Telcos Harness Open Models for Advanced AI Solutions

Telecom operators are shifting towards open models for their AI strategies, driven by the potential for effective customization and control, rather than just cost considerations. This trend is reflected in NVIDIA’s recent State of AI in Telecommunications report, which highlights that 89% of operators view open-source models as crucial for their AI initiatives.

The strategic significance of open models in the telecom sector stems from five key benefits:

1. Access to frontier-level intelligence at reduced costs enables operators to allocate proprietary models for use cases yielding the most value. Notably, benchmarks like the Artificial Analysis Intelligence Index v4.3.2 demonstrate that the performance of competitive open models is increasing across complex reasoning and coding tasks.

2. Customization capabilities for telecom-specific needs are significantly enhanced, as operators can utilize open weights and training recipes tailored for their networks, customers, and industry requirements.

3. Open models promote trustworthy AI practices. They enable greater visibility and control over model behavior and outputs, aligning these with business policies and regulatory frameworks.

4. The flexibility of deployment options allows teams to optimize and secure open models for use across public cloud environments, private strategies, and edge computing.

5. By fine-tuning open models, telcos can offer AI solutions that are specifically adapted for local enterprise and governmental needs.

Open Foundations Customized for Telecom

The NVIDIA Nemotron series exemplifies how open models are refined for telecom operations, providing capabilities that span frontier-level reasoning and speaker recognition, all based on open weights and training frameworks. For instance, SoftBank Corp. illustrates the application of these models to advance their unique AI capabilities.

Rajeev Koodli, principal fellow of SoftBank Corp., stated, “Open models allow us to leverage global foundation models while utilizing our operational insights.” This approach supports the development of the SoftBank Large Telecom Model, targeted at enhancing telecom-specific functions like network operations and management.

NVIDIA collaborates with various partners to create practical applications from these open models. The recently announced 30-billion-parameter Nemotron 3 Large Telco Model has been specifically fine-tuned on telecom datasets, offering a solid baseline that understands industry terminology and optimizes workflows for configurations and incident management.

Operators also benefit from the detailed fine-tuning recipes released by NVIDIA, designed to help them adapt the Nemotron 3 LTM to their specific operational data and requirements through the NVIDIA NeMo open libraries.

Deploying Open Models in Production

While open models are integral to building effective AI solutions, successfully transitioning these models into operational settings requires a comprehensive strategy. AT&T exemplifies the importance of choosing the right models tailored to business objectives. Andy Markus, AT&T’s chief data and AI officer, commented, “We view the future of AI as not about singular models but about precisely matching each workload with the optimal blend of performance and control.”

To facilitate this, telecom operators must develop robust pipelines that ensure data privacy and compliance during model fine-tuning. These include anonymizing sensitive data and creating synthetic datasets for training without revealing confidential information. NVIDIA’s end-to-end platform offers the necessary framework, spanning everything from data pipelines to agent orchestration, ensuring a safe and effective deployment environment.

By building collaborative partnerships across the platform’s ecosystem, telecom operators can transform the advantages of open models into real-world AI workflows.

Fostering Local AI Innovation

As telecom companies develop AI infrastructures aligned with national strategies, open models empower them to create AI services that meet local demands in languages, industry norms, and regulatory needs. For example, operators can tailor open models for unique linguistic and regional contexts while enabling clients to adapt them further for their specific uses.

In Indonesia, Indosat Ooredoo Hutchison is applying open models to create AI that resonates with local culture through their Sahabat-AI suite. Chirag Sukhadia, the chief data and AI officer, emphasized, “The significance of open models lies in their adaptability to our language and cultural context.” This application not only promotes AI adoption but also enhances local capabilities, allowing many to engage with and develop AI solutions relevant to their circumstances.

With a strong foundation in open models, telecom operators are positioned to not just adopt advanced AI technologies but also foster substantial growth and innovation within their local markets.