being used as the basis for those efforts.

Indosat Ooredoo Hutchison is using that approach in Indonesia through its Sahabat-AI family of open source models. The operator said the goal is to build AI systems that reflect local language and culture rather than relying solely on models developed elsewhere.

"For countries like Indonesia, the value of open models goes beyond access to powerful AI. It is about adapting that intelligence to our own language, culture, data and real-world needs," said Chirag Sukhadia, Chief Data and AI Officer of Indosat Ooredoo Hutchison.

"Sahabat-AI puts this into practice, using open models as a foundation to build AI that understands Indonesia and can be developed for local applications. This allows us not only to adopt AI, but to build capabilities around it and enable more Indonesians to create with AI on their own terms," Sukhadia said.