Telecom operators' strategy shifts to open weights as the 30 billion-parameter architecture reshapes network management

NVIDIA's latest report on the state of artificial intelligence in the telecommunications sector reveals a structural shift that goes beyond direct cost-reduction calculations, as its findings show that 89 % of the participating operators confirm that open-source models and software have become a core pillar of their corporate strategies. This trend reflects a clear desire among telecom companies to regain full control over sensitive data flows and complex network operations, rather than relying on closed-model systems that impose limits on transparency and curb internal customization.

The strategic value of open models stems from their ability to deliver efficiency comparable to leading models at lower operating cost, allowing operators to limit the use of closed models to highly specialized use cases. This approach is supported by independent benchmark measurements, such as the Artificial Analysis Intelligence Index v4.3.2, which showed that open models have become a genuine competitor in logical reasoning, programming and workflow automation tasks. In addition, the ability to train models and adjust their weights using network logs and subscriber data gives companies a superior capability to audit software behavior and align it with regulatory policies, while retaining distribution flexibility across private infrastructure and edge computing.

This practical path is embodied in NVIDIA's announcement of the Nemotron 3 Large Telco Model, a thirty-billion-parameter model that was finely tuned by AdaptKey using open telecom data sets to achieve higher accuracy in network management and customer-issue classification. The company also provided full programmatic tuning steps through the open-source NVIDIA NeMo libraries. In the same context, SoftBank is developing its own telecom model based on the Nemotron family to support network design and management, while AT&T is adopting a workload-balancing strategy that blends performance with corporate control, and Indosat Ooredoo Hutchison in Indonesia has built the open Sahabat-AI model suite to align applications with local language, culture and data.

Moving to production requires a strict governance platform to protect data before model training.Shifting the infrastructure to self-managing networks requires pipelines that anonymize sensitive records and generate synthetic data that preserve privacy, together with tools for orchestrating agents and securing runtime environments such as NVIDIA Agent Toolkit and the NVIDIA AI Enterprise suite, ensuring that open models move into corporate production environments without jeopardising network security.

This shift carries direct strategic implications for telecom operators and engineering teams in the Gulf, Egypt and the Levant. In markets that impose strict regulatory frameworks on data localisation and communications confidentiality, relying on closed, offshore APIs represents a regulatory risk and a rising financial burden in inference costs. What changes in practice for network engineers and technology leaders in the region is the need to build advanced capabilities in synthetic data engineering, to fine-tune open-model weights locally, and to manage agents in private data centres. This path reduces dependence on external subscriptions and redirects operating budgets toward strengthening digital sovereignty over infrastructure and sensitive corporate data.

The bet today is shifting from merely consuming ready-made AI models to owning their operational assets and managing them autonomously.Telecom operators who build their systems on open, auditable and adaptable weights gain technical flexibility and the ability to protect their records, making model control a natural extension of control over network cables and towers.