THE CRUNCH
Telecom operators are increasingly basing their AI strategies on open models, and NVIDIA argues the appeal goes beyond cost, per a company blog post. Open weights let operators fine-tune models on their own network and customer data, inspect and govern model behaviour, and deploy across public clouds, private infrastructure and edge environments. The claims come from NVIDIA, which sells the platform and models involved, so they deserve a sceptical read.
The case, as NVIDIA lays it out, rests on five claimed advantages: cheaper access to frontier-level intelligence, customisation for telecom-specific work using operators' own network and customer data, greater visibility and control over model behaviour, flexible deployment across clouds, private infrastructure and edge environments, and the chance to sell locally adapted AI services to enterprise and government customers.
Operators are already putting this into practice, according to the post. SoftBank Corp. says it is building its SoftBank Large Telecom Model on open foundations including NVIDIA's Nemotron models, while Indonesia's Indosat Ooredoo Hutchison has developed the Sahabat-AI family of open source models tuned to Indonesian language and culture. NVIDIA has also announced a 30-billion-parameter Nemotron 3 Large Telco Model, fine-tuned by AdaptKey on open source telecom datasets, and released a full recipe for operators to fine-tune open models with their own operational data using its NeMo libraries.
AT&T's chief data and AI officer Andy Markus framed model choice rather than model loyalty as the goal, saying open models are essential to matching each workload to the right combination of performance, cost and control. NVIDIA argues models alone are not enough, pointing to its own end-to-end platform spanning data pipelines, agent orchestration and secure runtimes.